Can machine learning predict resecability of a peritoneal carcinomatosis?

Area of Science:

• Oncology
• Surgical Oncology
• Artificial Intelligence in Medicine

Background:

• Approximately 20% of patients eligible for Hyperthermic Intraperitoneal Chemotherapy (HIPEC) undergo exploratory surgery ('open & close' or O&C).
• O&C procedures constitute up to 48% of surgeries, indicating a need for better patient selection.
• Predicting resectability is crucial for optimizing HIPEC candidacy and resource allocation.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting peritoneal carcinomatosis resectability.
• To provide decision-making support for surgeons evaluating patients for HIPEC.
• To reduce the rate of non-therapeutic 'open & close' procedures.

Main Methods:

• Intention-to-treat analysis of three prospective databases (2000-2015).
• Propensity score matching created comparable groups of patients.
• Evaluation of multiple classification algorithms including random forest, support vector machine, and conditional tree.

Main Results:

• A random forest model achieved the highest accuracy (approx. 98%) in predicting resectability.
• Nine non-resectability criteria related to organ involvement were identified and coded.
• The model demonstrated high performance with only two prediction errors in a test set of 92 patients.

Conclusions:

• Machine learning offers a promising approach to accurately predict peritoneal carcinomatosis resectability.
• Improved prediction can significantly decrease the incidence of 'open & close' surgeries.
• Future research should incorporate broader data (biologic, radiologic, laparoscopic) to further enhance predictive accuracy.