Construction of an Intelligent Decision-Making Model for Laparoscopic Common Bile Duct Repair

Area of Science:

• Hepatobiliary Surgery
• Medical Informatics
• Machine Learning in Medicine

Background:

• Laparoscopic common bile duct exploration (LCBDE) for common bile duct stones (CBDS) requires optimal surgical strategy selection.
• Primary duct closure (PDC) and T-tube drainage (TTD) are common repair methods with varying outcomes.
• A data-driven approach can enhance intraoperative decision-making.

Purpose of the Study:

• To develop a machine learning decision-support model for intraoperative selection of PDC versus TTD after LCBDE.
• To identify key clinical predictors for optimal surgical strategy in CBDS patients.

Main Methods:

• Retrospective analysis of 117 patients undergoing LCBDE for CBDS.
• Feature selection using Recursive Feature Elimination (RFE).
• Training and evaluation of machine learning classifiers, including Random Forest, using metrics like AUC, accuracy, precision, and recall.

Main Results:

• RFE identified age, white blood cells, C-reactive protein, total protein, and albumin as key predictive features.
• The Random Forest model with RFE achieved an AUC of 0.83, accuracy of 0.72, precision of 0.87, and recall of 0.62.
• Baseline characteristics were similar, with higher total protein and purulent bile in the TTD group.

Conclusions:

• An intelligent decision-making model for laparoscopic bile duct repair was successfully developed.
• The model leverages RFE and Random Forest to personalize surgical recommendations for CBDS.
• This approach may improve precision treatment outcomes in patients with common bile duct stones.