ML model combining MR pelvimetry and peritoneal reflection predicts surgical difficulty in rectal cancer surgery

In a retrospective cohort of 283 patients, an XGBoost model using MRI-based pelvic depth, tumor-to-peritoneal reflection distance, and gender achieved AUC 0.809 (95% CI 0.757–0.862) for predicting laparoscopic TME difficulty, outperforming logistic regression (AUC 0.623).

• Retrospective single-center study of 283 patients (204 male, mean age 61) who underwent rectal MRI before laparoscopic total mesorectal excision.
• The XGBoost model achieved AUC 0.809 versus 0.623 for logistic regression; key predictors were pelvic depth, distance from tumor to peritoneal reflection, and gender.
• No external validation; unsupervised clustering defined surgical difficulty, potentially limiting generalizability.

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