Automated Body Composition from Computed Tomography Scans Improves Survival Prediction in Colorectal Cancer Patients

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Colorectal cancer (CRC) presents survival outcome variability despite current staging.
• Body composition analysis from CT scans offers prognostic value but faces data generation challenges.
• Existing methods lack comprehensive predictive power for CRC patient survival.

Purpose of the Study:

• To develop and validate a deep learning model for predicting colorectal cancer patient survival.
• To integrate clinical and body composition biomarkers for enhanced prognostic accuracy.
• To investigate the association between specific body composition features and mortality risk in CRC patients.

Main Methods:

• A deep learning model was developed using features extracted from routine computed tomography (CT) scans.
• The model integrated both clinical data and quantitative body composition biomarkers.
• Survival prediction performance was evaluated using time-dependent concordance-index (C-index).

Main Results:

• The best model integrating clinical and body composition features achieved a C-index of 0.7298 (p < 0.001).
• This combined model significantly outperformed models using only clinical or body composition data.
• Increased skeletal muscle area/radiodensity correlated with lower mortality risk.
• Higher radiodensities in visceral and subcutaneous adipose tissues correlated with increased mortality risk.

Conclusions:

• Combining body composition and clinical markers via deep learning significantly improves colorectal cancer survival prediction.
• Skeletal muscle characteristics and adipose tissue radiodensity are critical prognostic indicators in CRC.
• This approach offers a more accurate and data-driven method for patient stratification and risk assessment.