Artificial intelligence-based detection of pre-operative surgical difficulty in laparoscopic cholecystectomy patients using gradient based model

Background: This study aimed to develop a gradient-based machine learning model to detect and assess potential challenges in laparoscopic cholecystectomy procedures. Methods: This prospective study included 146 patients diagnosed with gallstones or long-term gallbladder inflammation. Ultrasound imaging and shear wave elastography were used to evaluate various factors including gallbladder location, wall thickness, stone size, cystic duct length, common bile duct, adhesions and complications. Patients were categorized into three groups based on surgical difficulty using a manual model (MM) and a machine learning (ML) model. The ML model utilized a gradient boost algorithm and was trained, validated and tested using patient data. Laparoscopic gallbladder removal was performed and the surgeon evaluated the difficulty and complications of the procedure. Statistical analyses, including parametric measures, correlation analyses and diagnostic analyses of both models, were conducted. Results: Pericholecystic adhesions were the primary contributing factor to challenging laparoscopic cholecystectomies. The ML model achieved high accuracy (90%) for predicting preoperative surgical difficulty, with an area under the curve (AUC) of 1.0, for groups A and C and 0.89 for group B. Adhesion and cystic duct length were identified as the most significant factors in the ML model. Conclusions: The study concluded that the application of machine learning, specifically the Gradient Boosting Machine (GBM) model, enhanced the results of the manual model and demonstrated superior precision in predicting preoperative surgical difficulty.