Interpretable machine learning model to predict surgical difficulty in laparoscopic resection for rectal cancer.

Area of Science:

• Surgical Oncology
• Machine Learning in Medicine
• Rectal Cancer Treatment

Background:

• Laparoscopic total mesorectal excision (LaTME) is a standard but challenging procedure for rectal cancer.
• Predicting surgical difficulty is crucial for optimizing patient care and surgical planning.

Purpose of the Study:

• To develop and validate machine learning models for predicting the difficulty of LaTME in rectal cancer patients.
• To compare the performance of various machine learning models in predicting surgical difficulty.

Main Methods:

• Retrospective analysis of 626 rectal cancer patients undergoing LaTME.
• Feature selection using LASSO and logistic regression; SMOTE for class imbalance.
• Development and evaluation of six machine learning models (LGBM, CatBoost, XGBoost, LR, RF, MLP) using AUROC, accuracy, sensitivity, specificity, and F1 score.

Main Results:

• XGBoost model demonstrated the best performance with an AUROC of 0.855.
• Key predictors identified include tumor height, PNI, pelvic inlet/outlet dimensions, sacrococgeal distance, mesorectal fat area, and angle 5.
• Tumor height was identified as the most significant predictor in the XGBoost model.

Conclusions:

• An XGBoost model effectively predicts LaTME surgical difficulty in rectal cancer.
• This predictive tool can assist clinicians in anticipating surgical challenges and tailoring treatment strategies.