The Prediction of Peritoneal Carcinomatosis in Patients with Colorectal Cancer Using Machine Learning.

Area of Science:

• Oncology
• Computational Biology
• Medical Diagnostics

Background:

• Colorectal cancer (CRC) incidence is high, with late diagnosis common.
• Peritoneal carcinomatosis in CRC can be treated effectively if identified early.
• Prompt identification requires diagnostic factors for early detection.

Purpose of the Study:

• To identify factors for early diagnosis of peritoneal carcinomatosis in colorectal cancer.
• To evaluate machine learning models for detecting peritoneal carcinomatosis using routine blood tests.

Main Methods:

• Retrospective study (2010-2020) using routine blood test data.
• Applied artificial neural networks (ANN), support vector machines (SVM), and random forests (RF) with differential evolution (DE).
• DE optimized internal and structural parameters for ANN, SVM, and RF models.

Main Results:

• Random forests (RF) achieved the highest accuracy (0.75) in detecting peritoneal carcinomatosis.
• Sensitivity analysis identified key parameters influencing detection accuracy.
• Machine learning models show promise for diagnosing peritoneal carcinomatosis.

Conclusions:

• Routine blood tests combined with machine learning can aid in early peritoneal carcinomatosis detection.
• Random forests offer a robust approach for this diagnostic challenge.
• Early detection facilitates timely referral to specialized centers for improved patient outcomes.