Publikationen im NUM
Hier finden Sie eine Liste der Publikationen, die im Zusammenhang mit dem Netzwerk Universitätsmedizin in der ersten und zweiten Förderphase entstanden sind.
N. Bayerl,
J. Appel,
M. Schmidt,
A. Cavallaro,
R. Grimm,
H. Busch,
B. Wullich,
A. Hartmann,
M. Schlicht,
M. Uder,
M. S. May and
S. Ellmann,
"Diagnostic performance of AI-powered prostate MRI against biopsy ground truth: quasi-continuous risk scoring versus fixed PI-RADS thresholds",
Cancer imaging : the official publication of the International Cancer Imaging Society,
vol. 26,
no. 1,
Sep.
2026.
Abstract:
Multiparametric prostate MRI is central to prostate cancer (PCa) diagnostics, yet PI-RADS interpretation suffers from inter-reader variability. This study aimed to evaluate a commercial AI algorithm for cancer detection in prostate MRI against histopathology, comparing its standard PI-RADS classification with its quasi-continuous Level-of-Suspicion (LoS) score as decision variables. In this retrospective single-center study, 122 mpMRI examinations, each followed by systematic and/or fusion-targeted biopsy, were analyzed. Histopathology served as the reference standard for any PCa (Gleason ≥ 6) and clinically significant PCa (csPCa; Gleason ≥ 7). Diagnostic performance of the algorithm's PI-RADS and LoS outputs was assessed by ROC analysis. AUCs were compared using DeLong's test. The Youden-optimal LoS cutoff was determined, its optimism was quantified by bootstrap internal validation, and the calibration of the LoS score was assessed after logistic recalibration. For csPCa, PI-RADS yielded an AUC of 0.833 (95% CI: 0.764-0.901), versus 0.882 (95% CI: 0.820-0.943) for the LoS score (p = 0.106). At PI-RADS ≥ 4, sensitivity was 94.7% and specificity 63.1%. The Youden-optimal LoS cutoff was 81. At this exploratory threshold, sensitivity remained 94.7% while specificity was 70.8%, with positive and negative predictive values of 74.0% and 93.9%, respectively. Bootstrap internal validation indicated modest optimism, with corrected estimates of 92.9% for sensitivity and 69.1% for specificity. After logistic recalibration, the bootstrap-corrected calibration slope was 0.98 with an intercept of -0.01. In this retrospective single-center cohort, the AI algorithm achieved high diagnostic accuracy for csPCa. Its quasi-continuous LoS score provided an operating point combining high sensitivity with numerically higher specificity than the PI-RADS ≥ 4 threshold. Neither this difference (exact McNemar p = 0.063) nor the difference in AUC between the two classifiers (p = 0.106) reached statistical significance, so the finding should be regarded as a promising trend rather than a demonstrated advantage. Prospective confirmation is required before AI-assisted standardized interpretation is integrated into prostate MRI workflows on this basis.
| DOI: | 10.1186/s40644-026-01126-5 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42745323/ |
Abstract:
Multiparametric prostate MRI is central to prostate cancer (PCa) diagnostics, yet PI-RADS interpretation suffers from inter-reader variability. This study aimed to evaluate a commercial AI algorithm for cancer detection in prostate MRI against histopathology, comparing its standard PI-RADS classification with its quasi-continuous Level-of-Suspicion (LoS) score as decision variables. In this retrospective single-center study, 122 mpMRI examinations, each followed by systematic and/or fusion-targeted biopsy, were analyzed. Histopathology served as the reference standard for any PCa (Gleason ≥ 6) and clinically significant PCa (csPCa; Gleason ≥ 7). Diagnostic performance of the algorithm's PI-RADS and LoS outputs was assessed by ROC analysis. AUCs were compared using DeLong's test. The Youden-optimal LoS cutoff was determined, its optimism was quantified by bootstrap internal validation, and the calibration of the LoS score was assessed after logistic recalibration. For csPCa, PI-RADS yielded an AUC of 0.833 (95% CI: 0.764-0.901), versus 0.882 (95% CI: 0.820-0.943) for the LoS score (p = 0.106). At PI-RADS ≥ 4, sensitivity was 94.7% and specificity 63.1%. The Youden-optimal LoS cutoff was 81. At this exploratory threshold, sensitivity remained 94.7% while specificity was 70.8%, with positive and negative predictive values of 74.0% and 93.9%, respectively. Bootstrap internal validation indicated modest optimism, with corrected estimates of 92.9% for sensitivity and 69.1% for specificity. After logistic recalibration, the bootstrap-corrected calibration slope was 0.98 with an intercept of -0.01. In this retrospective single-center cohort, the AI algorithm achieved high diagnostic accuracy for csPCa. Its quasi-continuous LoS score provided an operating point combining high sensitivity with numerically higher specificity than the PI-RADS ≥ 4 threshold. Neither this difference (exact McNemar p = 0.063) nor the difference in AUC between the two classifiers (p = 0.106) reached statistical significance, so the finding should be regarded as a promising trend rather than a demonstrated advantage. Prospective confirmation is required before AI-assisted standardized interpretation is integrated into prostate MRI workflows on this basis.
M. Lindholz,
I. G. Schoots,
R. Bülow,
P. Asbach,
M. Kölln,
S. Nolte,
S. Alaro,
R. Ruppel,
N. Hosten,
T. Penzkofer and
C. A. Hamm,
"Reference values for PSA, PSA density, and MRI prostate volume in healthy German men",
European radiology,
Sep.
2026.
Abstract:
To establish stratified normative reference values for MRI-derived prostate volume (PV), prostate-specific antigen (PSA), and PSA density (PSAD) in German men with no known prostate pathologies, accounting for age, body-mass index, and height. This study included men from the prospective population-based Study of Health in Pomerania, Germany, with 10 years of follow-up. Men with diagnosed prostate cancer, benign prostatic hyperplasia, or lower urinary tract symptoms were excluded to derive reference values unaffected by disease or interventions. PV and PSAD were obtained from pelvic MRI using deep-learning segmentation. In a test set, segmentation-derived volumes were compared with manual PVs. PV, PSA, and PSAD were modeled using generalized additive models for location, scale, and shape (GAMLSS). Out of 1696 men, 1037 were analyzed (median age 46.00 years [IQR 38.00-56.00]). Median PV was 31.29 mL (5th-95th: 20.74-51.39), PSA 0.81 ng/mL (0.32-2.79), and PSAD 0.025 ng/mL² (0.011-0.071). PV and PSA increased progressively with age, whereas PSAD showed greater age-stability, peaking at 71 years, plateauing until 75, and then declining. Age-adjusted PSAD never exceeded 0.11 ng/mL², unlike PSA, which rose continuously. Body-mass index correlated positively with PV but inversely with PSAD; PSA showed a negative association with height. This study provides reference values and percentile curves for PV, PSA, and PSAD in men without prostate disease, along with an open-access calculator. PSAD showed age-stability, contrasting with continuously rising PSA. These reference values may help distinguish normal from pathological prostatic variation across age strata and inform PSAD thresholds in screening programs. Question Population-based reference values with long-term follow-up for PV, PSA, and PSAD are lacking, limiting the distinction between normal variation and disease-related changes. Findings Among 1037 men without prostate disease, 95th percentiles were: PV 51.39 mL, PSA 2.79 ng/mL, and PSAD 0.071 ng/mL². PSAD remained below 0.11 ng/mL² across all age ranges. Clinical relevance MRI-based percentiles for PV and PSAD may help identify normal variation, reducing unnecessary evaluations while supporting early detection of abnormalities. An open-access calculator enables personalized assessment for clinical and research purposes.
| DOI: | 10.1007/s00330-026-12937-2 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42789055/ |
Abstract:
To establish stratified normative reference values for MRI-derived prostate volume (PV), prostate-specific antigen (PSA), and PSA density (PSAD) in German men with no known prostate pathologies, accounting for age, body-mass index, and height. This study included men from the prospective population-based Study of Health in Pomerania, Germany, with 10 years of follow-up. Men with diagnosed prostate cancer, benign prostatic hyperplasia, or lower urinary tract symptoms were excluded to derive reference values unaffected by disease or interventions. PV and PSAD were obtained from pelvic MRI using deep-learning segmentation. In a test set, segmentation-derived volumes were compared with manual PVs. PV, PSA, and PSAD were modeled using generalized additive models for location, scale, and shape (GAMLSS). Out of 1696 men, 1037 were analyzed (median age 46.00 years [IQR 38.00-56.00]). Median PV was 31.29 mL (5th-95th: 20.74-51.39), PSA 0.81 ng/mL (0.32-2.79), and PSAD 0.025 ng/mL² (0.011-0.071). PV and PSA increased progressively with age, whereas PSAD showed greater age-stability, peaking at 71 years, plateauing until 75, and then declining. Age-adjusted PSAD never exceeded 0.11 ng/mL², unlike PSA, which rose continuously. Body-mass index correlated positively with PV but inversely with PSAD; PSA showed a negative association with height. This study provides reference values and percentile curves for PV, PSA, and PSAD in men without prostate disease, along with an open-access calculator. PSAD showed age-stability, contrasting with continuously rising PSA. These reference values may help distinguish normal from pathological prostatic variation across age strata and inform PSAD thresholds in screening programs. Question Population-based reference values with long-term follow-up for PV, PSA, and PSAD are lacking, limiting the distinction between normal variation and disease-related changes. Findings Among 1037 men without prostate disease, 95th percentiles were: PV 51.39 mL, PSA 2.79 ng/mL, and PSAD 0.071 ng/mL². PSAD remained below 0.11 ng/mL² across all age ranges. Clinical relevance MRI-based percentiles for PV and PSAD may help identify normal variation, reducing unnecessary evaluations while supporting early detection of abnormalities. An open-access calculator enables personalized assessment for clinical and research purposes.
A. Pfeil,
T. Hoffmann,
U. Teichgräber,
D. Renz,
D. Beckmann,
G. Wolf,
N. Schadewaldt and
B. Lassen-Schmidt,
"Beyond Visual Scoring: Computational Computed Tomography Analysis for High-Resolution Computed Tomography-Based Quantification of Interstitial Lung Disease in Inflammatory Rheumatic Disease",
Arthritis care & research,
Aug.
2026.
Abstract:
Interstitial lung disease (ILD) is a significant cause of morbidity and mortality in patients with inflammatory rheumatic disorders (IRDs). High-resolution computed tomography (HRCT) is widely considered the gold standard for the noninvasive assessment of ILD; however, its interpretation is constrained by substantial interobserver variability and the need for time-consuming expert evaluation. Computer-based image analysis, including artificial intelligence, has emerged as a promising approach for the automated, objective, and quantitative analysis of HRCT images. This method enables precise volumetric assessment of parenchymal alterations and facilitates pattern classification with unprecedented accuracy and efficiency. This review synthesizes current evidence on computational driven HRCT quantification in IRD-ILD, highlighting available technical approaches, validation strategies, and clinical applications.
| DOI: | 10.1002/acr.80142 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42605892/ |
Abstract:
Interstitial lung disease (ILD) is a significant cause of morbidity and mortality in patients with inflammatory rheumatic disorders (IRDs). High-resolution computed tomography (HRCT) is widely considered the gold standard for the noninvasive assessment of ILD; however, its interpretation is constrained by substantial interobserver variability and the need for time-consuming expert evaluation. Computer-based image analysis, including artificial intelligence, has emerged as a promising approach for the automated, objective, and quantitative analysis of HRCT images. This method enables precise volumetric assessment of parenchymal alterations and facilitates pattern classification with unprecedented accuracy and efficiency. This review synthesizes current evidence on computational driven HRCT quantification in IRD-ILD, highlighting available technical approaches, validation strategies, and clinical applications.
P. Hahlbohm,
S. Starke,
F. Schön,
L. Misera,
M. Erdmann,
R. Hoffmann,
A. Zwanenburg,
J. Kühn and
S. Löck,
"External validation of a top-ranked model from the RSNA pulmonary embolism detection challenge: assessment of generalizability",
Scientific reports,
vol. 16,
no. 1,
Aug.
2026.
Abstract:
To externally validate the RSNA 2020 challenge 2nd-place deep-learning (DL) algorithm for detecting pulmonary embolism (PE) on computed tomography pulmonary angiography (CTPA) scans with additional focus on subsegmental-only PEs (SSPE). Between 2015 and 2022 1,038 CTPAs were retrospectively enrolled. An experienced radiologist (> five years) labelled CTPAs for the presence, location (central, lobar and segmental, subsegmental-only) and side (right-sided, left-sided, bilateral) of PE. The DL model was tested for its ability to predict these labels by analyzing accuracy, sensitivity, specificity and area under the receiver operating characteristic curve (AUROC) using different model outcome probabilities for each analysis. Overall, 136 central, 375 peripheral PEs (303 lobar and segmental PEs, 72 SSPEs), and 527 patients without any PE were analyzed. The model correctly predicted the presence of any PE in 921/1,038 patients (88.7%), yielding a sensitivity of 0.80, specificity of 0.97 and AUROC of 0.94. No central PE was missed, whereas 100/375 (26.7%) peripheral PEs remained undetected. Using the corresponding model output probability, the central PE status was correctly identified in 999/1,038 (96.2%) patients, while peripheral PE was correctly identified in 785/1,038 (75.6%) patients. This corresponded to sensitivities of 0.94 and 0.77, specificities of 0.97 and 0.75, and AUROC values of 0.99 and 0.74, respectively. The model performed better in detecting right-sided compared to left-sided PEs (AUROC: right- vs. left-sided: 0.95 and 0.92, p < 0.05). SSPE status was correctly predicted in 796/1,038 (76.7%) patients, yielding a sensitivity of 0.94, a specificity of 0.22, and an AUROC of 0.48, respectively. The model demonstrated high performance in detecting any PE, performing best on central and slightly lower on lobar and segmental PE. Predicting the presence of SSPEs was difficult, likely due to the model not being explicitly trained for this subtype.
| DOI: | 10.1038/s41598-026-65551-z |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42570951/ |
Abstract:
To externally validate the RSNA 2020 challenge 2nd-place deep-learning (DL) algorithm for detecting pulmonary embolism (PE) on computed tomography pulmonary angiography (CTPA) scans with additional focus on subsegmental-only PEs (SSPE). Between 2015 and 2022 1,038 CTPAs were retrospectively enrolled. An experienced radiologist (> five years) labelled CTPAs for the presence, location (central, lobar and segmental, subsegmental-only) and side (right-sided, left-sided, bilateral) of PE. The DL model was tested for its ability to predict these labels by analyzing accuracy, sensitivity, specificity and area under the receiver operating characteristic curve (AUROC) using different model outcome probabilities for each analysis. Overall, 136 central, 375 peripheral PEs (303 lobar and segmental PEs, 72 SSPEs), and 527 patients without any PE were analyzed. The model correctly predicted the presence of any PE in 921/1,038 patients (88.7%), yielding a sensitivity of 0.80, specificity of 0.97 and AUROC of 0.94. No central PE was missed, whereas 100/375 (26.7%) peripheral PEs remained undetected. Using the corresponding model output probability, the central PE status was correctly identified in 999/1,038 (96.2%) patients, while peripheral PE was correctly identified in 785/1,038 (75.6%) patients. This corresponded to sensitivities of 0.94 and 0.77, specificities of 0.97 and 0.75, and AUROC values of 0.99 and 0.74, respectively. The model performed better in detecting right-sided compared to left-sided PEs (AUROC: right- vs. left-sided: 0.95 and 0.92, p < 0.05). SSPE status was correctly predicted in 796/1,038 (76.7%) patients, yielding a sensitivity of 0.94, a specificity of 0.22, and an AUROC of 0.48, respectively. The model demonstrated high performance in detecting any PE, performing best on central and slightly lower on lobar and segmental PE. Predicting the presence of SSPEs was difficult, likely due to the model not being explicitly trained for this subtype.
C. Römer,
J. Hölzen,
A. Pascher,
W. Heindel,
M. Künnemann,
R. Hosch,
K. Borys and
G. H. Pöhler,
"Prediction of Overall Survival in Patients With Oesophageal Cancer Using AI-Based 3D CT Body Composition Analysis",
Journal of cachexia, sarcopenia and muscle,
vol. 17,
no. 4,
pp. e70363,
Aug.
2026.
Abstract:
Oesophageal cancer remains the sixth most lethal malignancy, with 5-year survival rates around 22%. CT-derived 2D body composition analysis has emerged as a promising prognostic tool, but conventional, often manually created single-slice L3 measurements are unsuited for future clinical implementation. We evaluated fully automated AI-derived volumetric body composition indices as prognostic parameters using conventional L3 and BMI measurements as reference. This retrospective cohort study included patients with histologically confirmed oesophageal cancer treated between 2011 and 2024. Automated deep learning segmentation using the nnU-Net-based body and organ analysis pipeline quantified abdominal tissue volumes from staging CTs. Three normalized indices were calculated: sarcopenia index (SI, muscle/bone volume ratio), myosteatotic fat index (MFI, intramuscular/total adipose tissue volume ratio) and abdominal fat index (AFI, visceral/subcutaneous adipose tissue volume ratio). Cox proportional hazards models assessed prognostic value after sequential adjustment for age, sex, metastatic status, ECOG performance status, BMI and L3-derived indices. Kaplan-Meier analysis evaluated survival differences stratified by sex-specific median values. The cohort comprised 563 patients (19.7% female), median age 65.4 years (IQR: 58.8-71.3); 10.1% (n = 57) had metastatic disease and 68.7% (n = 387) underwent surgery. Males had higher sarcopenia index (2.54 ± 0.41 vs. 2.24 ± 0.45, p < 0.001) and abdominal fat index (median 0.72 vs. 0.36, p < 0.001), whereas MFI showed no difference (p = 0.89). During median follow-up of 22 months, 367 deaths (65.9%) occurred. Median overall survival was 28.6 months (95% CI: 23.7-33.3); 1-year survival 70.7% (95% CI: 67.2%-74.8%) and 5-year survival 32.8% (95% CI: 29.6%-38.3%), with marked differences by metastatic status (M0 vs. M1 1-year survival: 73.4% vs. 46.4%). In the fully adjusted model incorporating all three volumetric indices alongside clinical covariates, BMI and L3-derived parameters (n = 532), only sarcopenia index retained independent significance (HR = 0.56, 95% CI: 0.37-0.82, p = 0.003); all others showed p ≥ 0.26. Metastatic status (HR = 2.23, p < 0.001) and ECOG (HR = 1.29, p < 0.001) remained significant. All L3 indices lost significance alongside volumetric parameters (p > 0.14). Male patients with high sarcopenia index showed longer survival compared with patients with low sarcopenia index (33.3 vs. 21.8 months, p < 0.001); female patients showed larger descriptive contrasts (68.8 vs. 16.8 months, p < 0.001) that were formally confirmed by a significant sex-SI Cox interaction (LRT p = 0.026). Automatic AI-derived volumetric body composition parameters calculated from routine staging CTs predict overall survival in patients with oesophageal cancer and are associated with sex-related differences of prognostic impact, supporting further evaluation toward future clinical use.
| DOI: | 10.1002/jcsm.70363 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42581617/ |
Abstract:
Oesophageal cancer remains the sixth most lethal malignancy, with 5-year survival rates around 22%. CT-derived 2D body composition analysis has emerged as a promising prognostic tool, but conventional, often manually created single-slice L3 measurements are unsuited for future clinical implementation. We evaluated fully automated AI-derived volumetric body composition indices as prognostic parameters using conventional L3 and BMI measurements as reference. This retrospective cohort study included patients with histologically confirmed oesophageal cancer treated between 2011 and 2024. Automated deep learning segmentation using the nnU-Net-based body and organ analysis pipeline quantified abdominal tissue volumes from staging CTs. Three normalized indices were calculated: sarcopenia index (SI, muscle/bone volume ratio), myosteatotic fat index (MFI, intramuscular/total adipose tissue volume ratio) and abdominal fat index (AFI, visceral/subcutaneous adipose tissue volume ratio). Cox proportional hazards models assessed prognostic value after sequential adjustment for age, sex, metastatic status, ECOG performance status, BMI and L3-derived indices. Kaplan-Meier analysis evaluated survival differences stratified by sex-specific median values. The cohort comprised 563 patients (19.7% female), median age 65.4 years (IQR: 58.8-71.3); 10.1% (n = 57) had metastatic disease and 68.7% (n = 387) underwent surgery. Males had higher sarcopenia index (2.54 ± 0.41 vs. 2.24 ± 0.45, p < 0.001) and abdominal fat index (median 0.72 vs. 0.36, p < 0.001), whereas MFI showed no difference (p = 0.89). During median follow-up of 22 months, 367 deaths (65.9%) occurred. Median overall survival was 28.6 months (95% CI: 23.7-33.3); 1-year survival 70.7% (95% CI: 67.2%-74.8%) and 5-year survival 32.8% (95% CI: 29.6%-38.3%), with marked differences by metastatic status (M0 vs. M1 1-year survival: 73.4% vs. 46.4%). In the fully adjusted model incorporating all three volumetric indices alongside clinical covariates, BMI and L3-derived parameters (n = 532), only sarcopenia index retained independent significance (HR = 0.56, 95% CI: 0.37-0.82, p = 0.003); all others showed p ≥ 0.26. Metastatic status (HR = 2.23, p < 0.001) and ECOG (HR = 1.29, p < 0.001) remained significant. All L3 indices lost significance alongside volumetric parameters (p > 0.14). Male patients with high sarcopenia index showed longer survival compared with patients with low sarcopenia index (33.3 vs. 21.8 months, p < 0.001); female patients showed larger descriptive contrasts (68.8 vs. 16.8 months, p < 0.001) that were formally confirmed by a significant sex-SI Cox interaction (LRT p = 0.026). Automatic AI-derived volumetric body composition parameters calculated from routine staging CTs predict overall survival in patients with oesophageal cancer and are associated with sex-related differences of prognostic impact, supporting further evaluation toward future clinical use.
L. Garajová,
M. Theis,
S. Nowak,
D. Kütting,
W. Block,
J. A. Luetkens and
A. M. Sprinkart,
"Automated Scan Region Classification and Patient-specific Dose Modeling for Enhanced Dose Management in Computed Tomography",
Investigative radiology,
vol. 61,
no. 7,
pp. 477-485,
Jul.
2026.
Abstract:
Effective dose management in computed tomography is impeded by 2 key operational challenges: error-prone manual protocol mapping and the high volume of nonactionable alerts from fixed diagnostic reference levels (DRLs). This "alert fatigue" creates a risk of overlooking clinically significant dose deviations. This study aimed to develop and evaluate a novel artificial intelligence (AI)-assisted framework to automate scan classification and provide a patient-specific context for dose assessment. This retrospective study analyzed 2955 CT irradiation events. A processing pipeline was developed that first performs automated body segmentation using a deep learning model. A random forest classifier was then trained on the resulting organ volumes to identify 15 distinct scan regions. For 4 common examination types, linear regression models were established to predict the CT dose index (CTDI vol ) based on the patient's mean cross-sectional water-equivalent area. Cases were identified as statistical outliers if the absolute standardized residual was >2. The number of these outliers was compared with the number of conventional DRL exceedances. The automated scan region classifier achieved high accuracy, with a macro-averaged F1 score of 93.8% on the hold-out test set. The regression models demonstrated a clear linear correlation between patient anatomy and CTDI vol (r = 0.56 to 0.79). Crucially, the patient-specific models identified substantially fewer cases for review (60 statistical outliers) compared with the standard DRL-based method (170 exceedances). Manual analysis confirmed that all flagged cases were clinically justified. Our findings validate that an AI-assisted, patient-centered framework is a highly effective strategy for dose management. By shifting the paradigm from rigid, population-based thresholds to a dynamic, patient-specific assessment, our approach provides a more effective method for identifying potential dose deviations while substantially reducing the burden of nonactionable alerts. This work charts a course towards a new standard of radiation dose monitoring, advancing the field in the direction of a more efficient and reliable form of personalized dose monitoring.
| DOI: | 10.1097/rli.0000000000001247 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41169122/ |
Abstract:
Effective dose management in computed tomography is impeded by 2 key operational challenges: error-prone manual protocol mapping and the high volume of nonactionable alerts from fixed diagnostic reference levels (DRLs). This "alert fatigue" creates a risk of overlooking clinically significant dose deviations. This study aimed to develop and evaluate a novel artificial intelligence (AI)-assisted framework to automate scan classification and provide a patient-specific context for dose assessment. This retrospective study analyzed 2955 CT irradiation events. A processing pipeline was developed that first performs automated body segmentation using a deep learning model. A random forest classifier was then trained on the resulting organ volumes to identify 15 distinct scan regions. For 4 common examination types, linear regression models were established to predict the CT dose index (CTDI vol ) based on the patient's mean cross-sectional water-equivalent area. Cases were identified as statistical outliers if the absolute standardized residual was >2. The number of these outliers was compared with the number of conventional DRL exceedances. The automated scan region classifier achieved high accuracy, with a macro-averaged F1 score of 93.8% on the hold-out test set. The regression models demonstrated a clear linear correlation between patient anatomy and CTDI vol (r = 0.56 to 0.79). Crucially, the patient-specific models identified substantially fewer cases for review (60 statistical outliers) compared with the standard DRL-based method (170 exceedances). Manual analysis confirmed that all flagged cases were clinically justified. Our findings validate that an AI-assisted, patient-centered framework is a highly effective strategy for dose management. By shifting the paradigm from rigid, population-based thresholds to a dynamic, patient-specific assessment, our approach provides a more effective method for identifying potential dose deviations while substantially reducing the burden of nonactionable alerts. This work charts a course towards a new standard of radiation dose monitoring, advancing the field in the direction of a more efficient and reliable form of personalized dose monitoring.
O. Seidel,
M. Theis,
S. Nowak,
L. Garajová,
B. Wulff,
W. Block,
L. Müller,
T. Emrich,
J. A. Luetkens,
D. R. Hadizadeh,
A. M. Sprinkart and
D. Kravchenko,
"Deep learning-based detection of acute pancreatitis on abdominal contrast-enhanced CT",
European radiology experimental,
vol. 10,
no. 1,
Jul.
2026.
Abstract:
We developed and evaluated a deep learning (DL) model for image-based detection of acute pancreatitis (AP) on abdominal contrast-enhanced CT (CECT). A total of 552 patients from two university centers (January 2010-January 2026) were included. The internal dataset comprised 207 patients with clinically and radiologically confirmed AP (499 scans) and 250 control patients with suspected AP (368 scans). An independent external validation cohort included 95 patients. Convolutional neural network-based models were trained using monophasic and biphasic CECT data. The final model was evaluated on a 20% patient-level hold-out test set from the internal cohort and on the external cohort. Performance was assessed using the F1 score and area under the receiver operating characteristic curve (AUROC). A single-input multiphase model incorporating arterial and portal venous phase scans achieved the best performance, with ensembling applied for biphasic studies. On the internal hold-out test set (n = 116), the model achieved an F1 score of 0.83 (95% confidence interval 0.75-0.89) and an AUROC of 0.89 (0.82-0.95). Performance remained robust on external validation (n = 95), with an AUROC of 0.99 (0.96-1.00) and an F1 score of 0.92 (0.86-0.97). DL enabled accurate CECT-based identification of AP in this retrospective multicenter cohort, with performance maintained in an independent external dataset. Prospective validation using broader and independently adjudicated clinical populations remains necessary. The model showed promising performance for CECT-based acute pancreatitis detection but was not designed or tested as a triage system. Diagnostic uncertainty in acute pancreatitis often arises from nonspecific abdominal symptoms and inter-reader variability in CECT interpretation. The DL model achieved high internal accuracy (AUROC 0.89) and maintained robust performance in an independent external validation cohort (AUROC 0.99).
| DOI: | 10.1186/s41747-026-00775-2 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42536294/ |
Abstract:
We developed and evaluated a deep learning (DL) model for image-based detection of acute pancreatitis (AP) on abdominal contrast-enhanced CT (CECT). A total of 552 patients from two university centers (January 2010-January 2026) were included. The internal dataset comprised 207 patients with clinically and radiologically confirmed AP (499 scans) and 250 control patients with suspected AP (368 scans). An independent external validation cohort included 95 patients. Convolutional neural network-based models were trained using monophasic and biphasic CECT data. The final model was evaluated on a 20% patient-level hold-out test set from the internal cohort and on the external cohort. Performance was assessed using the F1 score and area under the receiver operating characteristic curve (AUROC). A single-input multiphase model incorporating arterial and portal venous phase scans achieved the best performance, with ensembling applied for biphasic studies. On the internal hold-out test set (n = 116), the model achieved an F1 score of 0.83 (95% confidence interval 0.75-0.89) and an AUROC of 0.89 (0.82-0.95). Performance remained robust on external validation (n = 95), with an AUROC of 0.99 (0.96-1.00) and an F1 score of 0.92 (0.86-0.97). DL enabled accurate CECT-based identification of AP in this retrospective multicenter cohort, with performance maintained in an independent external dataset. Prospective validation using broader and independently adjudicated clinical populations remains necessary. The model showed promising performance for CECT-based acute pancreatitis detection but was not designed or tested as a triage system. Diagnostic uncertainty in acute pancreatitis often arises from nonspecific abdominal symptoms and inter-reader variability in CECT interpretation. The DL model achieved high internal accuracy (AUROC 0.89) and maintained robust performance in an independent external validation cohort (AUROC 0.99).
R. Juglan,
M. Ligero,
Z. I. Carrero,
A. Rabasco Meneghetti,
T. Lenz,
L. Misera,
G. P. Veldhuizen,
P. Kuntke,
H. H. Kitzler,
S. Nebelung,
D. Truhn and
J. N. Kather,
"Generalizable and explainable deep learning for brain MRI: a multi-cohort evaluation of 3D architectures for age and sex prediction",
Brain informatics,
vol. 13,
no. 1,
Jul.
2026.
Abstract:
Deep learning (DL) methods increasingly outperform classical approaches in brain MRI analysis, yet their generalizability across independent imaging cohorts remains insufficiently evaluated. Because age and sex are fundamental neurobiological factors influencing brain structure and disease risk, this study systematically compares three three-dimensional architectures-Simple Fully Convolutional Network (SFCN), DenseNet121, and Swin Transformer-for age and sex prediction using T1-weighted MRI from four independent cohorts: UK Biobank (UKB, n = 47,390), Dallas Lifespan Brain Study (DLBS, n = 132), Parkinson's Progression Markers Initiative (PPMI, n = 108 controls), and Information eXtraction from Images (IXI, n = 319). SFCN consistently demonstrated the most robust performance. For sex classification, it achieved an AUC of 1.00 [1.00-1.00] in the UKB internal test set and 0.85-0.91 across external cohorts. For age prediction, SFCN achieved a mean absolute error (MAE) of 2.66 years (r = 0.89) internally and 4.98-5.81 years (r = 0.55-0.70) externally. Pairwise DeLong and Wilcoxon tests with Bonferroni correction confirmed significantly better performance of SFCN compared with Swin Transformer in most cohorts (p < 0.017). No significant demographic subgroup biases were observed. Explainability analyses further showed task-specific and spatially consistent attention patterns across cohorts. These findings demonstrate that simpler convolutional architectures can generalize more reliably than more complex attention-based models in multi-cohort settings. The study highlights the importance of external validation and emphasizes potential trade-offs between model complexity, robustness, and interpretability for clinically relevant neuroimaging applications.
| DOI: | 10.1186/s40708-026-00316-y |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42406258/ |
Abstract:
Deep learning (DL) methods increasingly outperform classical approaches in brain MRI analysis, yet their generalizability across independent imaging cohorts remains insufficiently evaluated. Because age and sex are fundamental neurobiological factors influencing brain structure and disease risk, this study systematically compares three three-dimensional architectures-Simple Fully Convolutional Network (SFCN), DenseNet121, and Swin Transformer-for age and sex prediction using T1-weighted MRI from four independent cohorts: UK Biobank (UKB, n = 47,390), Dallas Lifespan Brain Study (DLBS, n = 132), Parkinson's Progression Markers Initiative (PPMI, n = 108 controls), and Information eXtraction from Images (IXI, n = 319). SFCN consistently demonstrated the most robust performance. For sex classification, it achieved an AUC of 1.00 [1.00-1.00] in the UKB internal test set and 0.85-0.91 across external cohorts. For age prediction, SFCN achieved a mean absolute error (MAE) of 2.66 years (r = 0.89) internally and 4.98-5.81 years (r = 0.55-0.70) externally. Pairwise DeLong and Wilcoxon tests with Bonferroni correction confirmed significantly better performance of SFCN compared with Swin Transformer in most cohorts (p < 0.017). No significant demographic subgroup biases were observed. Explainability analyses further showed task-specific and spatially consistent attention patterns across cohorts. These findings demonstrate that simpler convolutional architectures can generalize more reliably than more complex attention-based models in multi-cohort settings. The study highlights the importance of external validation and emphasizes potential trade-offs between model complexity, robustness, and interpretability for clinically relevant neuroimaging applications.
P. Jacobs,
I. G. Steffen,
C. Ehrengut,
S. Bräuer,
H. Meyer and
T. Denecke,
"Interpretable Bayesian Modeling for Multireader Multicase Studies: Addressing Overdispersion and Limited Sample Size in Diagnostic Enhancement Evaluation",
Statistics in medicine,
vol. 45,
no. 15-17,
pp. e70666,
Jul.
2026.
Abstract:
The advent of machine and deep learning in the medical domain has led to significant advancements in diagnostic workflows and clinical decision-making, making rigorous evaluation of novel techniques essential for their integration into clinical practice. In this work, we introduce a Bayesian hierarchical Beta-Binomial modeling framework for estimating the effect of novel techniques in binary classification, with a particular focus on multireader, multicase study designs, motivated by applications in medical imaging. Some challenges in this context include, small sample sizes (i.e., only few readers particicipating in the study), pronounced overdispersion due to heterogeneity in reader performance, and class-imbalanced datasets. Addiotionally, the actual effect size of the novel technique may be small, further complicating robust estimation of model parameters. The proposed model explicitly accounts for overdispersion, addresses class imbalance within the test cohort, and incorporates prior information to regularize population-level parameter estimates across readers. Through simulation studies, the approach demonstrates improved robustness and lower estimation error compared to classical linear models, especially under high overdispersion and low sample sizes. Application to a real-world study of chest X-ray imaging with and without Bone Suppression Imaging enhancement illustrates the model's practical utility and highlights the importance of accounting for overdispersion and prior information in study design and analysis.
| DOI: | 10.1002/sim.70666 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42396812/ |
Abstract:
The advent of machine and deep learning in the medical domain has led to significant advancements in diagnostic workflows and clinical decision-making, making rigorous evaluation of novel techniques essential for their integration into clinical practice. In this work, we introduce a Bayesian hierarchical Beta-Binomial modeling framework for estimating the effect of novel techniques in binary classification, with a particular focus on multireader, multicase study designs, motivated by applications in medical imaging. Some challenges in this context include, small sample sizes (i.e., only few readers particicipating in the study), pronounced overdispersion due to heterogeneity in reader performance, and class-imbalanced datasets. Addiotionally, the actual effect size of the novel technique may be small, further complicating robust estimation of model parameters. The proposed model explicitly accounts for overdispersion, addresses class imbalance within the test cohort, and incorporates prior information to regularize population-level parameter estimates across readers. Through simulation studies, the approach demonstrates improved robustness and lower estimation error compared to classical linear models, especially under high overdispersion and low sample sizes. Application to a real-world study of chest X-ray imaging with and without Bone Suppression Imaging enhancement illustrates the model's practical utility and highlights the importance of accounting for overdispersion and prior information in study design and analysis.
M. Kesselmeier,
T. Thalheim,
F. Schmidt,
T. Peschel,
J. Palm,
A. Strübing,
A. Medek,
J. Przybilla,
A. M. Wermund,
R. Maas,
S. Härterich,
L. Redeker,
M. Federbusch,
D. Steinbach,
J. Gewehr,
M. Wurlitzer,
A. Riedel,
F. Meineke,
D. Neumann,
A. Scherag,
M. Loeffler and
O. B. O. Polar_mi,
"A distributed analysis approach for pharmacovigilance data from electronic medical records in German university hospitals: the POLAR_MI ETL Pipeline",
BMC Medical Informatics and Decision Making,
vol. 26,
no. 1,
Jun.
2026.
Abstract:
Das Projekt POLAR_MI untersuchte, welche Risiken durch die gleichzeitige Einnahme vieler Medikamente (Polypharmazie) bei erwachsenen Patientinnen und Patienten in deutschen Universitätskliniken entstehen können. Dafür wurden Routinedaten aus Krankenhäusern genutzt, ohne dass die Forschenden direkten Zugriff auf die persönlichen Patientendaten hatten, um den Datenschutz zu gewährleisten. Da die Daten und IT-Systeme der beteiligten Kliniken sehr unterschiedlich waren, mussten zunächst technische Lösungen entwickelt werden, um die Daten einheitlich auszuwerten. Mit dieser Methode konnten Daten von rund 500.000 Patientinnen und Patienten aus zehn Kliniken analysiert werden. Die Studie zeigt, dass große gemeinsame Analysen von Gesundheitsdaten in Deutschland möglich sind und liefert Erfahrungen sowie Werkzeuge, die auch für ähnliche Projekte in anderen Ländern genutzt werden können.
| DOI: | 10.1186/s12911-026-03550-w |
| Datei: | https://doi.org/10.1186/s12911-026-03550-w |
Abstract:
Das Projekt POLAR_MI untersuchte, welche Risiken durch die gleichzeitige Einnahme vieler Medikamente (Polypharmazie) bei erwachsenen Patientinnen und Patienten in deutschen Universitätskliniken entstehen können. Dafür wurden Routinedaten aus Krankenhäusern genutzt, ohne dass die Forschenden direkten Zugriff auf die persönlichen Patientendaten hatten, um den Datenschutz zu gewährleisten. Da die Daten und IT-Systeme der beteiligten Kliniken sehr unterschiedlich waren, mussten zunächst technische Lösungen entwickelt werden, um die Daten einheitlich auszuwerten. Mit dieser Methode konnten Daten von rund 500.000 Patientinnen und Patienten aus zehn Kliniken analysiert werden. Die Studie zeigt, dass große gemeinsame Analysen von Gesundheitsdaten in Deutschland möglich sind und liefert Erfahrungen sowie Werkzeuge, die auch für ähnliche Projekte in anderen Ländern genutzt werden können.
S. Sender,
P. J. Böttcher,
A. Klemenz,
T. Geyer,
J. Hillebrand,
M. Lütgens,
R. Lorbeer,
M. Manzke,
M. Weber and
F. G. Meinel,
"Deep-learning based image reconstruction enables reduced dose CT pulmonary angiography with non-inferior image quality",
Scientific reports,
vol. 16,
no. 1,
Jun.
2026.
Abstract:
To investigate whether a CT pulmonary angiography (CTPA) protocol with reduced radiation dose and deep-learning based image reconstruction (DLIR) is non-inferior in image quality to standard-dose CTPA using iterative reconstruction. A phantom study was conducted to estimate the additional radiation dose reduction enabled by high-strength deep learning-based image reconstruction (DLIR-H) compared to adaptive statistical iterative reconstruction (ASiR-V 90%). Medium and large phantoms were used to simulate different body sizes. Subsequently, we reduced radiation dose of our clinical CTPA protocol and transitioned to DLIR for image reconstruction. We retrospectively analyzed 307 consecutive patients who were examined before (n = 152) and after (n = 155) this clinically driven change in the CTPA protocol. Objective image quality was quantified and subjective image quality was rated by two radiologists. The non-inferiority margin was pre-specified as a < 5% difference in image quality parameters. In the phantom, DLIR-H allowed radiation dose to be reduced by up to 71% with equivalent or higher signal-to-noise-ratio (SNR) compared to standard-dose examinations reconstructed with ASiR-V 90%. In the patient cohort, radiation dose was reduced by 41% (median DLP 116 vs. 68 mGy*cm; effective dose 1.69 vs. 0.99 mSv, p < 0.001). In the modified protocol, median SNR was superior for the central pulmonary artery (13.6 vs. 22.3) and non-inferior for the segmental pulmonary arteries (16.4 vs. 16.8). Subjective image quality averaged over both readers was superior with the modified protocol. Compared to state-of-the-art iterative reconstruction, DLIR allows radiation dose for CTPA to be reduced by an additional 41% with non-inferior image quality.
| DOI: | 10.1038/s41598-026-56545-y |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42265315/ |
Abstract:
To investigate whether a CT pulmonary angiography (CTPA) protocol with reduced radiation dose and deep-learning based image reconstruction (DLIR) is non-inferior in image quality to standard-dose CTPA using iterative reconstruction. A phantom study was conducted to estimate the additional radiation dose reduction enabled by high-strength deep learning-based image reconstruction (DLIR-H) compared to adaptive statistical iterative reconstruction (ASiR-V 90%). Medium and large phantoms were used to simulate different body sizes. Subsequently, we reduced radiation dose of our clinical CTPA protocol and transitioned to DLIR for image reconstruction. We retrospectively analyzed 307 consecutive patients who were examined before (n = 152) and after (n = 155) this clinically driven change in the CTPA protocol. Objective image quality was quantified and subjective image quality was rated by two radiologists. The non-inferiority margin was pre-specified as a < 5% difference in image quality parameters. In the phantom, DLIR-H allowed radiation dose to be reduced by up to 71% with equivalent or higher signal-to-noise-ratio (SNR) compared to standard-dose examinations reconstructed with ASiR-V 90%. In the patient cohort, radiation dose was reduced by 41% (median DLP 116 vs. 68 mGy*cm; effective dose 1.69 vs. 0.99 mSv, p < 0.001). In the modified protocol, median SNR was superior for the central pulmonary artery (13.6 vs. 22.3) and non-inferior for the segmental pulmonary arteries (16.4 vs. 16.8). Subjective image quality averaged over both readers was superior with the modified protocol. Compared to state-of-the-art iterative reconstruction, DLIR allows radiation dose for CTPA to be reduced by an additional 41% with non-inferior image quality.
A. Jaiswal,
M. Rinneburger,
F. Meyer,
S. Arjune,
L. Oberlinkels,
P. A. Wawer Matos Reimer,
L. Lotter-Becker,
Ü. Akünal,
M. Bujotzek,
S. Denner,
K. Maier-Hein,
L. Stepansky,
M. S. May,
M. Habert,
A. Köhn,
M. Schöneck,
R. P. Wawer Matos Reimer,
R. Müller,
S. Lennartz,
N. Große Hokamp,
A. Bucher,
T. Persigehl and
L. Caldeira,
"Impact on Cost and Expert Time of Data-Efficient Deep Learning for Medical Image Segmentation",
Radiology. Artificial intelligence,
pp. e250200,
Jun.
2026.
Abstract:
Purpose To develop and systematically evaluate an iterative training approach, termed the expert-guided annotation loop, for efficient reference standard medical image segmentation, including assessment of two sample-selection strategies and real-world clinical implementation. Materials and Methods This retrospective study included ten datasets comprising 1948 CT or MRI examinations from autosomal dominant polycystic kidney disease, prostate cancer, uveal melanoma, thyroid eye disease, and non-small cell lung cancer patients. nnU-Net segmentation models were iteratively trained using an expert-guided annotation loop with random or active learning-based sample selection. In each iteration, additional samples were added to the training set, and model-generated presegmentations were corrected by expert radiologists to create reference standard annotations. Expert time required for manual segmentation versus presegmentations correction was measured. Model performance and efficiency were assessed using nonparametric tests, and cost savings were estimated for kidney and tumor segmentation using probabilistic sensitivity analysis. Feasibility of end-to-end no-code implementation was evaluated. Results Fifty-seven segmentation models were trained and evaluated. Final model Dice scores ranged from 0.67-0.97 for organ segmentation and from 0.64-0.69 for lung tumor segmentation across internal and external test sets. Maximum expert time savings were 90.3% for kidney and 48.2% for tumor segmentation (P < .001 and P = .003), corresponding to estimated per-examination cost savings of $14.30[95% CI: $5.94, $26.87] and $5.63[95% CI: $-7.26, $26.09], respectively. No-code execution of the expert-guided annotation loop was feasible. Conclusion The expert-guided annotation loop reduced expert annotation time and enabled estimated cost savings while producing high-quality reference standard CT and MRI segmentations. The no-code workflow was implemented in a clinical environment. © RSNA, 2026.
| DOI: | 10.1148/ryai.250200 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42340187/ |
Abstract:
Purpose To develop and systematically evaluate an iterative training approach, termed the expert-guided annotation loop, for efficient reference standard medical image segmentation, including assessment of two sample-selection strategies and real-world clinical implementation. Materials and Methods This retrospective study included ten datasets comprising 1948 CT or MRI examinations from autosomal dominant polycystic kidney disease, prostate cancer, uveal melanoma, thyroid eye disease, and non-small cell lung cancer patients. nnU-Net segmentation models were iteratively trained using an expert-guided annotation loop with random or active learning-based sample selection. In each iteration, additional samples were added to the training set, and model-generated presegmentations were corrected by expert radiologists to create reference standard annotations. Expert time required for manual segmentation versus presegmentations correction was measured. Model performance and efficiency were assessed using nonparametric tests, and cost savings were estimated for kidney and tumor segmentation using probabilistic sensitivity analysis. Feasibility of end-to-end no-code implementation was evaluated. Results Fifty-seven segmentation models were trained and evaluated. Final model Dice scores ranged from 0.67-0.97 for organ segmentation and from 0.64-0.69 for lung tumor segmentation across internal and external test sets. Maximum expert time savings were 90.3% for kidney and 48.2% for tumor segmentation (P < .001 and P = .003), corresponding to estimated per-examination cost savings of $14.30[95% CI: $5.94, $26.87] and $5.63[95% CI: $-7.26, $26.09], respectively. No-code execution of the expert-guided annotation loop was feasible. Conclusion The expert-guided annotation loop reduced expert annotation time and enabled estimated cost savings while producing high-quality reference standard CT and MRI segmentations. The no-code workflow was implemented in a clinical environment. © RSNA, 2026.
A. Surov,
A. Wienke,
M. Hinnerichs,
L. Sticklus,
R. Kloeckner,
M. Sieren,
F. G. Meinel,
J. Hillebrand,
N. Lubina,
C. Wollny,
T. Jorg,
M. Moos and
J. Borggrefe,
"Low Skeletal Muscle Mass Predicts 30-day Mortality in Patients With Acute Pulmonary Embolism. A Multicenter Study",
Academic radiology,
Jun.
2026.
Abstract:
Acute pulmonary embolism (APE) is a potentially life-threatening condition. We investigated the prognostic role of the skeletal musculature in patients with APE in a multicenter setting. The retrospective study comprised 1996 patients, 898 (45%) female; mean age 66.2 ± 15.8 years. Skeletal muscles were measured at the T12 level. Skeletal muscle area (SMA), skeletal muscle density (SMD), and intramuscular adipose tissue (IMAT) were quantified. SMA thresholds were used for the definition of low skeletal muscle mass (LSMM): 56.1 cm² for female patients and 92.3 cm² for male patients. Myosteatosis was defined as SMD < 31.3 HU for female patients and SMD < 37.5 HU for male patients. Sex-specific median cutoffs were used for IMAT. Uni- and multivariable logistic regression models were used to assess the impact of the skeletal musculature on clinical outcomes. LSMM was identified in 629 patients (31.5%) and myosteatosis in 1664 (83.4%). The 30-day mortality rate was 5%. LSMM was associated with 30-day mortality, OR= 2.21, 95%-CI (1.51 - 3.22), p=0.01 (multivariable analysis). The influence of LSMM on 30-day mortality was higher in female patients (OR= 2.89, 95%-CI (1.63 - 5.14), p<0.01) than in male patients (OR= 1.80, 95%-CI (1.11 - 2.94), p=0.02). Myosteatosis and IMAT were not associated with 30-day mortality in APE. LSMM is an independent and strong predictor of 30-day mortality in APE. LSMM has a higher influence on 30-day mortality in female patients. Thus, analysis of the skeletal musculature should be included in the clinical routine in APE.
| DOI: | 10.1016/j.acra.2026.05.013 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42259743/ |
Abstract:
Acute pulmonary embolism (APE) is a potentially life-threatening condition. We investigated the prognostic role of the skeletal musculature in patients with APE in a multicenter setting. The retrospective study comprised 1996 patients, 898 (45%) female; mean age 66.2 ± 15.8 years. Skeletal muscles were measured at the T12 level. Skeletal muscle area (SMA), skeletal muscle density (SMD), and intramuscular adipose tissue (IMAT) were quantified. SMA thresholds were used for the definition of low skeletal muscle mass (LSMM): 56.1 cm² for female patients and 92.3 cm² for male patients. Myosteatosis was defined as SMD < 31.3 HU for female patients and SMD < 37.5 HU for male patients. Sex-specific median cutoffs were used for IMAT. Uni- and multivariable logistic regression models were used to assess the impact of the skeletal musculature on clinical outcomes. LSMM was identified in 629 patients (31.5%) and myosteatosis in 1664 (83.4%). The 30-day mortality rate was 5%. LSMM was associated with 30-day mortality, OR= 2.21, 95%-CI (1.51 - 3.22), p=0.01 (multivariable analysis). The influence of LSMM on 30-day mortality was higher in female patients (OR= 2.89, 95%-CI (1.63 - 5.14), p<0.01) than in male patients (OR= 1.80, 95%-CI (1.11 - 2.94), p=0.02). Myosteatosis and IMAT were not associated with 30-day mortality in APE. LSMM is an independent and strong predictor of 30-day mortality in APE. LSMM has a higher influence on 30-day mortality in female patients. Thus, analysis of the skeletal musculature should be included in the clinical routine in APE.
M. Eicke,
K. Geissler,
J. Heepe,
M. B. Krueger,
H. B. Winther,
A. Heymann,
F. Knörr,
A. M. Bucher,
J. Herrmann,
C. Benoit,
H. Mentzel,
A. Schenk,
W. Wößmann,
D. M. Renz and
B. Lassen-Schmidt,
"Paediatric thoracal and abdominal organ segmentation in CT and MRI using global intensity non-linear augmentation",
European Journal of Radiology Artificial Intelligence,
vol. 6,
pp. 100096,
Jun.
2026.
| DOI: | 10.1016/j.ejrai.2026.100096 |
| Datei: | https://doi.org/10.1016/j.ejrai.2026.100096 |
M. P. Fischer,
A. Mayer,
A. Braun,
J. Kraft,
C. M. Hansen,
G. L. Baumgärtner,
P. Asbach,
A. R. Padhani,
T. Penzkofer,
F. König,
S. Awasthi,
M. R. Makowski,
S. Ripke and
C. A. Hamm,
"Polygenic risk scores for premagnetic resonance imaging risk stratification in men with clinically suspected prostate cancer",
Journal of the National Cancer Institute,
vol. 118,
no. 6,
pp. 1063-1072,
Jun.
2026.
Abstract:
Men with suspected prostate cancer undergo magnetic resonance imaging (MRI) before biopsy. However, approximately 30%-50% of MRIs are negative (Prostate Imaging-Reporting and Data System [PI-RADS] score 1-2), representing a challenge for MRI resource utilization. This study evaluates prostate cancer polygenic risk scores and clinical markers to optimize MRI utilization. In this prospective study, 500 cancer-suspected men of Western European descent scheduled for MRI (September 2017-December 2022) were enrolled. Exclusions included prior prostate cancer diagnosis, missing serum prostate-specific antigen (PSA), or PSA levels of at least 25 ng/mL. Patient-specific prostate cancer polygenic risk scores were calculated using genotype data obtained from saliva-derived DNA samples. Participants were grouped as MRI negative and positive (PI-RADS score 3-5). Logistic regression was used to calculate odds ratios (ORs) and to build multivariable risk models, including age, PSA, and polygenic risk scores for MRI positivity. Clinical utility was tested in a holdout test set using decision curve analysis. A total of 386 men (median age = 65 years, interquartile range [IQR] = 53-77 years) were eligible for analysis, which showed statistically significant associations between prostate cancer polygenic risk scores (OR = 1.56, 95% confidence interval [CI] = 1.23 to 1.98; P < .001) with MRI positivity, while PSA alone did not (OR = 1.17, 95% CI = 0.93 to 1.46; P = .18). The highest net benefit was shown using a multivariable age and prostate cancer polygenic risk score model, increasing the proportion of MRI-positive men by 14% compared with PSA alone (60% and 46%, respectively; P = .011). Genotype-informed risk stratification using prostate cancer polygenic risk scores could increase the proportion of cancer-suspicious findings at MRI, while identifying those who could safely avoid unnecessary MRI.
| DOI: | 10.1093/jnci/djag027 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41629765/ |
Abstract:
Men with suspected prostate cancer undergo magnetic resonance imaging (MRI) before biopsy. However, approximately 30%-50% of MRIs are negative (Prostate Imaging-Reporting and Data System [PI-RADS] score 1-2), representing a challenge for MRI resource utilization. This study evaluates prostate cancer polygenic risk scores and clinical markers to optimize MRI utilization. In this prospective study, 500 cancer-suspected men of Western European descent scheduled for MRI (September 2017-December 2022) were enrolled. Exclusions included prior prostate cancer diagnosis, missing serum prostate-specific antigen (PSA), or PSA levels of at least 25 ng/mL. Patient-specific prostate cancer polygenic risk scores were calculated using genotype data obtained from saliva-derived DNA samples. Participants were grouped as MRI negative and positive (PI-RADS score 3-5). Logistic regression was used to calculate odds ratios (ORs) and to build multivariable risk models, including age, PSA, and polygenic risk scores for MRI positivity. Clinical utility was tested in a holdout test set using decision curve analysis. A total of 386 men (median age = 65 years, interquartile range [IQR] = 53-77 years) were eligible for analysis, which showed statistically significant associations between prostate cancer polygenic risk scores (OR = 1.56, 95% confidence interval [CI] = 1.23 to 1.98; P < .001) with MRI positivity, while PSA alone did not (OR = 1.17, 95% CI = 0.93 to 1.46; P = .18). The highest net benefit was shown using a multivariable age and prostate cancer polygenic risk score model, increasing the proportion of MRI-positive men by 14% compared with PSA alone (60% and 46%, respectively; P = .011). Genotype-informed risk stratification using prostate cancer polygenic risk scores could increase the proportion of cancer-suspicious findings at MRI, while identifying those who could safely avoid unnecessary MRI.
W. Wällisch,
S. Dittrich,
A. Purbojo,
I. Schöffl,
T. Ganslandt,
H. Prokosch,
L. A. Kapsner and
J. M. Mang,
"Quality Assessment in Paediatric Cardiology: Experiences from Leveraging a Clinical Data Warehouse",
Life,
vol. 16,
no. 6,
pp. 941,
Jun.
2026.
Abstract:
Die Studie untersuchte die Daten von 1.700 Kindern und Jugendlichen, die zwischen 2011 und 2020 wegen angeborener Herzfehler operiert wurden. Ziel war es, einen Algorithmus zu entwickeln, der automatisch klinische Routinedaten auswertet und Risikofaktoren für Todesfälle nach Herzoperationen erkennt. Als besonders wichtige Risikofaktoren erwiesen sich der Einsatz einer Herz-Lungen-Unterstützung (ECMO), ein Geburts- bzw. Körpergewicht unter 2.500 Gramm, bestimmte schwere Formen angeborener Herzfehler sowie auffällige Nierenwerte (Kreatinin-Verhältnis). Neben der Bestätigung bereits bekannter Risikofaktoren zeigte die Studie, dass das Kreatinin-Verhältnis ein besonders aussagekräftiger Laborwert zur Abschätzung des Sterberisikos sein kann. Das entwickelte Verfahren könnte künftig dabei helfen, Behandlungsqualität und Patientenrisiken anhand von Routinedaten schnell, effizient und datenschutzkonform auch über mehrere Kliniken hinweg zu bewerten.
| DOI: | 10.3390/life16060941 |
| Datei: | https://doi.org/10.3390/life16060941 |
Abstract:
Die Studie untersuchte die Daten von 1.700 Kindern und Jugendlichen, die zwischen 2011 und 2020 wegen angeborener Herzfehler operiert wurden. Ziel war es, einen Algorithmus zu entwickeln, der automatisch klinische Routinedaten auswertet und Risikofaktoren für Todesfälle nach Herzoperationen erkennt. Als besonders wichtige Risikofaktoren erwiesen sich der Einsatz einer Herz-Lungen-Unterstützung (ECMO), ein Geburts- bzw. Körpergewicht unter 2.500 Gramm, bestimmte schwere Formen angeborener Herzfehler sowie auffällige Nierenwerte (Kreatinin-Verhältnis). Neben der Bestätigung bereits bekannter Risikofaktoren zeigte die Studie, dass das Kreatinin-Verhältnis ein besonders aussagekräftiger Laborwert zur Abschätzung des Sterberisikos sein kann. Das entwickelte Verfahren könnte künftig dabei helfen, Behandlungsqualität und Patientenrisiken anhand von Routinedaten schnell, effizient und datenschutzkonform auch über mehrere Kliniken hinweg zu bewerten.
M. A. Fink,
C. Römer,
M. Ebel,
E. Frodl,
D. P. Dos Santos,
H. Kauczor and
A. Bucher,
"Radiologists' memory as a data protection risk: a worst-case stress test for chest radiograph re-identification",
Insights into imaging,
vol. 17,
no. 1,
Jun.
2026.
Abstract:
To estimate an upper bound of memory-based re-identification risk for chest radiographs by testing radiologists under conditions that favor recognition. In this prospective, multicenter, web-based reader study, radiologists from 38 centers completed two reading phases. In Phase 1, each reader interpreted ten chest radiographs. After a minimum interval of 24 h, Phase 2 included six follow-up target examinations and six new non-target examinations (50% target prevalence). After each Phase-2 examination, readers indicated whether they remembered the patient. Following a positive response, they were asked separately whether they remembered the Phase-1 pseudonym and/or case position. Thirty-three readers with fully classifiable Phase-2 data contributed 396 Phase-2 examinations to the predefined primary analysis. Readers answered "remember" in 139 of 396 examinations (35.1%). Sensitivity for repeated target examinations was 50.0% (99/198), whereas 20.2% of new non-target examinations were nevertheless judged as remembered (40/198). Explicit identifiers were attempted in 23 of 396 examinations (5.8%). At least one explicit identifier, defined as the Phase-1 pseudonym and/or case position, was correct in five of 396 examinations (1.3%). In a low-prevalence model with one known patient per dataset, the positive predictive value of a "remember" response decreased from 2.44% in datasets of 100 radiographs to 0.25% in datasets of 1000 radiographs. Even in a design that favored memory, correct recall of explicit identifiers was rare, whereas false-positive recognition remained common. These findings support treating radiologists' memory as a limited, upper-bound component of re-identification risk, rather than assuming that familiarity routinely translates into identification. Even under conditions deliberately favoring memory, radiologists rarely converted familiarity with prior chest radiographs into correct explicit identifiers; in low-prevalence datasets, false-positive recognition dominated the practical meaning of a "remember" judgment. How often does a radiologist's feeling of familiarity with a previously seen chest radiograph translate into correct recall of a usable explicit identifier? In a deliberately memory-favoring design, sensitivity was 50.0%, but 20.2% of new examinations were false positives and only 1.3% yielded a correct explicit identifier.
| DOI: | 10.1186/s13244-026-02336-y |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42301366/ |
Abstract:
To estimate an upper bound of memory-based re-identification risk for chest radiographs by testing radiologists under conditions that favor recognition. In this prospective, multicenter, web-based reader study, radiologists from 38 centers completed two reading phases. In Phase 1, each reader interpreted ten chest radiographs. After a minimum interval of 24 h, Phase 2 included six follow-up target examinations and six new non-target examinations (50% target prevalence). After each Phase-2 examination, readers indicated whether they remembered the patient. Following a positive response, they were asked separately whether they remembered the Phase-1 pseudonym and/or case position. Thirty-three readers with fully classifiable Phase-2 data contributed 396 Phase-2 examinations to the predefined primary analysis. Readers answered "remember" in 139 of 396 examinations (35.1%). Sensitivity for repeated target examinations was 50.0% (99/198), whereas 20.2% of new non-target examinations were nevertheless judged as remembered (40/198). Explicit identifiers were attempted in 23 of 396 examinations (5.8%). At least one explicit identifier, defined as the Phase-1 pseudonym and/or case position, was correct in five of 396 examinations (1.3%). In a low-prevalence model with one known patient per dataset, the positive predictive value of a "remember" response decreased from 2.44% in datasets of 100 radiographs to 0.25% in datasets of 1000 radiographs. Even in a design that favored memory, correct recall of explicit identifiers was rare, whereas false-positive recognition remained common. These findings support treating radiologists' memory as a limited, upper-bound component of re-identification risk, rather than assuming that familiarity routinely translates into identification. Even under conditions deliberately favoring memory, radiologists rarely converted familiarity with prior chest radiographs into correct explicit identifiers; in low-prevalence datasets, false-positive recognition dominated the practical meaning of a "remember" judgment. How often does a radiologist's feeling of familiarity with a previously seen chest radiograph translate into correct recall of a usable explicit identifier? In a deliberately memory-favoring design, sensitivity was 50.0%, but 20.2% of new examinations were false positives and only 1.3% yielded a correct explicit identifier.
C. Römer,
W. Heindel,
A. Helfen,
M. Köhler,
R. Eßeling,
N. Meier,
J. Varghese and
G. H. Pöhler,
"RAVEN: An open-source assessment framework for automated entity extraction in structured radiology reporting",
Informatics in Medicine Unlocked,
vol. 63,
pp. 101755,
Jun.
2026.
| DOI: | 10.1016/j.imu.2026.101755 |
| Datei: | https://doi.org/10.1016/j.imu.2026.101755 |
C. A. Hamm,
A. Padhani and
S. Ripke,
"Response to Sud and Houlston",
Journal of the National Cancer Institute,
Jun.
2026.
Abstract:
N/A
| DOI: | 10.1093/jnci/djag180 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42283081/ |
Abstract:
N/A
C. A. Hamm,
E. Yuan,
G. L. Baumgärtner,
A. B. Dias,
J. Zawaideh,
G. Brembilla,
Y. Arita,
M. Rooij,
R. Cuocolo and
A. Ponsiglione,
"Shifting from black box decisions to informed decision-making in using artificial intelligence to analyze prostate MRI",
Abdominal radiology (New York),
Jun.
2026.
Abstract:
Prostate magnetic resonance imaging (MRI) is central to the detection and localization of clinically significant prostate cancers. However, its interpretation is time-consuming and variable. Artificial intelligence (AI) tools for prostate MRI are rapidly emerging, with their performance approaching that of expert readers in select settings, although robust external validation in unselected real-world cohorts remains incomplete. Moreover, many deep learning models are limited in terms of transparency; that is, the "black box" problem hinders trust, safe deployment, and regulatory acceptance. This review summarizes the core concepts of interpretability and explainable artificial intelligence (XAI), highlights commonly used approaches to explain the output of AI models, and discusses how the lack of transparency hinders clinical deployment and how this can be addressed using standardized reporting frameworks. Ultimately, the purpose of this review is to draw attention to the pressing need for XAI and to spark interest and awareness of how standardized frameworks may help us to shift from "black box" decisions to informed decision-making in clinical practice.
| DOI: | 10.1007/s00261-026-05644-5 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42364039/ |
Abstract:
Prostate magnetic resonance imaging (MRI) is central to the detection and localization of clinically significant prostate cancers. However, its interpretation is time-consuming and variable. Artificial intelligence (AI) tools for prostate MRI are rapidly emerging, with their performance approaching that of expert readers in select settings, although robust external validation in unselected real-world cohorts remains incomplete. Moreover, many deep learning models are limited in terms of transparency; that is, the "black box" problem hinders trust, safe deployment, and regulatory acceptance. This review summarizes the core concepts of interpretability and explainable artificial intelligence (XAI), highlights commonly used approaches to explain the output of AI models, and discusses how the lack of transparency hinders clinical deployment and how this can be addressed using standardized reporting frameworks. Ultimately, the purpose of this review is to draw attention to the pressing need for XAI and to spark interest and awareness of how standardized frameworks may help us to shift from "black box" decisions to informed decision-making in clinical practice.
S. Eminovic,
R. Schmidt,
B. Levita,
M. Lindholz,
A. Haack,
A. Burdenski,
M. Bui,
I. T. Schobert,
A. Dell'Orco,
J. Nawabi and
T. Penzkofer,
"Supporting Radiology Resident Education and Clinical Decision-Making With Large Language Models: Comparative Study of Reasoning Models DeepSeek-R1 and ChatGPT-o1",
JMIR AI,
vol. 5,
pp. e86974,
Jun.
2026.
Abstract:
Radiology trainees require efficient, accurate, and accessible resources to master complex imaging techniques and identify findings that guide clinical decision-making. Large language models (LLMs) are emerging as promising tools for medical education and clinical workflows, offering the potential to enhance learning by providing instant feedback, aiding in diagnostic accuracy, and offering personalized learning experiences. However, systematic comparisons of LLMs for radiology education and clinical support remain limited, particularly regarding differences across subspecialties and resident experience levels. This study aimed to evaluate and compare the response quality of 2 state-of-the-art reasoning-based LLMs, namely DeepSeek-R1 and ChatGPT-o1, as clinical and radiology residency support tools, comparing performance across clinical and didactic dimensions, including text- and image-based responses. Overall, 27 radiology questions covering 9 radiological subspecialties were answered by both LLMs. Additionally, 6 image-based questions were presented only to ChatGPT-o1 due to its image processing capabilities. Responses were independently rated by 7 radiology residents (postgraduate years 2-5) across 9 rating criteria grouped into 3 dimensions (factual accuracy, clinical practicality, and didactic value), using a 5-point Likert scale. Statistics compared LLMs, reader experience, and response types for text- as well as image-based for ChatGPT-o1 queries. DeepSeek-R1 consistently outperformed ChatGPT-o1 across all rating dimensions, with highly significant differences across all criteria (mean ratings: DeepSeek-R1 4.51, SD 0.73 vs ChatGPT-o1 3.73, SD 0.98; P<.001). In an exploratory subspecialty-level analysis, DeepSeek-R1 descriptively outperformed ChatGPT-o1 across all subspecialties. For both LLMs accumulated, junior residents tended to rate slightly higher than seniors in 7 of 9 criteria, although differences were not statistically significant. However, for ChatGPT-o1, junior residents rated significantly higher in overall score across all criteria (juniors 3.81, SD 0.64 vs seniors 3.63, SD 0.65; P=.02). Image-based responses by ChatGPT-o1 scored significantly lower than text-based (mean 3.19, SD 1.42; P=.007), particularly in factual accuracy (mean 2.75, SD 1.45; P<.001) and clinical practicality (mean 3.11, SD 1.47; P=.03). Both DeepSeek-R1 and ChatGPT-o1 demonstrate promising potential on simulated radiology question sets designed for educational and clinical contexts, with DeepSeek-R1 outperforming ChatGPT-o1 across all evaluated criteria. These results emphasize the value of open-source models for educational use and provide early evidence that LLMs may support radiology resident training under controlled conditions; however, their real-world educational and clinical effects require further investigation. Future research should prospectively evaluate how LLMs can be integrated into radiology training, assess their impact alongside conventional teaching methods, and investigate multimodal capabilities to better reflect realistic clinical scenarios.
| DOI: | 10.2196/86974 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42361338/ |
Abstract:
Radiology trainees require efficient, accurate, and accessible resources to master complex imaging techniques and identify findings that guide clinical decision-making. Large language models (LLMs) are emerging as promising tools for medical education and clinical workflows, offering the potential to enhance learning by providing instant feedback, aiding in diagnostic accuracy, and offering personalized learning experiences. However, systematic comparisons of LLMs for radiology education and clinical support remain limited, particularly regarding differences across subspecialties and resident experience levels. This study aimed to evaluate and compare the response quality of 2 state-of-the-art reasoning-based LLMs, namely DeepSeek-R1 and ChatGPT-o1, as clinical and radiology residency support tools, comparing performance across clinical and didactic dimensions, including text- and image-based responses. Overall, 27 radiology questions covering 9 radiological subspecialties were answered by both LLMs. Additionally, 6 image-based questions were presented only to ChatGPT-o1 due to its image processing capabilities. Responses were independently rated by 7 radiology residents (postgraduate years 2-5) across 9 rating criteria grouped into 3 dimensions (factual accuracy, clinical practicality, and didactic value), using a 5-point Likert scale. Statistics compared LLMs, reader experience, and response types for text- as well as image-based for ChatGPT-o1 queries. DeepSeek-R1 consistently outperformed ChatGPT-o1 across all rating dimensions, with highly significant differences across all criteria (mean ratings: DeepSeek-R1 4.51, SD 0.73 vs ChatGPT-o1 3.73, SD 0.98; P<.001). In an exploratory subspecialty-level analysis, DeepSeek-R1 descriptively outperformed ChatGPT-o1 across all subspecialties. For both LLMs accumulated, junior residents tended to rate slightly higher than seniors in 7 of 9 criteria, although differences were not statistically significant. However, for ChatGPT-o1, junior residents rated significantly higher in overall score across all criteria (juniors 3.81, SD 0.64 vs seniors 3.63, SD 0.65; P=.02). Image-based responses by ChatGPT-o1 scored significantly lower than text-based (mean 3.19, SD 1.42; P=.007), particularly in factual accuracy (mean 2.75, SD 1.45; P<.001) and clinical practicality (mean 3.11, SD 1.47; P=.03). Both DeepSeek-R1 and ChatGPT-o1 demonstrate promising potential on simulated radiology question sets designed for educational and clinical contexts, with DeepSeek-R1 outperforming ChatGPT-o1 across all evaluated criteria. These results emphasize the value of open-source models for educational use and provide early evidence that LLMs may support radiology resident training under controlled conditions; however, their real-world educational and clinical effects require further investigation. Future research should prospectively evaluate how LLMs can be integrated into radiology training, assess their impact alongside conventional teaching methods, and investigate multimodal capabilities to better reflect realistic clinical scenarios.
S. Eminovic,
M. Lindholz,
A. Dell'Orco,
A. Meddeb,
G. L. Baumgärtner,
S. Schulze-Weddige,
E. Siebert,
M. P. Wattjes,
T. Penzkofer and
J. Nawabi,
"Comparing performance of seven fine-tuned open-source large language models in summarizing and predicting outcome-relevant information from mechanical thrombectomy reports in patients with acute ischemic stroke",
European radiology,
vol. 36,
no. 5,
pp. 3661-3675,
Mai
2026.
Abstract:
This study evaluates seven open-source Large Language Models (LLMs) in summarizing radiology reports of acute ischemic stroke patients treated with mechanical thrombectomy and predicting angiography-based outcome measures relevant to post-thrombectomy reperfusion. 2000 mechanical thrombectomy reports (findings and summarizing impression section as gold standard) were split into training set (N = 1900) for model fine-tuning and test set (N = 100). A two-step evaluation was performed: (1) Quantitative analyses of seven LLMs with metrics ROUGE-1, -2, -L, METEOR, BERTScore (F1) and BLEU comparing LLM-generated summaries against gold-standard impressions. (2) Qualitative manual evaluation of the four best-performing models by two radiologists, assessing correctness and completeness across key parameters: outcome-relevant scores, vessel information, occlusion side, number of passes, relevant additional information, hallucinations, and grammar quality. Statistical significance was assessed via a two-tailed, four-sample χ² test, followed by post hoc pairwise χ² comparisons. BioMistral-7b scored highest across most quantitative metrics (ROUGE-1: 0.47, ROUGE-2: 0.30, ROUGE-L: 0.43, METEOR: 0.46, BERTScore (F1): 0.82). Manual evaluation revealed gemma-2-9b most frequently documented pass counts (56 out of 100 cases (56%); p < 0.02 vs. Llama-3.1-8b/mistral-7b-instruct), while mistral-7b-instruct described them most often correctly (29 out of 38 mentioned passes (76.32%); p < 0.02 vs. BioMistral-7b and p < 0.01 vs. gemma-2-9b). All four manually evaluated LLMs performed moderately well in predicting "Thrombolysis-In-Cerebral-Ischemia (TICI)" Score (correctness rate ranging from 66 to 71%; p = 0.89). All four manually evaluated LLMs effectively summarized thrombectomy reports and demonstrated moderate accuracy predicting TICI scores. Their integration into radiology workflows could enhance efficiency, warranting further clinical validation. Question Specifically fine-tuned Large Language Models (LLMs) can improve radiology workflow by automatically summarizing thrombectomy reports and inferring angiographic classifications from textual descriptions. Findings Fine-tuned LLMs achieve similar performance in summarizing thrombectomy reports, with each model performing best in specific categories and showing moderate accuracy in correct "Thrombolysis-In-Cerebral-Ischemia (TICI)" Score prediction (66-71%). Clinical relevance Integrating fine-tuned LLMs into radiology workflows may accelerate decision-making and improve patient outcomes by automatically summarizing reports and assessing recanalization success, while future work should enhance contextual understanding, address ambiguous inputs, and limit hallucinations.
| DOI: | 10.1007/s00330-025-12122-x |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41249547/ |
Abstract:
This study evaluates seven open-source Large Language Models (LLMs) in summarizing radiology reports of acute ischemic stroke patients treated with mechanical thrombectomy and predicting angiography-based outcome measures relevant to post-thrombectomy reperfusion. 2000 mechanical thrombectomy reports (findings and summarizing impression section as gold standard) were split into training set (N = 1900) for model fine-tuning and test set (N = 100). A two-step evaluation was performed: (1) Quantitative analyses of seven LLMs with metrics ROUGE-1, -2, -L, METEOR, BERTScore (F1) and BLEU comparing LLM-generated summaries against gold-standard impressions. (2) Qualitative manual evaluation of the four best-performing models by two radiologists, assessing correctness and completeness across key parameters: outcome-relevant scores, vessel information, occlusion side, number of passes, relevant additional information, hallucinations, and grammar quality. Statistical significance was assessed via a two-tailed, four-sample χ² test, followed by post hoc pairwise χ² comparisons. BioMistral-7b scored highest across most quantitative metrics (ROUGE-1: 0.47, ROUGE-2: 0.30, ROUGE-L: 0.43, METEOR: 0.46, BERTScore (F1): 0.82). Manual evaluation revealed gemma-2-9b most frequently documented pass counts (56 out of 100 cases (56%); p < 0.02 vs. Llama-3.1-8b/mistral-7b-instruct), while mistral-7b-instruct described them most often correctly (29 out of 38 mentioned passes (76.32%); p < 0.02 vs. BioMistral-7b and p < 0.01 vs. gemma-2-9b). All four manually evaluated LLMs performed moderately well in predicting "Thrombolysis-In-Cerebral-Ischemia (TICI)" Score (correctness rate ranging from 66 to 71%; p = 0.89). All four manually evaluated LLMs effectively summarized thrombectomy reports and demonstrated moderate accuracy predicting TICI scores. Their integration into radiology workflows could enhance efficiency, warranting further clinical validation. Question Specifically fine-tuned Large Language Models (LLMs) can improve radiology workflow by automatically summarizing thrombectomy reports and inferring angiographic classifications from textual descriptions. Findings Fine-tuned LLMs achieve similar performance in summarizing thrombectomy reports, with each model performing best in specific categories and showing moderate accuracy in correct "Thrombolysis-In-Cerebral-Ischemia (TICI)" Score prediction (66-71%). Clinical relevance Integrating fine-tuned LLMs into radiology workflows may accelerate decision-making and improve patient outcomes by automatically summarizing reports and assessing recanalization success, while future work should enhance contextual understanding, address ambiguous inputs, and limit hallucinations.
C. A. Hamm,
T. Rüterhenke,
G. L. Baumgärtner,
C. Mayerosch,
S. Schallenberg,
M. Lindholz,
N. L. Beetz,
P. Asbach and
T. Penzkofer,
"Imaging-based clinical decision tree enables risk stratification of extraprostatic extension before radical prostatectomy in prostate cancer patients",
Insights into imaging,
vol. 17,
no. 1,
Mai
2026.
Abstract:
Extraprostatic extension (EPE) significantly impacts surgical planning for prostate cancer (PCa) patients, influencing nerve-sparing surgery and neoadjuvant therapy decisions. However, Likert-scale-based radiological (r)EPE assessment lacks sufficient diagnostic accuracy for reliable clinical decision-making. Therefore, the aim was to evaluate rEPE scoring alongside clinical parameters to develop a clinically feasible decision tree for preoperative risk stratification. This retrospective single-center study included 429 consecutive PCa patients undergoing radical prostatectomy between January 2012 and October 2018. All patients underwent multiparametric MRI with PI-RADS scoring and rEPE grading (grades 0-3). Clinical parameters included PSA density (PSAD) and ISUP grade group (GG) at biopsy. Univariate and multivariate logistic regression identified predictors of EPE. A clinical decision tree was developed using binary classification to stratify patients into risk groups. EPE was confirmed in 145 patients (33.8%). Multivariate analysis identified rEPE grade (OR 2.64, p < 0.001) and GG at biopsy (OR 1.41, p < 0.001) as independent predictors. The decision tree assigned 48% of patients to the high-risk (rEPE grade 3: 89% EPE risk) and low-risk group (rEPE < 3 + PSAD < 0.2 ng/mL² + GG < 4: 13% EPE risk), while 52% showed intermediate risk (28-45% EPE risk). The developed decision tree combining MRI-derived rEPE grading, PSAD, and biopsy GG enables reliable identification of patients at high and low risk for EPE. This tool supports informed decision-making regarding nerve-sparing surgery and neoadjuvant therapy, potentially contributing to personalized treatment planning. Decision tree combining routine MRI-based and clinical markers reliably stratifies prostate cancer patients into high-risk and low-risk groups for EPE, supporting personalized surgical planning. EPE affects surgical planning decisions in prostate cancer patients. Combining EPE grade at MRI, PSAD, and biopsy grade improves risk stratification. The developed decision tree reliably stratified every second patient into distinct EPE-risk groups, potentially improving personalized surgical planning.
| DOI: | 10.1186/s13244-026-02305-5 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42129021/ |
Abstract:
Extraprostatic extension (EPE) significantly impacts surgical planning for prostate cancer (PCa) patients, influencing nerve-sparing surgery and neoadjuvant therapy decisions. However, Likert-scale-based radiological (r)EPE assessment lacks sufficient diagnostic accuracy for reliable clinical decision-making. Therefore, the aim was to evaluate rEPE scoring alongside clinical parameters to develop a clinically feasible decision tree for preoperative risk stratification. This retrospective single-center study included 429 consecutive PCa patients undergoing radical prostatectomy between January 2012 and October 2018. All patients underwent multiparametric MRI with PI-RADS scoring and rEPE grading (grades 0-3). Clinical parameters included PSA density (PSAD) and ISUP grade group (GG) at biopsy. Univariate and multivariate logistic regression identified predictors of EPE. A clinical decision tree was developed using binary classification to stratify patients into risk groups. EPE was confirmed in 145 patients (33.8%). Multivariate analysis identified rEPE grade (OR 2.64, p < 0.001) and GG at biopsy (OR 1.41, p < 0.001) as independent predictors. The decision tree assigned 48% of patients to the high-risk (rEPE grade 3: 89% EPE risk) and low-risk group (rEPE < 3 + PSAD < 0.2 ng/mL² + GG < 4: 13% EPE risk), while 52% showed intermediate risk (28-45% EPE risk). The developed decision tree combining MRI-derived rEPE grading, PSAD, and biopsy GG enables reliable identification of patients at high and low risk for EPE. This tool supports informed decision-making regarding nerve-sparing surgery and neoadjuvant therapy, potentially contributing to personalized treatment planning. Decision tree combining routine MRI-based and clinical markers reliably stratifies prostate cancer patients into high-risk and low-risk groups for EPE, supporting personalized surgical planning. EPE affects surgical planning decisions in prostate cancer patients. Combining EPE grade at MRI, PSAD, and biopsy grade improves risk stratification. The developed decision tree reliably stratified every second patient into distinct EPE-risk groups, potentially improving personalized surgical planning.
C. Fischer,
C. Gulden,
D. Quell,
J. Ziegler,
D. Hellwig,
T. Ganslandt,
T. Pukrop and
P. Pallaoro,Multi-Centric Quality Analysis of Oncological Data in FHIR.
....
Mai
2026.
Abstract:
Für aussagekräftige Forschung mit Daten ist eine hohe Datenqualität sehr wichtig. Je komplexer die Systeme sind, aus denen die Daten stammen, und je mehr Verarbeitungsschritte die Daten durchlaufen, desto größer ist das Risiko für Fehler oder unvollständige Informationen. Solche Qualitätsprobleme können die Ergebnisse von Studien beeinträchtigen. Auch bei standardisierten medizinischen Datensätzen, wie dem onkologischen Basisdatensatz (oBDS), muss deshalb überprüft werden, ob die Daten tatsächlich zuverlässig sind. In dieser Studie wurde die Qualität von Krebsdaten untersucht, die aus sechs deutschen Universitätskliniken stammen und in das FHIR-Format überführt wurden. Dafür wurden automatische Prüfungen mit den Tools Pathling und Great Expectations eingesetzt. Diese überprüften verschiedene Aspekte der Datenqualität, zum Beispiel Vollständigkeit und Plausibilität der Daten. Die Untersuchung zeigte, dass die Datenqualität zwischen den einzelnen Kliniken unterschiedlich ausfällt. Obwohl alle Einrichtungen denselben standardisierten Datensatz verwenden, kann also nicht automatisch von einer gleich hohen Datenqualität ausgegangen werden. Die Ergebnisse sollen künftig in die Real-World-Data-Plattform des Bayerischen Zentrums für Krebsforschung (BZKF) integriert werden. Dadurch erhalten Forschende besseren Zugang zu qualitativ hochwertigen Krebsdaten für wissenschaftliche Studien.
| DOI: | 10.3233/shti260400 |
| Datei: | https://doi.org/10.3233/shti260400 |
Abstract:
Für aussagekräftige Forschung mit Daten ist eine hohe Datenqualität sehr wichtig. Je komplexer die Systeme sind, aus denen die Daten stammen, und je mehr Verarbeitungsschritte die Daten durchlaufen, desto größer ist das Risiko für Fehler oder unvollständige Informationen. Solche Qualitätsprobleme können die Ergebnisse von Studien beeinträchtigen. Auch bei standardisierten medizinischen Datensätzen, wie dem onkologischen Basisdatensatz (oBDS), muss deshalb überprüft werden, ob die Daten tatsächlich zuverlässig sind. In dieser Studie wurde die Qualität von Krebsdaten untersucht, die aus sechs deutschen Universitätskliniken stammen und in das FHIR-Format überführt wurden. Dafür wurden automatische Prüfungen mit den Tools Pathling und Great Expectations eingesetzt. Diese überprüften verschiedene Aspekte der Datenqualität, zum Beispiel Vollständigkeit und Plausibilität der Daten. Die Untersuchung zeigte, dass die Datenqualität zwischen den einzelnen Kliniken unterschiedlich ausfällt. Obwohl alle Einrichtungen denselben standardisierten Datensatz verwenden, kann also nicht automatisch von einer gleich hohen Datenqualität ausgegangen werden. Die Ergebnisse sollen künftig in die Real-World-Data-Plattform des Bayerischen Zentrums für Krebsforschung (BZKF) integriert werden. Dadurch erhalten Forschende besseren Zugang zu qualitativ hochwertigen Krebsdaten für wissenschaftliche Studien.
J. Bankole,
M. Anywar,
J. Heykendorf,
K. Lütje,
H. Riemenschneider,
J. Rupp,
P. Ross and
B. Schreiweis,Timely, AQL-Driven clinical cohort identification in OpenEHR infrastructures.
....
Mai
2026.
Abstract:
Die zeitnahe Identifizierung von Kohorten stellt in Krankenhäusern aufgrund fragmentierter Systeme, verzögerter Kodierung und begrenzter IT-Kapazitäten nach wie vor eine Herausforderung dar. Wir stellen CohortMailer vor, ein schlankes Tool, das die Extraktion von Patientenkohorten aus openEHR-Repositorien mithilfe der Archetype Query Language (AQL) automatisiert und gefilterte Falllisten per E-Mail nach einem festgelegten Zeitplan versendet. Das System wurde an zwei Standorten des Universitätsklinikums Schleswig-Holstein (UKSH) eingesetzt und führte strukturierte Abfragen durch, wandte regelbasierte Filter an und verteilte tabellarische Zusammenfassungen an die Studienteams. Während einer 30-tägigen Evaluierung erreichte CohortMailer eine Verfügbarkeit von 83,3 % und identifizierte 274 einzelne Patienten. In der Atemwegskohorte erreichte es einen positiven prädiktiven Wert (PPV) von 62,9 % und eine Sensitivität von 50,3 %. Die Ergebnisse zeigen, wie modulare, standardbasierte Automatisierung den manuellen Screening-Aufwand reduzieren und klinische Arbeitsabläufe in Echtzeit unterstützen kann.
| DOI: | 10.3233/shti260415 |
| Datei: | https://doi.org/10.3233/shti260415 |
Abstract:
Die zeitnahe Identifizierung von Kohorten stellt in Krankenhäusern aufgrund fragmentierter Systeme, verzögerter Kodierung und begrenzter IT-Kapazitäten nach wie vor eine Herausforderung dar. Wir stellen CohortMailer vor, ein schlankes Tool, das die Extraktion von Patientenkohorten aus openEHR-Repositorien mithilfe der Archetype Query Language (AQL) automatisiert und gefilterte Falllisten per E-Mail nach einem festgelegten Zeitplan versendet. Das System wurde an zwei Standorten des Universitätsklinikums Schleswig-Holstein (UKSH) eingesetzt und führte strukturierte Abfragen durch, wandte regelbasierte Filter an und verteilte tabellarische Zusammenfassungen an die Studienteams. Während einer 30-tägigen Evaluierung erreichte CohortMailer eine Verfügbarkeit von 83,3 % und identifizierte 274 einzelne Patienten. In der Atemwegskohorte erreichte es einen positiven prädiktiven Wert (PPV) von 62,9 % und eine Sensitivität von 50,3 %. Die Ergebnisse zeigen, wie modulare, standardbasierte Automatisierung den manuellen Screening-Aufwand reduzieren und klinische Arbeitsabläufe in Echtzeit unterstützen kann.
S. Mayr,
S. Zicha,
D. Giese,
M. Kopp,
S. Schmidt,
S. Arndt,
L. Sommerfeld,
J. Kleiss,
M. Uder and
M. S. May,
"Cardiac Magnetic Resonance Imaging at Lower Field Strength: A Comparison of Biventricular Cine and Quantitative T1 and T2 Maps at 0.55T and 1.5T",
Investigative radiology,
vol. 61,
no. 4,
pp. 251-260,
Apr.
2026.
Abstract:
This study aimed to compare biventricular cine measurements and quantitative maps of myocardial T1 and T2 relaxation times at field strengths ranging from 0.55T to 1.5T in patients with a clinical indication for cardiac magnetic resonance imaging (CMR). Establishing the feasibility of low-field CMR may improve accessibility due to easier siting and lower cost of the low-field systems. Thirteen patients underwent same-day comparative CMR at 0.55T and on a commercial scanner at 1.5T. We examined all individuals with breath-held segmented bSSFP cine sequences for volumetric assessment of the left ventricle (LV), right ventricle (RV), and visual assessment of wall motion abnormalities (WMA) and valve pathologies. The quantitative T1 and T2 maps were acquired in 3 short-axis views for tissue characterization. Blinded readers scored the image quality on a 3-point Likert scale. Ten healthy volunteers were additionally examined at 0.55T to obtain reference values for the parametric maps. Functional analysis of both ventricles at low-field CMR has a reasonable correlation (r=0.94 to 0.99) with conventional 1.5T measurements. LV ejection fraction (EF) ( P =0.62) and RV measurements (all P >0.05) were highly reproducible, but LV absolute volumetric measurements were slightly lower at the low-field strength (all P <0.05). T1 and T2 relaxation times correlated strongly between field strengths (r=0.79; P <0.01 and r=0.63; P =0.02). Reference values from the volunteers were 678.6±13.5 ms (T1) and 66.5±4.1 ms (T2). Overall, good image quality was achieved, and visual assessment showed excellent agreement with 1.5T. Trigger artifacts occurred more frequently during 0.55T scans compared with 1.5T. Correct clinical stratification of volumetric parameters, WMA, valve pathologies, and parametric maps was possible in 94% of all cases. CMR at 0.55T provides a comprehensive assessment of function, structure, and tissue characterization comparable to that of 1.5T. The diagnostic accuracy in this clinical cohort is high. Identification of the specific trigger problems and optimization of the trigger technique could help to overcome the slight inaccuracies in the LV absolute volumetric measurements. Low-field CMR holds promise for expanding access to this valuable diagnostic tool, benefiting patients worldwide and on an individual level.
| DOI: | 10.1097/RLI.0000000000001225 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/40692436/ |
Abstract:
This study aimed to compare biventricular cine measurements and quantitative maps of myocardial T1 and T2 relaxation times at field strengths ranging from 0.55T to 1.5T in patients with a clinical indication for cardiac magnetic resonance imaging (CMR). Establishing the feasibility of low-field CMR may improve accessibility due to easier siting and lower cost of the low-field systems. Thirteen patients underwent same-day comparative CMR at 0.55T and on a commercial scanner at 1.5T. We examined all individuals with breath-held segmented bSSFP cine sequences for volumetric assessment of the left ventricle (LV), right ventricle (RV), and visual assessment of wall motion abnormalities (WMA) and valve pathologies. The quantitative T1 and T2 maps were acquired in 3 short-axis views for tissue characterization. Blinded readers scored the image quality on a 3-point Likert scale. Ten healthy volunteers were additionally examined at 0.55T to obtain reference values for the parametric maps. Functional analysis of both ventricles at low-field CMR has a reasonable correlation (r=0.94 to 0.99) with conventional 1.5T measurements. LV ejection fraction (EF) ( P =0.62) and RV measurements (all P >0.05) were highly reproducible, but LV absolute volumetric measurements were slightly lower at the low-field strength (all P <0.05). T1 and T2 relaxation times correlated strongly between field strengths (r=0.79; P <0.01 and r=0.63; P =0.02). Reference values from the volunteers were 678.6±13.5 ms (T1) and 66.5±4.1 ms (T2). Overall, good image quality was achieved, and visual assessment showed excellent agreement with 1.5T. Trigger artifacts occurred more frequently during 0.55T scans compared with 1.5T. Correct clinical stratification of volumetric parameters, WMA, valve pathologies, and parametric maps was possible in 94% of all cases. CMR at 0.55T provides a comprehensive assessment of function, structure, and tissue characterization comparable to that of 1.5T. The diagnostic accuracy in this clinical cohort is high. Identification of the specific trigger problems and optimization of the trigger technique could help to overcome the slight inaccuracies in the LV absolute volumetric measurements. Low-field CMR holds promise for expanding access to this valuable diagnostic tool, benefiting patients worldwide and on an individual level.
M. Bachl,
T. Rüttinger,
L. Siegler,
M. Schöniger,
S. Arndt,
L. A. Kapsner,
M. Uder and
M. S. May,
"CT Fingerprinting of Pulmonary Diseases: A Visual Analysis of Structured Reporting in RACOON-RECO",
RoFo : Fortschritte auf dem Gebiete der Rontgenstrahlen und der Nuklearmedizin,
Apr.
2026.
Abstract:
Pulmonary diseases can be analyzed with computed tomography (CT) with regard to their severity, distribution, and morphology. Up to now, this has mainly been done descriptively by radiologists, resulting in a high variance of results. Standardized structured reporting has the great potential to systematically record pulmonary changes and make them available for machine-readable reuse. However, it is still unclear how these structured findings can be analyzed in a simple and communicable way. The aim of this study was therefore to develop a method for comprehensible visualization and to apply it to a quality-assured data set. For the retrospective evaluation, chest CTs of patients from a total of 15 different disease groups such as infectious diseases, structural and vascular lung diseases, malignant lung diseases, and pleural diseases were included monocentrically. Structured reporting was performed in the RACOON infrastructure using the RECO (Phase 1) template. The systematically reported changes were evaluated both for relative frequency and for location, severity, distribution, and specific additional criteria. The results were summarized into a "CT fingerprint" using a self-developed visual RGB schema and systematically evaluated regarding the differences across the disease groups. A total of 359 patients were included. The relative frequency of pathological CT changes shows some overlap between the disease groups. In particular, the differentiation of subgroups is therefore limited. The use of fingerprints allows for better differentiation, as here not only the frequency but also the severity, localization, and distribution of the CT changes and additional criteria are taken into account. For instance, pneumonias can be differentiated depending on the causative pathogen: Based on the frequency distribution of pathological CT changes alone, Covid-19 pneumonia can be quickly classified as a pneumonia, particularly as a subgroup of viral pneumonia, but it is not possible to clearly differentiate it from other viral pneumonias. In contrast, visual analysis of the CT fingerprint allows for a better distinction. Overall, each of the 15 diseases shows an individual fingerprint. CT fingerprints enable the visualization and differentiation of pulmonary diseases based on typical CT patterns, making them a potential contribution to pandemic preparedness. · CT fingerprints allow for a visually clear presentation of structured reports from chest CTs.. · Pulmonary disease groups can thus be compared and categorized based on CT characteristics.. · The retrospective analysis of Covid-19 pneumonia shows that even previously unknown lung diseases could be quickly assigned to known disease groups.. · Pneumonias show pathogen-specific patterns.. · Bachl M, Rüttinger T, Siegler L etal. CT Fingerprinting of Pulmonary Diseases: A Visual Analysis of Structured Reporting in RACOON-RECO. Rofo 2026; DOI 10.1055/a-2818-5917.
| DOI: | 10.1055/a-2818-5917 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41980587/ |
Abstract:
Pulmonary diseases can be analyzed with computed tomography (CT) with regard to their severity, distribution, and morphology. Up to now, this has mainly been done descriptively by radiologists, resulting in a high variance of results. Standardized structured reporting has the great potential to systematically record pulmonary changes and make them available for machine-readable reuse. However, it is still unclear how these structured findings can be analyzed in a simple and communicable way. The aim of this study was therefore to develop a method for comprehensible visualization and to apply it to a quality-assured data set. For the retrospective evaluation, chest CTs of patients from a total of 15 different disease groups such as infectious diseases, structural and vascular lung diseases, malignant lung diseases, and pleural diseases were included monocentrically. Structured reporting was performed in the RACOON infrastructure using the RECO (Phase 1) template. The systematically reported changes were evaluated both for relative frequency and for location, severity, distribution, and specific additional criteria. The results were summarized into a "CT fingerprint" using a self-developed visual RGB schema and systematically evaluated regarding the differences across the disease groups. A total of 359 patients were included. The relative frequency of pathological CT changes shows some overlap between the disease groups. In particular, the differentiation of subgroups is therefore limited. The use of fingerprints allows for better differentiation, as here not only the frequency but also the severity, localization, and distribution of the CT changes and additional criteria are taken into account. For instance, pneumonias can be differentiated depending on the causative pathogen: Based on the frequency distribution of pathological CT changes alone, Covid-19 pneumonia can be quickly classified as a pneumonia, particularly as a subgroup of viral pneumonia, but it is not possible to clearly differentiate it from other viral pneumonias. In contrast, visual analysis of the CT fingerprint allows for a better distinction. Overall, each of the 15 diseases shows an individual fingerprint. CT fingerprints enable the visualization and differentiation of pulmonary diseases based on typical CT patterns, making them a potential contribution to pandemic preparedness. · CT fingerprints allow for a visually clear presentation of structured reports from chest CTs.. · Pulmonary disease groups can thus be compared and categorized based on CT characteristics.. · The retrospective analysis of Covid-19 pneumonia shows that even previously unknown lung diseases could be quickly assigned to known disease groups.. · Pneumonias show pathogen-specific patterns.. · Bachl M, Rüttinger T, Siegler L etal. CT Fingerprinting of Pulmonary Diseases: A Visual Analysis of Structured Reporting in RACOON-RECO. Rofo 2026; DOI 10.1055/a-2818-5917.
S. Saclier,
Y. He,
J. Klockner,
L. Noack,
J. Ramtke,
J. Chapiro,
R. Ruppel,
F. Hesse,
S. A. S. Abosabie,
J. K. Unger,
R. Schmidt,
B. Gebauer and
L. J. Savic,
"Establishing selection criteria for male New Zealand white rabbits in interventional radiology research using organ volumes and vessel diameters",
European radiology experimental,
vol. 10,
no. 1,
Apr.
2026.
Abstract:
New Zealand White (NZW) rabbits are widely used in interventional radiology research due to their suitability for human-sized treatment and imaging equipment, offering high translational potential. This study aims to define selection criteria for rabbits by correlating body weight (BW) and age with abdominal organ and vessel dimensions measured on cross-sectional imaging. Computed tomography and magnetic resonance imaging scans of 80 male NZW rabbits were analyzed using 3D Slicer to measure abdominal organ volumes and vessel diameters. Additionally, an in-house nnU-Net was built for liver volumetry and validated against manual segmentations. Imaging-based measurements were confirmed by gross anatomy in five animals. Statistics included normality testing and Pearson correlation. BW ranged from 2.0 to 4.5 kg (median [IQR]: 3.5 [2.9-3.8]) and age from 10.0 to 24.9 weeks (17.7 [15.0-21.4]); age correlated strongly with BW (p < 0.001). Organ volumes (liver, both kidneys) correlated with BW and age (all p < 0.001), respectively. Additionally, several vessel diameters (left common and internal iliac arteries, inferior vena cava, right common carotid artery) significantly correlated with BW and age, while the celiac trunk (p = 0.010), common hepatic (p = 0.011), and right renal artery (p = 0.031) correlated with BW only. The liver segmentation model achieved a Dice Similarity Coefficient of 0.91. BW and age in NZW rabbits correlate with both organ volumes and large vessels relevant to interventional procedures, supporting the use of biometric data as selection criteria to improve standardization, reduce complications, and enhance preclinical research quality. Weight- and age-based selection of NZW rabbits improves anatomical suitability for image-guided interventions, enhancing technical success and reproducibility. It may reduce complications and dropouts, avoiding false attribution of adverse events to the technique rather than biometric unsuitability. Lack of standardized selection criteria in the VX2 rabbit model increases procedural risks and impairs reproducibility in interventional radiology research. Biometric data correlate with organ and vessel dimensions, enabling estimations of anatomical suitability for image-guided procedures in interventional radiological research. This study establishes anatomical reference data providing a quantitative basis to standardize and refine future VX2 rabbit research.
| DOI: | 10.1186/s41747-026-00708-z |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/42047748/ |
Abstract:
New Zealand White (NZW) rabbits are widely used in interventional radiology research due to their suitability for human-sized treatment and imaging equipment, offering high translational potential. This study aims to define selection criteria for rabbits by correlating body weight (BW) and age with abdominal organ and vessel dimensions measured on cross-sectional imaging. Computed tomography and magnetic resonance imaging scans of 80 male NZW rabbits were analyzed using 3D Slicer to measure abdominal organ volumes and vessel diameters. Additionally, an in-house nnU-Net was built for liver volumetry and validated against manual segmentations. Imaging-based measurements were confirmed by gross anatomy in five animals. Statistics included normality testing and Pearson correlation. BW ranged from 2.0 to 4.5 kg (median [IQR]: 3.5 [2.9-3.8]) and age from 10.0 to 24.9 weeks (17.7 [15.0-21.4]); age correlated strongly with BW (p < 0.001). Organ volumes (liver, both kidneys) correlated with BW and age (all p < 0.001), respectively. Additionally, several vessel diameters (left common and internal iliac arteries, inferior vena cava, right common carotid artery) significantly correlated with BW and age, while the celiac trunk (p = 0.010), common hepatic (p = 0.011), and right renal artery (p = 0.031) correlated with BW only. The liver segmentation model achieved a Dice Similarity Coefficient of 0.91. BW and age in NZW rabbits correlate with both organ volumes and large vessels relevant to interventional procedures, supporting the use of biometric data as selection criteria to improve standardization, reduce complications, and enhance preclinical research quality. Weight- and age-based selection of NZW rabbits improves anatomical suitability for image-guided interventions, enhancing technical success and reproducibility. It may reduce complications and dropouts, avoiding false attribution of adverse events to the technique rather than biometric unsuitability. Lack of standardized selection criteria in the VX2 rabbit model increases procedural risks and impairs reproducibility in interventional radiology research. Biometric data correlate with organ and vessel dimensions, enabling estimations of anatomical suitability for image-guided procedures in interventional radiological research. This study establishes anatomical reference data providing a quantitative basis to standardize and refine future VX2 rabbit research.
B. Schreiweis,
B. Kinast,
H. Ulrich,
T. Bronsch,
A. Kock-Schoppenhauer and
B. Bergh,
"A Data-Centric Approach for Health Care and Research in a Health Knowledge Management Platform: Implementation and Requirement-Based Evaluation Study",
JMIR Medical Informatics,
vol. 14,
pp. e83608,
Mä.
2026.
Abstract:
Patientendaten liegen verteilt in vielen verschiedenen IT-Systemen vor und lassen sich kaum gemeinsam nutzen. Am MeDIC des Universitätsklinikums Schleswig-Holstein wurde deshalb eine Plattform entwickelt, die Daten aus Patientenversorgung und Forschung zusammenführt und vereinheitlicht. Die Plattform integriert mittlerweile Daten von etwa 1,5 Millionen Patienten, darunter Millionen von Nachrichten und Bilddatensätze aus verschiedenen Kliniksystemen. Eine Überprüfung anhand von 39 Anforderungen zeigte, dass 77 % davon erfüllt wurden – in der Praxis wurde die Plattform anhand zweier Szenarien evaluiert. Insgesamt zeigt die Studie, dass eine solche Plattform die Lücke zwischen Klinik und Forschung schließen kann – weitere Funktionen wie KI-gestützte Analysen sind bereits in Planung.
| DOI: | 10.2196/83608 |
| Datei: | https://doi.org/10.2196/83608 |
Abstract:
Patientendaten liegen verteilt in vielen verschiedenen IT-Systemen vor und lassen sich kaum gemeinsam nutzen. Am MeDIC des Universitätsklinikums Schleswig-Holstein wurde deshalb eine Plattform entwickelt, die Daten aus Patientenversorgung und Forschung zusammenführt und vereinheitlicht. Die Plattform integriert mittlerweile Daten von etwa 1,5 Millionen Patienten, darunter Millionen von Nachrichten und Bilddatensätze aus verschiedenen Kliniksystemen. Eine Überprüfung anhand von 39 Anforderungen zeigte, dass 77 % davon erfüllt wurden – in der Praxis wurde die Plattform anhand zweier Szenarien evaluiert. Insgesamt zeigt die Studie, dass eine solche Plattform die Lücke zwischen Klinik und Forschung schließen kann – weitere Funktionen wie KI-gestützte Analysen sind bereits in Planung.
T. C. Walter-Rittel,
B. Gorodetski,
A. Hartenstein,
J. Rogasch,
I. Schatka,
H. Amthauer,
M. Makowski,
C. A. Hamm and
T. Penzkofer,
"CT Radiomics Models Did Not Outperform Experts in Predicting [68Ga]Ga-PSMA-PET Positivity in Prostate Cancer Lymph Node Staging",
Current oncology (Toronto, Ont.),
vol. 33,
no. 3,
Mä.
2026.
Abstract:
Background: The use of [68Ga]Ga-PSMA-PET/CT for prostate cancer (PCa) staging is limited by cost and availability. This study evaluates whether radiomic features from contrast-enhanced (CE) CT can predict PSMA-positive lymph nodes (LNs) as a surrogate for metastasis. Methods: A retrospective study of 447 patients included 2537 segmented LNs (425 PET-positive, 2112 PET-negative). Two uroradiologists assessed 417 LNs on CE-CT using a four-point Likert scale. Radiomic features were extracted, selected using four algorithms, and analyzed with six model-building methods. Model performance was compared to radiologist ratings. Results: Radiomic models achieved an accuracy of 0.77-0.85, sensitivity of 0.85-0.91, and specificity of 0.74-0.85. Compared to radiologists, models had higher NPV (0.97-0.98 vs. 0.96) and sensitivity (0.85-0.91 vs. 0.76), but radiologists had superior accuracy (0.95 vs. 0.77-0.85) and specificity (0.97-0.98 vs. 0.74-0.85). In a subanalysis of LNs rated as probably benign or malignant, expert radiologists outperformed the algorithm with greater specificity and PPV (p < 0.005). A density threshold of >27 HU predicted PSMA-positive LNs with 0.79 accuracy, 0.87 sensitivity, and 0.78 specificity. Conclusions: While radiomics did not outperform expert radiologists, the single first-order parameter CT density >27 HU was predictive of PSMA-positive LNs. Clinical Relevance Statement: Radiomic models did not outperform expert uroradiologists. However, in high-volume or resource-limited settings lacking access to [68Ga]Ga-PSMA-PET/CT, they may help improve LN assessment in PCa patients with CT alone.
| DOI: | 10.3390/curroncol33030146 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41892174/ |
Abstract:
Background: The use of [68Ga]Ga-PSMA-PET/CT for prostate cancer (PCa) staging is limited by cost and availability. This study evaluates whether radiomic features from contrast-enhanced (CE) CT can predict PSMA-positive lymph nodes (LNs) as a surrogate for metastasis. Methods: A retrospective study of 447 patients included 2537 segmented LNs (425 PET-positive, 2112 PET-negative). Two uroradiologists assessed 417 LNs on CE-CT using a four-point Likert scale. Radiomic features were extracted, selected using four algorithms, and analyzed with six model-building methods. Model performance was compared to radiologist ratings. Results: Radiomic models achieved an accuracy of 0.77-0.85, sensitivity of 0.85-0.91, and specificity of 0.74-0.85. Compared to radiologists, models had higher NPV (0.97-0.98 vs. 0.96) and sensitivity (0.85-0.91 vs. 0.76), but radiologists had superior accuracy (0.95 vs. 0.77-0.85) and specificity (0.97-0.98 vs. 0.74-0.85). In a subanalysis of LNs rated as probably benign or malignant, expert radiologists outperformed the algorithm with greater specificity and PPV (p < 0.005). A density threshold of >27 HU predicted PSMA-positive LNs with 0.79 accuracy, 0.87 sensitivity, and 0.78 specificity. Conclusions: While radiomics did not outperform expert radiologists, the single first-order parameter CT density >27 HU was predictive of PSMA-positive LNs. Clinical Relevance Statement: Radiomic models did not outperform expert uroradiologists. However, in high-volume or resource-limited settings lacking access to [68Ga]Ga-PSMA-PET/CT, they may help improve LN assessment in PCa patients with CT alone.
M. Balzer,
A. Elsakka,
G. Luijten,
K. Schormann,
S. Jovic,
M. Gerigk,
L. Mazilescu,
S. Schmitz,
J. Keyl,
S. Liszio,
O. Basu,
B. M. Goraus,
S. Kasper,
J. Kleesiek,
B. Hinrichs-Puladi and
J. Egger,
"OncoVR - Virtual Reality in Oncology for Patient-centered Care: A Systematic Review and Meta-Analysis",
Current oncology reports,
vol. 28,
no. 1,
Mä.
2026.
Abstract:
PURPOSE OF REVIEW: Due to significant research efforts in the field of virtual reality (VR) and the increasing interest of the gaming industry, technical possibilities have developed rapidly. VR applications have become more extensive in recent years, with devices becoming more convenient and less complex to handle. The purpose of this review was to summarize the current state of research on the use of VR in oncology and to discuss its potential role in patient-centered care. RECENT FINDINGS: As a result of technological progress, the benefits of VR have entered the clinical field and are being used in various areas. However, since VR has not yet become part of everyday life, most patients are inexperienced in using these devices. Previous studies have shown that VR applications for oncology patients can improve well-being and reduce stress, anxiety, and nausea during therapy, surgery, and diagnostic procedures. Oncology is particularly well suited for psychological VR applications because cancer can be a potentially traumatic experience with a strong impact on mental health, which in turn can influence therapeutic success. In addition, oncological treatments are usually long-term, making them well suited for studies involving extended VR use. VR represents a promising supportive tool in oncology, particularly for improving psychological well-being and enhancing patient-centered care. Ongoing research suggests that its use may positively influence both the mental state of patients and the overall treatment experience.
| DOI: | 10.1007/s11912-026-01758-7 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41886179/ |
Abstract:
PURPOSE OF REVIEW: Due to significant research efforts in the field of virtual reality (VR) and the increasing interest of the gaming industry, technical possibilities have developed rapidly. VR applications have become more extensive in recent years, with devices becoming more convenient and less complex to handle. The purpose of this review was to summarize the current state of research on the use of VR in oncology and to discuss its potential role in patient-centered care. RECENT FINDINGS: As a result of technological progress, the benefits of VR have entered the clinical field and are being used in various areas. However, since VR has not yet become part of everyday life, most patients are inexperienced in using these devices. Previous studies have shown that VR applications for oncology patients can improve well-being and reduce stress, anxiety, and nausea during therapy, surgery, and diagnostic procedures. Oncology is particularly well suited for psychological VR applications because cancer can be a potentially traumatic experience with a strong impact on mental health, which in turn can influence therapeutic success. In addition, oncological treatments are usually long-term, making them well suited for studies involving extended VR use. VR represents a promising supportive tool in oncology, particularly for improving psychological well-being and enhancing patient-centered care. Ongoing research suggests that its use may positively influence both the mental state of patients and the overall treatment experience.
J. Kleiss,
S. Arndt,
L. Sommerfeld,
M. S. Md,
F. Putz,
T. Graetz,
L. Stepansky,
K. Türkan,
S. Mayr,
M. Uder and
M. S. May,
"Performance analysis of liver segmentation using nn-UNet TotalSegmentator: Focus on atypical livers, pathologies, and variants",
European journal of radiology,
vol. 196,
pp. 112674,
Mä.
2026.
Abstract:
This study evaluates the accuracy of the nn-UNet TotalSegmentator (TS) by Wasserthal et al. (2023) in segmenting atypical livers with pathologies and variants in CT scans. CT scans were retrospectively collected from our RIS and divided into two cohorts: a reference group (67 healthy livers) and a study group (55 scans across eleven pathology and variant subgroups). TS performed automatic segmentation for all groups. For reference, the images were then manually segmented, with corrections reviewed by two radiologists. Accuracy was assessed using Dice similarity score, Hausdorff distance (HD), mean surface distance (MSD), volume difference, and clinical ratings. Automatic segmentation underestimated liver volume by a mean of 48.11 ml (3.1%) in the reference group and overestimated it in 84% of study group cases by 79.09 ml (4%). The average Dice score was 0.980 ± 0.007 for the reference group and 0.933 ± 0.113 for the study group. Hepatomegaly achieved the highest score (0.979 ± 0.006), Polycystic liver disease (PLD) the lowest (0.656 ± 0.230). Cirrhosis with Ascites, Beavertail, and PLD had significantly lower Dice scores than the reference group. Clinical ratings were often lower than Dice scores suggested, especially in Beavertail, Cirrhosis with Ascites, Ablation defects, Metastases, and Hemihepatectomy. TS performs excellently on healthy and well on most pathological livers. Despite high Dice scores in many pathological cases, clinical ratings reveal limitations. Clinical evaluation remains essential. Inclusion of PLD and Beavertail cases in training data may reduce bias and improve performance.
| DOI: | 10.1016/j.ejrad.2026.112674 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41581492/ |
Abstract:
This study evaluates the accuracy of the nn-UNet TotalSegmentator (TS) by Wasserthal et al. (2023) in segmenting atypical livers with pathologies and variants in CT scans. CT scans were retrospectively collected from our RIS and divided into two cohorts: a reference group (67 healthy livers) and a study group (55 scans across eleven pathology and variant subgroups). TS performed automatic segmentation for all groups. For reference, the images were then manually segmented, with corrections reviewed by two radiologists. Accuracy was assessed using Dice similarity score, Hausdorff distance (HD), mean surface distance (MSD), volume difference, and clinical ratings. Automatic segmentation underestimated liver volume by a mean of 48.11 ml (3.1%) in the reference group and overestimated it in 84% of study group cases by 79.09 ml (4%). The average Dice score was 0.980 ± 0.007 for the reference group and 0.933 ± 0.113 for the study group. Hepatomegaly achieved the highest score (0.979 ± 0.006), Polycystic liver disease (PLD) the lowest (0.656 ± 0.230). Cirrhosis with Ascites, Beavertail, and PLD had significantly lower Dice scores than the reference group. Clinical ratings were often lower than Dice scores suggested, especially in Beavertail, Cirrhosis with Ascites, Ablation defects, Metastases, and Hemihepatectomy. TS performs excellently on healthy and well on most pathological livers. Despite high Dice scores in many pathological cases, clinical ratings reveal limitations. Clinical evaluation remains essential. Inclusion of PLD and Beavertail cases in training data may reduce bias and improve performance.
J. Nawabi,
S. Schulze-Weddige,
G. L. Baumgärtner,
T. Orth,
A. Dell’Orco,
A. Morotti,
F. Mazzacane,
H. Kniep,
C. H. Nolte,
U. Hanning,
M. Scheel,
J. Fiehler,
M. P. Wattjes and
T. Penzkofer,
"Reliability and predictors of automated volume quantification with neural networks in intracerebral hemorrhage",
European Journal of Radiology Artificial Intelligence,
vol. 5,
pp. 100067,
Mä.
2026.
| DOI: | 10.1016/j.ejrai.2025.100067 |
| Datei: | https://doi.org/10.1016/j.ejrai.2025.100067 |
M. Lindholz,
R. Bülow,
I. G. Schoots,
M. Kölln,
S. Nolte,
G. L. Baumgärtner,
J. Chenot,
C. O. Schmidt,
M. Burchardt,
M. Nauck,
H. Völzke,
M. O. Wielpütz,
T. Penzkofer,
N. Hosten and
C. A. Hamm,
"Clinical and Liquid Biomarkers of 20-Year Prostate Cancer Risk in Men Aged 45 to 70 Years",
JAMA network open,
vol. 9,
no. 2,
pp. e2556732,
Feb.
2026.
Abstract:
Prostate cancer incidence is projected to double by 2040, yet optimal risk stratification approaches for screening remain unclear despite recent international endorsements of organized programs that call for risk-adapted algorithms using readily available biomarkers. To identify biomarkers for risk-adapted prostate cancer screening in men without prostate cancer. This cohort study used data from the Study of Health in Pomerania (SHIP), a prospective population-based research initiative in Germany that randomly invited participants aged 20 to 79 years who underwent comprehensive examinations with structured 20-year follow-up. Data were collected from October 17, 1997, through September 14, 2021, with final analysis completed November 16, 2025. Baseline clinical and liquid biomarkers, including body mass index, waist-to-hip ratio, and glycated hemoglobin, together with prostate cancer-specific biomarkers of serum prostate-specific antigen (PSA) and magnetic resonance imaging-derived prostate volume and PSA density. The primary outcome was long-term prostate cancer incidence. Cumulative incidence functions for prostate cancer and death were treated as competing risks. Cause-specific Cox models were used to estimate risk and assess the association between baseline biomarkers and long-term incidence. Among 2651 men included in the study (median [IQR] age, 54.0 [48.0-62.0] years), 1482 (55.9%) had low baseline PSA levels (<1.00 ng/mL), with a cumulative prostate cancer incidence of 0.1% (95% CI, 0.0%-0.4%), 0.6% (95% CI, 0.3%-1.2%), and 3.3% (95% CI, 2.1%-4.8%) at 5, 10, and 20 years, respectively. In 958 men (36.1%) with intermediate PSA levels (1.00-3.00 ng/mL), prostate cancer incidence was 1.4% (95% CI, 0.7%-2.3%), 5.0% (95% CI, 3.6%-6.6%), and 11.8% (95% CI, 9.2%-14.8%) at 5, 10, and 20 years, respectively. In 211 men (8.0%) with high PSA levels (>3.00 ng/mL), prostate cancer incidence was 14.5% (95% CI, 10.1%-19.7%), 28.3% (95% CI, 22.2%-34.8%), and 34.8% (95% CI, 27.5%-42.2%) at 5, 10, and 20 years, respectively. Cumulative prostate cancer incidence differed significantly among the PSA groups (P < .001). In univariable and multivariable Cox regression, age (hazard ratio [HR], 1.04; 95% CI, 1.02-1.07), PSA (HR, 1.06; 95% CI, 1.04-1.07), and PSA density (HR, 1.41; 95% CI, 1.30-1.52) remained consistently associated with prostate cancer risk, whereas other variables showed either no association or inconsistent results across models. This cohort study found that a low baseline PSA level was associated with low long-term prostate cancer risk among men aged 45 to 70 years, supporting risk-adapted screening with extended intervals for men with low PSA levels.
| DOI: | 10.1001/jamanetworkopen.2025.56732 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41627816/ |
Abstract:
Prostate cancer incidence is projected to double by 2040, yet optimal risk stratification approaches for screening remain unclear despite recent international endorsements of organized programs that call for risk-adapted algorithms using readily available biomarkers. To identify biomarkers for risk-adapted prostate cancer screening in men without prostate cancer. This cohort study used data from the Study of Health in Pomerania (SHIP), a prospective population-based research initiative in Germany that randomly invited participants aged 20 to 79 years who underwent comprehensive examinations with structured 20-year follow-up. Data were collected from October 17, 1997, through September 14, 2021, with final analysis completed November 16, 2025. Baseline clinical and liquid biomarkers, including body mass index, waist-to-hip ratio, and glycated hemoglobin, together with prostate cancer-specific biomarkers of serum prostate-specific antigen (PSA) and magnetic resonance imaging-derived prostate volume and PSA density. The primary outcome was long-term prostate cancer incidence. Cumulative incidence functions for prostate cancer and death were treated as competing risks. Cause-specific Cox models were used to estimate risk and assess the association between baseline biomarkers and long-term incidence. Among 2651 men included in the study (median [IQR] age, 54.0 [48.0-62.0] years), 1482 (55.9%) had low baseline PSA levels (<1.00 ng/mL), with a cumulative prostate cancer incidence of 0.1% (95% CI, 0.0%-0.4%), 0.6% (95% CI, 0.3%-1.2%), and 3.3% (95% CI, 2.1%-4.8%) at 5, 10, and 20 years, respectively. In 958 men (36.1%) with intermediate PSA levels (1.00-3.00 ng/mL), prostate cancer incidence was 1.4% (95% CI, 0.7%-2.3%), 5.0% (95% CI, 3.6%-6.6%), and 11.8% (95% CI, 9.2%-14.8%) at 5, 10, and 20 years, respectively. In 211 men (8.0%) with high PSA levels (>3.00 ng/mL), prostate cancer incidence was 14.5% (95% CI, 10.1%-19.7%), 28.3% (95% CI, 22.2%-34.8%), and 34.8% (95% CI, 27.5%-42.2%) at 5, 10, and 20 years, respectively. Cumulative prostate cancer incidence differed significantly among the PSA groups (P < .001). In univariable and multivariable Cox regression, age (hazard ratio [HR], 1.04; 95% CI, 1.02-1.07), PSA (HR, 1.06; 95% CI, 1.04-1.07), and PSA density (HR, 1.41; 95% CI, 1.30-1.52) remained consistently associated with prostate cancer risk, whereas other variables showed either no association or inconsistent results across models. This cohort study found that a low baseline PSA level was associated with low long-term prostate cancer risk among men aged 45 to 70 years, supporting risk-adapted screening with extended intervals for men with low PSA levels.
M. Chaturvedi,
A. Bartz,
C. Denkinger,
C. Klett-Tammen,
M. Kretzschmar,
A. Kuhlmann,
B. Lange,
F. Marx,
R. Mikolajczyk,
I. Monsef,
H. Nguyen,
J. Suer,
N. Skoetz,
V. Jaeger and
A. Karch,
"Guidelines on reporting and assessing dynamic mathematical models of infectious diseases: a scoping review",
BMC Infect Dis,
vol. 26,
no. 1,
pp. 182,
Jan.
2026.
| DOI: | 10.1186/s12879-025-12211-8 |
M. Glock,
A. Erdekian,
M. Rueb,
F. Uhl,
R. Husemann,
J. Stoffers-Winterling,
S. Lindner,
O. Tüscher,
L. Hölzel,
K. Lieb,
K. Adorjan and
H. Wiegand,
"Utilization of mental health services during the first year of the COVID-19 pandemic - a systematic review and meta-analysis.",
Eur Psychiatry,
vol. 69,
no. 1,
pp. e10,
Jan.
2026.
| DOI: | 10.1192/j.eurpsy.2025.10119. |
T. Best,
L. Oestereich,
F. Colavita,
J. Michel,
Z. Igloi,
Z. Kis,
S. Metzger,
T. Bleicker,
B. van der Veer,
M. Schmidt,
N. Smolka,
K. Heimsch,
F. Carletti,
L. Widerspick,
L. Subissi,
P. Tober-Lau,
F. Kurth,
R. Molenkamp,
L. Sander,
B. Palyi,
L. Presser,
A. Nitsche,
S. Becker,
S. Günther,
J. Aarum,
V. Corman,
M. Koopmans and
C. Drosten,
"Accelerated qualification of diagnostic tests during an ongoing outbreak of Bundibugyo virus disease: a collaborative analytical validation study",
Lancet Infect Dis,
vol. S1473-3099(26)00465-2,
pp. Online ahead of print.,
Sep.
2026.
| DOI: | 10.1016/S1473-3099(26)00465-2 |
M. Riera-Marin,
J. Kleiss,
A. Aubanell,
A. Antolin,
J. Moreno-Vedia,
J. Rodriguez-Comas,
S. O K,
M. May,
J. Garcia-Lopez,
A. Galdran and
M. A. González Ballester,
"A MultiRater MultiOrgan Abdominal CT Dataset for Calibration Analysis and Uncertainty Modeling in Segmentation",
Scientific data,
vol. 13,
no. 1,
pp. 156,
Jan.
2026.
Abstract:
In medical imaging, deep learning (DL) models often struggle to delineate ambiguous structures such as tumors or organ boundaries, leading to uncertainty in defining precise contours. This challenge is amplified by inter-rater variability, where experts may disagree on boundary delineations, resulting in inconsistent segmentation outcomes. Addressing these issues requires robust algorithms capable of quantifying uncertainty, standardizing annotation practices, and improving calibration to ensure reliable predictions, particularly in multi-class and multi-rater scenarios. When models are miscalibrated and overconfident, their outputs can mislead clinical decision-making, potentially influencing radiologists to over- or under-estimate malignancy risks. The CURVAS challenge (Calibration and Uncertainty for multiRater Volume Assessment in multiorgan Segmentation) was established to address these challenges by jointly assessing uncertainty, calibration, and segmentation quality, as well as promoting clinical relevance by evaluating organ volumes while accounting for annotation variability. To support this, a dataset of 90 contrast-enhanced CT scans from University Hospital Erlangen was curated, containing pancreas, liver, and kidney segmentations annotated by three experts. This resource provides a foundation for developing and benchmarking algorithms that balance segmentation accuracy, calibration, and reliability. A quantitative analysis of the annotations shows that kidney and liver segmentations exhibit strong consistency, whereas the pancreas remains challenging, emphasizing the need for refined labeling protocols and improved training strategies.
| DOI: | 10.1038/s41597-025-06473-9 |
| Datei: | https://pubmed.ncbi.nlm.nih.gov/41513702/ |
Abstract:
In medical imaging, deep learning (DL) models often struggle to delineate ambiguous structures such as tumors or organ boundaries, leading to uncertainty in defining precise contours. This challenge is amplified by inter-rater variability, where experts may disagree on boundary delineations, resulting in inconsistent segmentation outcomes. Addressing these issues requires robust algorithms capable of quantifying uncertainty, standardizing annotation practices, and improving calibration to ensure reliable predictions, particularly in multi-class and multi-rater scenarios. When models are miscalibrated and overconfident, their outputs can mislead clinical decision-making, potentially influencing radiologists to over- or under-estimate malignancy risks. The CURVAS challenge (Calibration and Uncertainty for multiRater Volume Assessment in multiorgan Segmentation) was established to address these challenges by jointly assessing uncertainty, calibration, and segmentation quality, as well as promoting clinical relevance by evaluating organ volumes while accounting for annotation variability. To support this, a dataset of 90 contrast-enhanced CT scans from University Hospital Erlangen was curated, containing pancreas, liver, and kidney segmentations annotated by three experts. This resource provides a foundation for developing and benchmarking algorithms that balance segmentation accuracy, calibration, and reliability. A quantitative analysis of the annotations shows that kidney and liver segmentations exhibit strong consistency, whereas the pancreas remains challenging, emphasizing the need for refined labeling protocols and improved training strategies.
S. Grube,
S. Latus,
D. Moebius,
M. Fischer,
V. Raudonis,
A. Heinemann,
B. Ondruschka and
A. Schlaefer,
"An AI Based Mobile Robot for Autonomous Ultrasound in Legal Medicine",
IEEE Access,
vol. 14,
pp. 108851-108865,
Jul.
2026.
Abstract:
Ein Team von Wissenschaftler*innen aus dem Institut für Rechtsmedizin am Universitätsklinikum Hamburg-Eppendorf sowie der Technischen Universität Hamburg hat ein KI-basiertes mobiles Robotersystem entwickelt, das Ultraschalluntersuchungen von Verstorbenen vollständig autonom durchführen kann. Das System lokalisiert den Körper, identifiziert relevante anatomische Zielstrukturen (z.B. Organe, Gefäße), plant selbstständig den Scanpfad und erstellt reproduzierbare Ultraschallaufnahmen. Die Ergebnisse aus Phantomversuchen und ersten Anwendungen in der postmortalen Praxis zeigen die Machbarkeit des Ansatzes. Neben der Unterstützung bei der Verletzungsdokumentation eröffnet die Technologie auch Perspektiven für den Einsatz in der Telemedizin sowie in infektiösen Umgebungen, in denen eine kontaktarme Diagnostik von Vorteil ist.
| DOI: | 10.1109/ACCESS.2026.3713969 |
Abstract:
Ein Team von Wissenschaftler*innen aus dem Institut für Rechtsmedizin am Universitätsklinikum Hamburg-Eppendorf sowie der Technischen Universität Hamburg hat ein KI-basiertes mobiles Robotersystem entwickelt, das Ultraschalluntersuchungen von Verstorbenen vollständig autonom durchführen kann. Das System lokalisiert den Körper, identifiziert relevante anatomische Zielstrukturen (z.B. Organe, Gefäße), plant selbstständig den Scanpfad und erstellt reproduzierbare Ultraschallaufnahmen. Die Ergebnisse aus Phantomversuchen und ersten Anwendungen in der postmortalen Praxis zeigen die Machbarkeit des Ansatzes. Neben der Unterstützung bei der Verletzungsdokumentation eröffnet die Technologie auch Perspektiven für den Einsatz in der Telemedizin sowie in infektiösen Umgebungen, in denen eine kontaktarme Diagnostik von Vorteil ist.
[en]
K. Schultebraucks,
S. Gershov,
F. Fischer,
K. Wingenfeld,
S. Schmidt,
S. Steinbrecher,
T. Zoller,
F. Steinbeis,
S. M. P"utz,
J. Deckert,
M. Scherer,
I. Br"ohl,
P. Wagner,
K. S. Appel,
M. Kohls,
S. Jiru-Hillmann,
M. Nauck,
B. Lorenz-Depiereux,
S. Blaschke,
A. Muzalyova,
C. Stellbrink,
A. Steinmetz,
M. M. Addo,
E. Dahl,
M. Zettler,
S. Hansch,
A. Dinkel,
V. Keitel,
M. J. G. T. Vehreschild,
J. J. Vehreschild,
F. Paul,
M. Witzenrath,
M. Rose and
C. Otte,
"An externally validated machine learning algorithm for
predicting mental and physical health outcomes three months
post-hospitalization for severe viral infection with
SARS-CoV-2",
Brain Behav. Immun. Health,
vol. 54,
no. 101267,
pp. 101267,
Jul.
2026.
Elsevier BV.
Abstract:
Many individuals hospitalized due to severe viral infections develop mental and physical sequelae, which could potentially be prevented by targeted interventions for those at risk. Our goal was to develop and externally validate an algorithm for predicting mental and physical symptoms after SARS-CoV-2 hospitalization utilizing routinely collected clinical data. Participants were included from two independent samples of the German National Pandemic Cohort Network (NAPKON): a model development sample (SUEP; N = 451; mean age: 55.6 $\pm$ 15.3; 36.2% female) and an external validation sample (HAP: N = 158; mean age: 55.1 $\pm$ 12.1; 39.9% female). Machine learning models leveraging demographic, clinical and biological variables collected at the time of admission were employed to predict Patient-Reported Outcomes Measurement Information System scores (PROMIS) across 7 domains (physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, and pain) three months after SARS-CoV-2 hospitalization. Shapley Additive exPlanation values were used to provide interpretable information about key predictive factors. Approximately 15-20% of participants reported moderate to severe impairment in at least one PROMIS domain three months after hospitalization. For the mental health composite score, the best-performing model achieved RMSE = 1.833 $\pm$ 0.341 and R2 = 0.927 $\pm$ 0.031 in SUEP and RMSE = 3.131 and R2 = 0.893 in HAP. For the physical health composite, the best-performing model achieved RMSE = 2.908 $\pm$ 0.703 and R2 = 0.824 $\pm$ 0.052 in SUEP and RMSE = 3.019 and R2 = 0.850 in HAP. Furthermore, the models achieved high predictive performance across all individual PROMIS domain scores in both samples. We provide an externally validated methodology for accurately predicting mental and physical symptomatology following hospitalization due to a severe viral infection. This approach may facilitate the development of a brief risk stratification tool at the point of hospitalization, enabling early identification of at-risk patients, improving the prediction accuracy of subsequent psychological and physical sequelae, and supporting timely preventive interventions.
Abstract:
Many individuals hospitalized due to severe viral infections develop mental and physical sequelae, which could potentially be prevented by targeted interventions for those at risk. Our goal was to develop and externally validate an algorithm for predicting mental and physical symptoms after SARS-CoV-2 hospitalization utilizing routinely collected clinical data. Participants were included from two independent samples of the German National Pandemic Cohort Network (NAPKON): a model development sample (SUEP; N = 451; mean age: 55.6 $\pm$ 15.3; 36.2% female) and an external validation sample (HAP: N = 158; mean age: 55.1 $\pm$ 12.1; 39.9% female). Machine learning models leveraging demographic, clinical and biological variables collected at the time of admission were employed to predict Patient-Reported Outcomes Measurement Information System scores (PROMIS) across 7 domains (physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, and pain) three months after SARS-CoV-2 hospitalization. Shapley Additive exPlanation values were used to provide interpretable information about key predictive factors. Approximately 15-20% of participants reported moderate to severe impairment in at least one PROMIS domain three months after hospitalization. For the mental health composite score, the best-performing model achieved RMSE = 1.833 $\pm$ 0.341 and R2 = 0.927 $\pm$ 0.031 in SUEP and RMSE = 3.131 and R2 = 0.893 in HAP. For the physical health composite, the best-performing model achieved RMSE = 2.908 $\pm$ 0.703 and R2 = 0.824 $\pm$ 0.052 in SUEP and RMSE = 3.019 and R2 = 0.850 in HAP. Furthermore, the models achieved high predictive performance across all individual PROMIS domain scores in both samples. We provide an externally validated methodology for accurately predicting mental and physical symptomatology following hospitalization due to a severe viral infection. This approach may facilitate the development of a brief risk stratification tool at the point of hospitalization, enabling early identification of at-risk patients, improving the prediction accuracy of subsequent psychological and physical sequelae, and supporting timely preventive interventions.
[en]
J. Leberzammer,
W. T. Abplanalp,
K. Grikscheit,
E. G. Solomonidis,
S. Glaser,
B. Schuhmacher,
M. Merten,
G. Jeremijev,
A. Wilken-Schmitz,
L. Korth,
D. John,
S. G"unther,
C. Kuenne,
M. Looso,
C. Valasarajan,
V. Benes,
F. Jung,
V. Niehaus,
G. Anton,
C. Stellbrink,
C. R"ommele,
S. G"opel,
S. S. Pullamsetti,
J. Vehreschild,
M. Vehreschild,
S. Ciesek,
D. M. Leistner,
A. M. Zeiher,
S. Dimmeler and
S. Cremer,
"Atherosclerosis licenses for an exceeding immune response in
COVID-19 disease by interferon priming in circulating myeloid
cells",
Cardiovasc. Res.,
vol. 122,
no. 1,
pp. 81—97,
Feb.
2026.
Oxford University Press (OUP).
Abstract:
AIMS: Patients with cardiovascular disease (CVD) have an increased risk of developing severe respiratory infections, including COVID-19. However, the underlying molecular mechanisms are not completely understood. It has been previously shown that CVD predisposes to an altered responsiveness to subsequent inflammatory triggers by an imprinted epigenetic memory in innate immune cells. Therefore, we hypothesized that patients with pre-existing atherosclerotic cardiovascular disease (ASCVD) and COVID-19 display a dysregulated inflammatory response compared to patients without ASCVD due to epigenetically altered immune cells leading to increased disease severity. METHODS AND RESULTS: Single-cell RNA sequencing revealed a dysregulated myeloid immune response with hyperinflammatory and immunosuppressive features in patients with ASCVD and moderate COVID-19. Assay for Transposase-Accessible Chromatin sequencing and in vitro experiments with isolated monocytes infected with SARS-CoV-2 showed epigenetic priming of monocytes from patients with ASCVD towards increased expression of inflammatory mediators and type I interferon signalling. In a German nationwide cohort (NAPKON), using multiplex cytokine assays, enzyme-linked immunosorbent assays, and bulk-RNA sequencing, we confirmed that patients with ASCVD display an exaggerated inflammatory response during moderate COVID-19. CONCLUSION: This study demonstrates that patients with ASCVD show a dysregulated myeloid immune response in moderate COVID-19 disease. Mechanistically, epigenetic imprinting sensitizes myeloid cells of patients with ASCVD to an exaggerated type I interferon-associated immune response.
Abstract:
AIMS: Patients with cardiovascular disease (CVD) have an increased risk of developing severe respiratory infections, including COVID-19. However, the underlying molecular mechanisms are not completely understood. It has been previously shown that CVD predisposes to an altered responsiveness to subsequent inflammatory triggers by an imprinted epigenetic memory in innate immune cells. Therefore, we hypothesized that patients with pre-existing atherosclerotic cardiovascular disease (ASCVD) and COVID-19 display a dysregulated inflammatory response compared to patients without ASCVD due to epigenetically altered immune cells leading to increased disease severity. METHODS AND RESULTS: Single-cell RNA sequencing revealed a dysregulated myeloid immune response with hyperinflammatory and immunosuppressive features in patients with ASCVD and moderate COVID-19. Assay for Transposase-Accessible Chromatin sequencing and in vitro experiments with isolated monocytes infected with SARS-CoV-2 showed epigenetic priming of monocytes from patients with ASCVD towards increased expression of inflammatory mediators and type I interferon signalling. In a German nationwide cohort (NAPKON), using multiplex cytokine assays, enzyme-linked immunosorbent assays, and bulk-RNA sequencing, we confirmed that patients with ASCVD display an exaggerated inflammatory response during moderate COVID-19. CONCLUSION: This study demonstrates that patients with ASCVD show a dysregulated myeloid immune response in moderate COVID-19 disease. Mechanistically, epigenetic imprinting sensitizes myeloid cells of patients with ASCVD to an exaggerated type I interferon-associated immune response.
K. Allgoewer,
A. Lavenia,
L. Secco,
T. Schaller,
A. Fitzek,
C. Kriebel,
A. T. Azeke,
T. Kondo,
R. Tse,
P. Chui,
S. Schmiedel and
B. Ondruschka,
"Autopsy practices for high-consequence infectious diseases: Global guidelines, alternatives, and the BSL-4 gap",
Emerging Microbes & Infections,
vol. 15,
no. 1,
pp. 2678656,
Dez.
2026.
Abstract:
Der Review befasst sich mit Obduktionen bei hochgefährlichen Infektionskrankheiten (High Consequence Infectious Diseases, HCIDs) wie Ebola, Marburg oder Lassa-Fieber und zeigt, dass Obduktionen entscheidend zum Verständnis neuer und bekannter Infektionskrankheiten beitragen können. Gerade für Erkrankungen, die durch Erreger der höchsten Risikogruppe (RG 4) verursacht werden, ist ein verbessertes Verständnis der Pathophysiologie essentiell, um bislang fehlende spezifische Therapeutika und Impfstoffe zu entwickeln. Doch gerade hier besteht weltweit eine erhebliche Lücke: Während für Obduktionen unter Biosicherheitsstufe 3 detaillierte Empfehlungen existieren, fehlen Leitlinien für Obduktionen unter BSL-4-Bedingungen. Die Publikation vergleicht bestehende internationale Vorgaben, diskutiert alternative Verfahren wie minimalinvasive Gewebeentnahmen und entwickelt erstmals einen strukturierten Vorschlag für die Durchführung von BSL-4-Obduktionen. Die Arbeit unterstreicht die Bedeutung international harmonisierter Standards und leistungsfähiger Obduktionsinfrastrukturen als Bestandteil der Pandemic Preparedness.
| DOI: | 10.1080/22221751.2026.2678656 |
| Datei: | https://doi.org/10.1080/22221751.2026.2678656 |
Abstract:
Der Review befasst sich mit Obduktionen bei hochgefährlichen Infektionskrankheiten (High Consequence Infectious Diseases, HCIDs) wie Ebola, Marburg oder Lassa-Fieber und zeigt, dass Obduktionen entscheidend zum Verständnis neuer und bekannter Infektionskrankheiten beitragen können. Gerade für Erkrankungen, die durch Erreger der höchsten Risikogruppe (RG 4) verursacht werden, ist ein verbessertes Verständnis der Pathophysiologie essentiell, um bislang fehlende spezifische Therapeutika und Impfstoffe zu entwickeln. Doch gerade hier besteht weltweit eine erhebliche Lücke: Während für Obduktionen unter Biosicherheitsstufe 3 detaillierte Empfehlungen existieren, fehlen Leitlinien für Obduktionen unter BSL-4-Bedingungen. Die Publikation vergleicht bestehende internationale Vorgaben, diskutiert alternative Verfahren wie minimalinvasive Gewebeentnahmen und entwickelt erstmals einen strukturierten Vorschlag für die Durchführung von BSL-4-Obduktionen. Die Arbeit unterstreicht die Bedeutung international harmonisierter Standards und leistungsfähiger Obduktionsinfrastrukturen als Bestandteil der Pandemic Preparedness.
[en]
M. Zoch,
C. Gierschner,
J. Weidner,
M. Sedlmayr,
G. M"uller,
D. Choukair,
G. F. Hoffmann,
N. Toepfner,
R. Berner,
F. Prasser,
J. Schepers and
H. Hebestreit,
"Decentralized rare disease studies in Germany: first results and
hurdles of secondary use of patient data",
BMC Med. Inform. Decis. Mak.,
vol. 26,
no. 1,
Aug.
2026.
Springer Science and Business Media LLC.
Abstract:
Der Artikel zeigt, dass Daten aus verschiedenen deutschen Universitätskliniken gemeinsam für die Forschung an seltenen Erkrankungen genutzt werden können, ohne dass die persönlichen Patientendaten zentral zusammengeführt werden. Dafür wurden Daten zu vier seltenen Erkrankungen aus 20 Kliniken ausgewertet. Die Ergebnisse zeigen, dass dieser Ansatz grundsätzlich funktioniert und neue Erkenntnisse für die Forschung und Behandlung liefern kann. Gleichzeitig gibt es noch einige Schwierigkeiten: Die Daten werden in den Kliniken teilweise unterschiedlich oder unvollständig dokumentiert, manche Erkrankungen werden nicht korrekt erkannt und sehr kleine Patientenzahlen müssen aus Datenschutzgründen verborgen werden. Dadurch lassen sich manche Fragen nicht zuverlässig beantworten. Insgesamt kommt die Studie zu dem Schluss, dass die gemeinsame, aber dezentrale Auswertung von Patientendaten großes Potenzial für die Forschung zu seltenen Erkrankungen hat. Dafür müssen Datenqualität, technische Abläufe und Datenschutz jedoch weiter verbessert werden.
Abstract:
Der Artikel zeigt, dass Daten aus verschiedenen deutschen Universitätskliniken gemeinsam für die Forschung an seltenen Erkrankungen genutzt werden können, ohne dass die persönlichen Patientendaten zentral zusammengeführt werden. Dafür wurden Daten zu vier seltenen Erkrankungen aus 20 Kliniken ausgewertet. Die Ergebnisse zeigen, dass dieser Ansatz grundsätzlich funktioniert und neue Erkenntnisse für die Forschung und Behandlung liefern kann. Gleichzeitig gibt es noch einige Schwierigkeiten: Die Daten werden in den Kliniken teilweise unterschiedlich oder unvollständig dokumentiert, manche Erkrankungen werden nicht korrekt erkannt und sehr kleine Patientenzahlen müssen aus Datenschutzgründen verborgen werden. Dadurch lassen sich manche Fragen nicht zuverlässig beantworten. Insgesamt kommt die Studie zu dem Schluss, dass die gemeinsame, aber dezentrale Auswertung von Patientendaten großes Potenzial für die Forschung zu seltenen Erkrankungen hat. Dafür müssen Datenqualität, technische Abläufe und Datenschutz jedoch weiter verbessert werden.
[en]
A. Abolhassani,
T. M. Alawathurage,
A. Schmidt,
F. Brand,
L. L. Kilarski,
H. Altmann,
E. Dahl,
S. Frank,
S. G"opel,
F. Hanses,
J. C. Hellmuth,
C. Herr,
A. J. Kaasch,
R. Kobbe,
M. J. Konik,
I. Pink,
C. R"ommele,
J. Rupp,
C. S. Scheer,
M. A. Schneider,
C. Stellbrink,
H. C. Stubbe,
P. Tepasse,
A. Teufel,
I. Vadasz,
M. J. G. T. Vehreschild,
M. Witzenrath,
G. Anton,
I. Br"ohl,
S. Herold,
T. Illig,
S. Jiru-Hillmann,
P. Krawitz,
L. Mitrov,
A. Philipsen,
S. M. P"utz,
M. M. Noethen,
P. Nuernberg,
J. Reese,
O. Riess,
S. Schreiber,
J. Schultze,
F. Steinbeis,
J. J. Vehreschild,
C. Wildberg,
K. U. Ludwig and
E. C. Schulte,
"Genetic contribution to severe COVID-19 in adults under 60
years without major comorbidities in the German National Pandemic
Cohort Network (NAPKON)",
Hum. Genomics,
vol. 20,
no. 1,
pp. 23,
Jan.
2026.
Abstract:
While genome-wide association studies (GWAS) have linked common genetic variants to COVID-19 susceptibility and severity, rare high-impact variants may also contribute to phenotypic heterogeneity. Inborn errors of type I interferon immunity (IFN-I-IEIs), including X-linked TLR7 deficiency, account for ~ 2% of critical COVID-19 cases. In this study, we investigated rare potentially deleterious variants in IFN-I-IEI and GWAS-prioritized genes in young, severely affected COVID-19 patients from the German National Pandemic Cohort Network (NAPKON). Genome sequencing was performed on 110 hospitalized COVID-19 patients, including 82 males and 28 females, all under 60 years of age and without relevant pre-existing medical conditions. Rare potentially deleterious variants in TLR7 and 25 additional IFN-I-IEI genes, as well as 23 GWAS risk genes for COVID-19 severity, were analyzed based on allele frequency, predicted functional impact, and inheritance pattern models and subsequently classified based on the American College of Medical Genetics and Genomics (ACMG) criteria. Polygenic Risk Scores (PRS) were additionally calculated as an exploratory and case-only analysis to assess the contribution of common variant-derived genetic predisposition for severe COVID-19. Consistent with prior findings from other studies in German cohorts, no candidate variants or large deletions were identified in TLR7. However, 7 variants of uncertain significance in IFN-I-IEI genes as well as 13 candidate variants of potential deleterious effect in GWAS risk genes were present in 19 individuals (17.3%). We observed nominally significant differences in PRS distributions, with younger individuals (< 40 years) having higher PRS (p = 0.045) compared to older individuals, and carriers of rare variants having lower PRS compared to non-carriers (p = 0.037). These patterns are consistent with an age-dependent contribution of polygenic risk to severe COVID-19 and a potentially lower polygenic burden among rare-variant carriers, although confirmation in larger well-controlled cohorts will be required. The candidate variants identified in IFN-I-IEI and GWAS risk genes represent targets for further functional studies to clarify their potential contribution to disease risk. These findings highlight the need for future integrative genomic approaches to better understand the joint contribution of common and rare variants to COVID-19 severity.
Abstract:
While genome-wide association studies (GWAS) have linked common genetic variants to COVID-19 susceptibility and severity, rare high-impact variants may also contribute to phenotypic heterogeneity. Inborn errors of type I interferon immunity (IFN-I-IEIs), including X-linked TLR7 deficiency, account for ~ 2% of critical COVID-19 cases. In this study, we investigated rare potentially deleterious variants in IFN-I-IEI and GWAS-prioritized genes in young, severely affected COVID-19 patients from the German National Pandemic Cohort Network (NAPKON). Genome sequencing was performed on 110 hospitalized COVID-19 patients, including 82 males and 28 females, all under 60 years of age and without relevant pre-existing medical conditions. Rare potentially deleterious variants in TLR7 and 25 additional IFN-I-IEI genes, as well as 23 GWAS risk genes for COVID-19 severity, were analyzed based on allele frequency, predicted functional impact, and inheritance pattern models and subsequently classified based on the American College of Medical Genetics and Genomics (ACMG) criteria. Polygenic Risk Scores (PRS) were additionally calculated as an exploratory and case-only analysis to assess the contribution of common variant-derived genetic predisposition for severe COVID-19. Consistent with prior findings from other studies in German cohorts, no candidate variants or large deletions were identified in TLR7. However, 7 variants of uncertain significance in IFN-I-IEI genes as well as 13 candidate variants of potential deleterious effect in GWAS risk genes were present in 19 individuals (17.3%). We observed nominally significant differences in PRS distributions, with younger individuals (< 40 years) having higher PRS (p = 0.045) compared to older individuals, and carriers of rare variants having lower PRS compared to non-carriers (p = 0.037). These patterns are consistent with an age-dependent contribution of polygenic risk to severe COVID-19 and a potentially lower polygenic burden among rare-variant carriers, although confirmation in larger well-controlled cohorts will be required. The candidate variants identified in IFN-I-IEI and GWAS risk genes represent targets for further functional studies to clarify their potential contribution to disease risk. These findings highlight the need for future integrative genomic approaches to better understand the joint contribution of common and rare variants to COVID-19 severity.
[en]
W. Hoffmann,
N. Berg,
J. Schmitt,
P. Ihle,
G. Anton,
T. Datzmann,
B. Franke,
M. Geraedts,
S. Gloystein,
S. Hanß,
P. Heinrich,
E. Henke,
A. Hennings,
P. Heuschmann,
T. Illig,
M. Kraus,
C. Lang,
G. M"uller,
A. Neumann,
S. Ortmann,
V. Priya,
J. Reese,
U. Sax,
S. Scheithauer,
O. Schoffer,
M. Sedlmayr,
S. C. Semler,
E. Swart,
F. Wissing,
S. R. Zeißig and
R. Heyder,
"Health research requires the linking of healthcare-related data",
Gesundheitswesen,
Apr.
2026.
Georg Thieme Verlag KG.
Abstract:
Linking project data and data from routine clinical practice with healthcare-related data is essential for German healthcare research in order to answer complex questions validly and efficiently. Currently, fragmented data sources, heterogeneous legal requirements, and a lack of infrastructure prevent the optimal use and linking of these data. The Network University Medicine (NUM) is therefore developing a dedicated data infrastructure to link primary study data and routine clinical data with external healthcare-related data (e. g., data from statutory and private health insurance companies, data from cancer registries of the federal states, and data from registration offices). The position paper describes the various data worlds, including examples from epidemiological and clinical research that illustrate the added value and challenges of data linkage. In particular, it presents the new structures of the data acceptance and processing center (DAAeD) in the NUM, which is intended to enable quality-assured, data protection-compliant receipt and linkage of data. Standardized application and approval procedures as well as innovative privacy-preserving record linkage procedures are central to this. International experience, including from Scandinavia and the UK, demonstrates the benefits of such infrastructures for research and healthcare. In conclusion, we advocate rapid political and institutional implementation of the recommendations described in order to make health research in Germany competitive in international comparison and to ensure sustainable, patient-centered health care.
Abstract:
Linking project data and data from routine clinical practice with healthcare-related data is essential for German healthcare research in order to answer complex questions validly and efficiently. Currently, fragmented data sources, heterogeneous legal requirements, and a lack of infrastructure prevent the optimal use and linking of these data. The Network University Medicine (NUM) is therefore developing a dedicated data infrastructure to link primary study data and routine clinical data with external healthcare-related data (e. g., data from statutory and private health insurance companies, data from cancer registries of the federal states, and data from registration offices). The position paper describes the various data worlds, including examples from epidemiological and clinical research that illustrate the added value and challenges of data linkage. In particular, it presents the new structures of the data acceptance and processing center (DAAeD) in the NUM, which is intended to enable quality-assured, data protection-compliant receipt and linkage of data. Standardized application and approval procedures as well as innovative privacy-preserving record linkage procedures are central to this. International experience, including from Scandinavia and the UK, demonstrates the benefits of such infrastructures for research and healthcare. In conclusion, we advocate rapid political and institutional implementation of the recommendations described in order to make health research in Germany competitive in international comparison and to ensure sustainable, patient-centered health care.
[en]
T. Kraus,
H. Napierala,
A. Schrimpf,
S. Joos,
S. Blaschke,
F. Hanses,
M. J. G. T. Vehreschild,
E. Hamelmann,
J. Schmidt,
C. S. Scheer,
C. R"ommele,
S. G"opel,
M. Almahfoud,
E. Dahl,
A. Hamprecht,
A. J. Kaasch,
J. Wendel,
B. Ole Jensen,
P. Tepasse,
H. Dashti,
J. J. Vehreschild,
M. Scherer,
I. Br"ohl,
S. Raquib,
C. N"urnberger,
J. Reese,
K. S. Appel,
D. Krefting,
C. F"orster and
N. S. Group,
"Impact of COVID-19 vaccination on post-COVID syndrome across
clinical definitions and phenotypes: a prospective multicenter
cohort study",
Mayo Clin. Proc.,
Mä.
2026.
Elsevier BV.
Abstract:
OBJECTIVES: To assess the impact of COVID-19 vaccination on distinct clinical definitions and phenotypes of post-COVID syndrome (PCS) and identify risk factors for PCS despite vaccination. METHODS: Data were drawn from the German National Pandemic Cohort Network (NAPKON), including adult COVID-19 patients with known vaccination status recruited between December 2, 2020, and February 13, 2023. PCS prevalence was assessed using 3 clinical definitions: the broad WHO definition (any sequelae at 3 months), symptom clusters (Fatigue, Respiratory, Cognitive), and a symptom-based PCS score reflecting clinical severity. Multivariable logistic regression was employed to estimate the protective effect of vaccination. RESULTS: Among 756 patients, 26% were fully vaccinated ($\geq$2 doses) before infection. Vaccination was associated with a significantly reduced risk of PCS according to the WHO definition (OR 0.555, 95% CI 0.339-0.906), the PCS score (OR 0.536, 95% CI 0.335-0.856), the Respiratory Cluster (OR 0.508, 95% CI 0.295-0.875), and the Cognitive Cluster (OR 0.443, 95% CI 0.213-0.923). In contrast, no protective association was observed for the Fatigue Cluster (OR 0.917, 95% CI 0.554-1.519). The favorable association with vaccination was particularly observed in patients with mild acute symptoms, regardless of hospitalization. The protective effect of vaccination persisted at 12-month follow-up, although fatigue remained unaffected. CONCLUSION: Findings indicate that broader PCS definitions may mask clinically relevant heterogeneity and support the need for differentiated, phenotype-oriented definitions that reflect clinical presentations of PCS, including differential responses to vaccination. Such refined clinical definitions may facilitate exploration of whether these phenotypes reflect distinct underlying pathophysiological mechanisms.
Abstract:
OBJECTIVES: To assess the impact of COVID-19 vaccination on distinct clinical definitions and phenotypes of post-COVID syndrome (PCS) and identify risk factors for PCS despite vaccination. METHODS: Data were drawn from the German National Pandemic Cohort Network (NAPKON), including adult COVID-19 patients with known vaccination status recruited between December 2, 2020, and February 13, 2023. PCS prevalence was assessed using 3 clinical definitions: the broad WHO definition (any sequelae at 3 months), symptom clusters (Fatigue, Respiratory, Cognitive), and a symptom-based PCS score reflecting clinical severity. Multivariable logistic regression was employed to estimate the protective effect of vaccination. RESULTS: Among 756 patients, 26% were fully vaccinated ($\geq$2 doses) before infection. Vaccination was associated with a significantly reduced risk of PCS according to the WHO definition (OR 0.555, 95% CI 0.339-0.906), the PCS score (OR 0.536, 95% CI 0.335-0.856), the Respiratory Cluster (OR 0.508, 95% CI 0.295-0.875), and the Cognitive Cluster (OR 0.443, 95% CI 0.213-0.923). In contrast, no protective association was observed for the Fatigue Cluster (OR 0.917, 95% CI 0.554-1.519). The favorable association with vaccination was particularly observed in patients with mild acute symptoms, regardless of hospitalization. The protective effect of vaccination persisted at 12-month follow-up, although fatigue remained unaffected. CONCLUSION: Findings indicate that broader PCS definitions may mask clinically relevant heterogeneity and support the need for differentiated, phenotype-oriented definitions that reflect clinical presentations of PCS, including differential responses to vaccination. Such refined clinical definitions may facilitate exploration of whether these phenotypes reflect distinct underlying pathophysiological mechanisms.
[en]
M. L. Benesch Vidal,
A. Haack,
S. Anders-Lohner,
J. Munzinger,
L. D. Keil,
J. N. Albrecht,
L. Borger,
R. Twerenbold,
B. Ondruschka,
S. Blankenberg and
P. M. Becher,
"Implementation of post-mortem examinations within a large population-based cohort: results from the Hamburg City Health Study",
Clinical Research in Cardiology,
pp. Epub ahead of print,
Jan.
2026.
Abstract:
Die Studie untersucht die Durchführung konventioneller Obduktionen innerhalb der Hamburg City Health Study (HCHS) und deren Bedeutung für die Bestimmung von Todesursachen. Insgesamt verstarben während der Beobachtungszeit 354 Teilnehmende, von denen jedoch nur ein kleiner Teil (23 Fälle) obduziert wurde. In etwa 43 % dieser Fälle bestätigte die Autopsie die zuvor angenommene Todesursache, während in rund 17 % Abweichungen oder Unklarheiten zwischen klinischer Diagnose und Obduktionsbefund bestanden. In den übrigen Fällen war die Todesursache allein anhand äußerer Befunde nicht eindeutig feststellbar, weshalb eine Obduktion durchgeführt wurde. Die Ergebnisse zeigen, dass Obduktionen zwar organisatorisch und ethisch herausfordernd sind, aber wichtige zusätzliche Informationen zur genaueren Bestimmung von Todesursachen liefern.
| DOI: | 10.1007/s00392-025-02829-8 |
| Datei: | https://doi.org/10.1007/s00392-025-02829-8 |
Abstract:
Die Studie untersucht die Durchführung konventioneller Obduktionen innerhalb der Hamburg City Health Study (HCHS) und deren Bedeutung für die Bestimmung von Todesursachen. Insgesamt verstarben während der Beobachtungszeit 354 Teilnehmende, von denen jedoch nur ein kleiner Teil (23 Fälle) obduziert wurde. In etwa 43 % dieser Fälle bestätigte die Autopsie die zuvor angenommene Todesursache, während in rund 17 % Abweichungen oder Unklarheiten zwischen klinischer Diagnose und Obduktionsbefund bestanden. In den übrigen Fällen war die Todesursache allein anhand äußerer Befunde nicht eindeutig feststellbar, weshalb eine Obduktion durchgeführt wurde. Die Ergebnisse zeigen, dass Obduktionen zwar organisatorisch und ethisch herausfordernd sind, aber wichtige zusätzliche Informationen zur genaueren Bestimmung von Todesursachen liefern.
[en]
A. K. Kuderna,
N. K. Schneider,
R. Heyder,
G. Schmidt and
P. Boor,
"Infrastructure and policy reforms to facilitate collaboration in medical research",
Nature Reviews Nephrology,
pp. Epub ahead of print,
Jun.
2026.
Abstract:
Der Kommentar adressiert, welche infrastrukturellen und politischen Reformen notwendig sind, um den Austausch großer Datenmengen und die Zusammenarbeit vieler Standorte in der medizinischen Forschung effizienter zu gestalten. Anhand der Entwicklungen in Deutschland beschreiben die Autor*innen, wie mit dem Netzwerk Universitätsmedizin (NUM), dem Nationalen Obduktionsregister (NAREG), optimierten Ethikverfahren (one study, one vote) sowie digitalen Plattformen wichtige Voraussetzungen für eine moderne, datengetriebene Forschung geschaffen wurden. Das Ziel sollte sein, administrative Hürden abzubauen, Forschungsdaten besser nutzbar zu machen und multizentrische Studien schneller umzusetzen. Gleichzeitig wird deutlich, dass nationale Maßnahmen künftig durch internationale und europäische Regelungen ergänzt werden müssen, um medizinische Forschung in Deutschland nachhaltig zu stärken. Der Beitrag unterstreicht die Bedeutung leistungsfähiger Forschungsinfrastrukturen und harmonisierter Rahmenbedingungen für die Medizin der Zukunft.
| DOI: | 10.1038/s41581-026-01090-0 |
| Datei: | https://www.nature.com/articles/s41581-026-01090-0 |
Abstract:
Der Kommentar adressiert, welche infrastrukturellen und politischen Reformen notwendig sind, um den Austausch großer Datenmengen und die Zusammenarbeit vieler Standorte in der medizinischen Forschung effizienter zu gestalten. Anhand der Entwicklungen in Deutschland beschreiben die Autor*innen, wie mit dem Netzwerk Universitätsmedizin (NUM), dem Nationalen Obduktionsregister (NAREG), optimierten Ethikverfahren (one study, one vote) sowie digitalen Plattformen wichtige Voraussetzungen für eine moderne, datengetriebene Forschung geschaffen wurden. Das Ziel sollte sein, administrative Hürden abzubauen, Forschungsdaten besser nutzbar zu machen und multizentrische Studien schneller umzusetzen. Gleichzeitig wird deutlich, dass nationale Maßnahmen künftig durch internationale und europäische Regelungen ergänzt werden müssen, um medizinische Forschung in Deutschland nachhaltig zu stärken. Der Beitrag unterstreicht die Bedeutung leistungsfähiger Forschungsinfrastrukturen und harmonisierter Rahmenbedingungen für die Medizin der Zukunft.
[en]
B. Sailer,
S. Sari-Yavuz,
S. Biergans,
R. Verb"ucheln,
L. Achauer,
P. Rosenberger,
M. Hardt and
M. Koeppen,
"Interpretable machine learning models for predicting
perioperative myocardial injury in non-cardiac surgery",
Eur. Heart J. Digit. Health,
vol. 7,
no. 6,
pp. ztag093,
Jul.
2026.
Oxford University Press (OUP).
Abstract:
Aims: Perioperative myocardial injury (PMI) is a frequent and often asymptomatic complication after non-cardiac surgery and is associated with increased short- and long-term mortality. Conventional risk scores, such as the Revised Cardiac Risk Index (RCRI), have limited predictive accuracy and are infrequently used in clinical practice. We aimed to develop and temporally validate an interpretable machine learning model using Explainable Boosting Machines (EBMs) to predict PMI from routine pre-operative data. Methods and results: In this retrospective cohort study at a tertiary care centre in Germany, we included 9323 adult patients undergoing 9824 non-cardiac surgical procedures between 2014 and 2023 who received post-operative high-sensitivity cardiac troponin testing as part of routine care. PMI was defined as a post-operative elevation of high-sensitivity cardiac troponin above the upper reference limit. An EBM was trained on structured pre-operative data from 2014 to 2021 and evaluated in a temporally independent test cohort from 2022 to 2023, with performance compared with logistic regression, random forest, XGBoost, and a modified RCRI. Model discrimination, calibration, and Brier scores were assessed. Feature contributions were examined using internal shape functions and SHAP values. PMI occurred in 2804 procedures (28.5%). The EBM achieved the highest predictive performance (AUROC 0.730, 95% CI 0.720-0.740), outperforming all comparators. Calibration was robust across clinically relevant risk ranges. Key predictors included age, leukocyte count, renal function, potassium, and platelet count. The EBM identified high-risk patients more efficiently than the modified RCRI and ESC guideline-based strategies (Number Needed to Evaluate 3.0 vs. 3.5) and reduced troponin assays by 18.2% in the temporally independent cohort. Conclusion: An interpretable machine learning model trained on routine clinical data can accurately predict PMI and outperform existing risk scores. The EBM supports individualized risk stratification and may enhance perioperative decision-making and resource allocation within a guideline-directed testing population. Prospective and external validation is required before clinical implementation.
Abstract:
Aims: Perioperative myocardial injury (PMI) is a frequent and often asymptomatic complication after non-cardiac surgery and is associated with increased short- and long-term mortality. Conventional risk scores, such as the Revised Cardiac Risk Index (RCRI), have limited predictive accuracy and are infrequently used in clinical practice. We aimed to develop and temporally validate an interpretable machine learning model using Explainable Boosting Machines (EBMs) to predict PMI from routine pre-operative data. Methods and results: In this retrospective cohort study at a tertiary care centre in Germany, we included 9323 adult patients undergoing 9824 non-cardiac surgical procedures between 2014 and 2023 who received post-operative high-sensitivity cardiac troponin testing as part of routine care. PMI was defined as a post-operative elevation of high-sensitivity cardiac troponin above the upper reference limit. An EBM was trained on structured pre-operative data from 2014 to 2021 and evaluated in a temporally independent test cohort from 2022 to 2023, with performance compared with logistic regression, random forest, XGBoost, and a modified RCRI. Model discrimination, calibration, and Brier scores were assessed. Feature contributions were examined using internal shape functions and SHAP values. PMI occurred in 2804 procedures (28.5%). The EBM achieved the highest predictive performance (AUROC 0.730, 95% CI 0.720-0.740), outperforming all comparators. Calibration was robust across clinically relevant risk ranges. Key predictors included age, leukocyte count, renal function, potassium, and platelet count. The EBM identified high-risk patients more efficiently than the modified RCRI and ESC guideline-based strategies (Number Needed to Evaluate 3.0 vs. 3.5) and reduced troponin assays by 18.2% in the temporally independent cohort. Conclusion: An interpretable machine learning model trained on routine clinical data can accurately predict PMI and outperform existing risk scores. The EBM supports individualized risk stratification and may enhance perioperative decision-making and resource allocation within a guideline-directed testing population. Prospective and external validation is required before clinical implementation.
[en]
B. Sailer,
S. Sari-Yavuz,
S. Biergans,
R. Verb"ucheln,
L. Achauer,
P. Rosenberger,
M. Hardt and
M. Koeppen,
"Interpretable machine learning models for predicting
perioperative myocardial injury in non-cardiac surgery",
Eur. Heart J. Digit. Health,
vol. 7,
no. 6,
pp. ztag093,
Jul.
2026.
Oxford University Press (OUP).
Abstract:
Perioperativer Myokardschaden ist eine häufige, oft asymptomatische Komplikation nach nicht-kardialen Operationen und mit einer erhöhten Sterblichkeit und Krankheitslast assoziiert. In dieser retrospektiven Studie wurden pre-operative Daten von 9.323 Patientinnen und Patienten genutzt, um mögliche Risikofaktoren mittels Machine Learning zu untersuchen. Ein besonderer Fokus lag dabei auf sog. erklärbaren Modellen wie der Explainable Boosting Machine (EBM), welche einfach nachvollziehbare Entscheidungen liefern. Die EBM erreichte von allen getesteten Modellen die höchste prädiktive Performanz (AUROC 0.730, 95% CI 0.720–0.740). Zu den Risikofaktoren zählten fortgeschrittenes Alter, auffällige Werte bei den weißen Blutkörperchen, der Nierenfunktion, dem Kaliumspiegel und der Thrombozytenzahl. Dieser Ansatz könnte Ärztinnen und Ärzten helfen die Untersuchungen auf diejenigen Patienten zu konzentrieren, bei denen die Wahrscheinlichkeit von Komplikationen am höchsten ist.
Abstract:
Perioperativer Myokardschaden ist eine häufige, oft asymptomatische Komplikation nach nicht-kardialen Operationen und mit einer erhöhten Sterblichkeit und Krankheitslast assoziiert. In dieser retrospektiven Studie wurden pre-operative Daten von 9.323 Patientinnen und Patienten genutzt, um mögliche Risikofaktoren mittels Machine Learning zu untersuchen. Ein besonderer Fokus lag dabei auf sog. erklärbaren Modellen wie der Explainable Boosting Machine (EBM), welche einfach nachvollziehbare Entscheidungen liefern. Die EBM erreichte von allen getesteten Modellen die höchste prädiktive Performanz (AUROC 0.730, 95% CI 0.720–0.740). Zu den Risikofaktoren zählten fortgeschrittenes Alter, auffällige Werte bei den weißen Blutkörperchen, der Nierenfunktion, dem Kaliumspiegel und der Thrombozytenzahl. Dieser Ansatz könnte Ärztinnen und Ärzten helfen die Untersuchungen auf diejenigen Patienten zu konzentrieren, bei denen die Wahrscheinlichkeit von Komplikationen am höchsten ist.