Machine learning combined with body composition predicts surgical difficulty in mid-low rectal cancer surgery

Area of Science:

• Colorectal Surgery
• Surgical Oncology
• Medical Imaging & Body Composition Analysis

Background:

• Laparoscopic Total Mesorectal Excision (LaTME) is a key procedure for rectal cancer.
• Identifying factors predicting surgical difficulty is crucial for patient outcomes.

Purpose of the Study:

• To identify body composition characteristics associated with LaTME surgical difficulty.
• To develop and validate an interpretable machine learning model for predicting LaTME difficulty.

Main Methods:

• Utilized LASSO regression to identify predictive clinical features in 387 rectal cancer patients.
• Developed and validated seven machine learning algorithms, including logistic regression (LR).
• Employed SHapley Additive exPlanations (SHAP) for feature interpretability and Kaplan-Meier analysis for survival.

Main Results:

• Visceral fat area (VFA), visceral fat ratio (VFR), subcutaneous fat area (SFA), and other factors predicted LaTME difficulty.
• The LR model showed optimal predictive performance.
• Difficult LaTME was associated with significantly reduced overall survival rates.

Conclusions:

• Body composition metrics are significant predictors of surgical difficulty in LaTME.
• Machine learning models, particularly LR, can effectively predict LaTME difficulty.
• Surgical difficulty in LaTME correlates with poorer patient prognosis.