Preoperative CT deep learning of tumor and visceral fat predicts colorectal cancer recurrence, with added pathology boosting performance

Preoperative CT deep learning of tumor and visceral adipose tissue predicted colorectal cancer recurrence with C-index 0.734–0.772 externally. Adding postoperative pathology increased to 0.772–0.810 (2-year AUC 0.836).

• Tumor and visceral adipose tissue deep learning scores were each independently associated with recurrence-free survival after multivariable adjustment.
• The model used a 3D convolutional neural network for tumor features and attention-based multiple instance learning for whole-abdominal visceral adipose tissue.
• Training-derived risk tertiles consistently stratified patients into low, intermediate, and high risk in all cohorts (log-rank P<0.001).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.