Explainable artificial intelligence prediction-based model in laparoscopic liver surgery for segments 7 and 8: an...

Area of Science:

• Hepatobiliary surgery
• Minimally invasive surgery
• Artificial intelligence in medicine

Background:

• Artificial intelligence (AI) is increasingly valuable for predicting surgical outcomes.
• AI models were developed to predict surgical complexity and postoperative course for laparoscopic liver surgery targeting segments 7 and 8.

Purpose of the Study:

• To develop and interpret AI models for predicting surgical complexity and postoperative outcomes in laparoscopic liver surgery.
• To identify key predictive variables using SHapley Additive exPlanations (SHAP).

Main Methods:

• Utilized an international multi-institutional database of patients undergoing minimally invasive liver surgery for segments 7 and 8.
• Employed AI models (Multi-layer Perceptron and Random Forest) and SHAP for interpretability.
• Analyzed differences between converted and non-converted surgeries.

Main Results:

• Multi-layer Perceptron (MLP) and Random Forest (RF) models demonstrated high performance in predicting complexity and outcomes, respectively.
• SHAP analysis identified "resection type" and "largest tumor size" as crucial predictors.
• Significant differences were found between converted and non-converted surgeries in tumor location, blood loss, complications, and operation time.

Conclusions:

• SHAP successfully elucidates AI model predictions for surgical complexity and outcomes in laparoscopic liver surgery.
• AI and interpretability methods offer valuable insights into predicting patient trajectories in complex liver resections.