PP287 - AI BODY COMPOSITION ANALYSIS FROM ROUTINE IMAGING PREDICTS TOXICITY IN COLORECTAL CANCER PATIENTS - ENABLING RISK STRATIFICATION AND TARGETED PREHABILITATION

PP287

AI BODY COMPOSITION ANALYSIS FROM ROUTINE IMAGING PREDICTS TOXICITY IN COLORECTAL CANCER PATIENTS - ENABLING RISK STRATIFICATION AND TARGETED PREHABILITATION

M. Cook^1,*, E. B. Mark², M. Landgrebe³, A. Carus³

¹Centre for Nutrition and Intestinal Failure, Department of Gastroenterology, ²Clinical Cancer Research Centre, Aalborg University Hospital, ³Department of Oncology, Aalborg University Hospital, Aalborg, Denmark

Rationale: Reduced skeletal muscle mass (SMM), reflecting nutritional impairment related to cancer and ageing, is associated with increased risk of adverse outcomes in oncology and represents a marker of nutritional vulnerability. Despite routine availability, CT-based assessment of SMM is not used in clinical practise. This study evaluated an AI-based 3D segmentation tool to extract SMM from diagnostic imaging and assess its association with early severe toxicity, with potential application in nutritional risk stratification and prehabilitation.

Methods: This retrospective study included colorectal cancer patients undergoing FOLFOX/CAPOX treatment at Aalborg University Hospital from 2014-2024. Baseline routine diagnostic CT scans were retrieved along with demographic and treatment-related data.

A fully automated 3D segmentation tool (DAFS, Voronoi Inc. Canada) was used to quantify SMM from routine imaging.

Primary endpoint was early severe toxicity within first two cycles. Differences in imaging-derived muscle measures between patients with and without early severe toxicity were assessed using two-sample t-tests.

Results: A total of 254 patients were included (52% female, median age 65), of whom 49% experienced severe toxicity. The segmentation tool had a mean processing time of 6 min per patient and performed batch analyses.

Patients experiencing early severe toxicity had significantly lower skeletal muscle volume and estimated lean body mass compared to those without toxicity (1665 vs 1901 cm³, p=0.0001; 45.0 vs 49.3 kg, p=0.0007).

Conclusion: Automated body composition analysis from routine imaging identifies patients with reduced muscle reserves who have an increased risk of early severe toxicity. This approach may support nutritional risk stratification and targeted prehabilitation without additional patient burden.

Disclosure of Interest: None declared