RACOON-PAIN

The RACOON-PAIN study has the following key objectives:

1. To develop disease-specific AI algorithms for the diagnosis of five common acute conditions associated with abdominal pain: hollow organ perforation, acute pancreatitis, acute cholecystitis, colonic diverticulitis and appendicitis. To evaluate the predictive and prognostic performance of AI algorithms in large multicentre cohorts, utilising the existing infrastructure of the Radiological Cooperative Network (RACOON).

2. Development of disease-specific AI-based risk models that take into account not only demographic data (age, sex) but also laboratory parameters (C-reactive protein, leukocytes) as well as imaging features, including organ segmentations and body composition parameters – amongst which are the skeletal muscle index (indicative of sarcopenia), skeletal muscle density (indicative of myosteatosis), intramuscular adipose tissue, visceral adipose tissue (visceral obesity) and subcutaneous adipose tissue.

3. Improving the prediction of treatment-related complications and adverse clinical outcomes by integrating AI-based imaging parameters into existing clinical decision-making pathways.

4. AI-based prediction of long-term patient outcomes based on initial imaging and the data it contains – which has not yet been fully utilised – as well as the longitudinal clinical courses of the patient cohort.

The greatest challenge facing RACOON-PAIN lies in harmonising highly heterogeneous data: 37 partner sites, comprising 39 university hospitals, operate using different CT protocols, with variable image quality and site-specific documentation practices. Standardised series selection, uniform ECRFs and continuous quality monitoring with automated plausibility checks are therefore essential prerequisites for generalisable AI models.

A particular technological challenge concerns the future-proofing of the infrastructure: RACOON is set to be equipped with significantly more powerful hardware for AI-assisted image analysis. The PAIN-specific workflows, data flows and storage solutions must therefore be planned and pre-built today based on specifications that currently exist only on paper. The project addresses this risk with a consistently modular, scalable architecture that enables the subsequent transition to the new generation of hardware without interrupting running pipelines. At the same time, PAIN-specific adaptations are being implemented on the current infrastructure, particularly within the analysis environment, before the final datasets and hardware configuration are ready.
In addition, there is a parallel regulatory process: ethics and data protection approvals must be obtained at each individual site. This time-consuming process may delay the start of data collection at individual sites; however, internal project-wide balancing across sites with higher-than-planned case numbers keeps the overall timetable on track.

To ensure that AI decisions can be accepted and validated in time-critical acute care, RACOON-PAIN relies from the outset on explainable methods (including gradient-boosted decision trees with SHAP scores), whose predictions remain comprehensible to clinical practitioners.

RACOON-PAIN is a collaborative project within Thematic Area 6 of the third funding phase of the NUM, running from 1 February 2026 to 31 July 2028. Operational coordination has deliberately been assigned to two centres (UKRUB Minden/Bochum and Rostock University Medical Centre) to ensure continuity throughout the entire project period. The project is strategically managed by a Steering Committee chaired by Prof. Jan Borggrefe; the patient perspective is enshrined through a permanent seat on the Steering Committee.

Technically, RACOON-PAIN builds on the established RACOON infrastructure: CT data and clinical information are collected locally at the sites, automatically pseudonymised and transferred via the RACOON upload gateway to the central, data-protection-compliant analysis environment, RACOON-CENTRAL. Mint Medical GmbH provides the data collection, upload gateway and ECRF documentation (Mint-Lesion); the German Cancer Research Centre (DKFZ) provides the central AI analysis environment via the Joint Imaging Platform. The analysis environment is hosted on the servers of Essen University Hospital and developed by Mint Medical GmbH.

The project work is divided into six work packages: (1) Ethical and legal framework and onboarding, (2) Data integration and central data management, (3) Analysis and publication, (4) Project management, (5) AI-assisted prediction of complications, (6) AI-based prognosis of longitudinal outcomes.
Since the project’s launch in February 2026, the key milestones have been met on schedule: The primary ethics and data protection approvals for the coordinating centre have been in place since March 2026, the universal ECRF template for all sites has been finalised, and the communication and reporting structures have been established. Over 1,800 cases have already been identified; a first test case has been successfully uploaded to RACOON-CENTRAL.