Machine learning for predicting surgical difficulty of laparoscopic total mesorectal excision for rectal cancer: integrating MR-based pelvimetry and peritoneal reflection.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiology, Changhai Hospital, Naval Medical University, Shanghai, China.
- Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China.
- Department of Colorectal Surgery, Changhai Hospital, Shanghai, China.
Abstract
To investigate the potential of a machine learning (ML) model integrating MR-based pelvimetry and the peritoneal reflection (PR) in predicting the surgical difficulty of laparoscopic total mesorectal excision (LaTME) for rectal cancer (RC). RC patients who underwent rectal magnetic resonance imaging (MRI) before LaTME between December 2020 and December 2022 were included in this retrospective study. The duration of surgery, blood loss, postoperative hospital stay, and postoperative complications were selected to evaluate the surgical difficulty. Using unsupervised clustering analysis, all patients were divided into surgical difficulty group and non-surgical difficulty group based on four indicators of surgical difficulty. Least absolute shrinkage and selection operator (LASSO) regression and logistic regression were used to identify factors influencing the surgical difficulty of LaTME, and logistic regression (LR) and the extreme-gradient boosting (XGB) models were constructed. The receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were utilized to compare the performance of the two models. Data from 283 patients (60.98 ± 11.17 years old, 204 males and 79 females) were evaluated. The LASSO regression suggested that pelvic depth, the distance from the lower tumor margin to PR (T-PR), and gender were associated with the operative difficulty. The XGB model with an area under curve (AUC) of 0.809 (95% CI: 0.757-0.862) demonstrated a better performance than the LR model with an AUC of 0.623 (95%CI: 0.553-0.694). DCA confirmed that the XGB model was superior to the LR approach. The ML model integrating the pelvic depth, T-PR, and gender can evaluate the surgical difficulty of LaTME preoperatively and non-invasively, thus facilitating the predictive, preventive, and personalized treatment process in RC treatment.