Preoperative Prediction of Perineural Invasion and Prognosis in Gastric Cancer Based on Machine Learning through a...

Area of Science:

• Oncology
• Radiology
• Machine Learning
• Medical Informatics

Background:

• Perineural invasion (PNI) is a critical prognostic factor in gastric cancer.
• Accurate preoperative prediction of PNI is essential for treatment planning and improving patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning-based nomogram for preoperative prediction of PNI in gastric cancer.
• To assess the impact of PNI on the overall survival (OS) of gastric cancer patients.

Main Methods:

• Retrospective analysis of 162 gastric cancer patients' data.
• Extraction of radiomics features from contrast-enhanced computed tomography (CECT) scans.
• Development of a nomogram integrating clinicopathological factors and radiomics scores (radscore) using machine learning algorithms (LASSO, mRMR).
• Validation using an independent cohort of 42 patients.
• Kaplan-Meier analysis to evaluate the effect of PNI on OS.

Main Results:

• T stage, N stage, and radscore were identified as independent predictors of PNI.
• The developed nomogram demonstrated high predictive accuracy with AUC values of 0.851 (training), 0.842 (testing), and 0.813 (validation).
• Gastric cancer patients with PNI showed significantly poorer OS compared to those without PNI.

Conclusions:

• A machine learning-based radiomics-clinicopathological model offers a non-invasive and effective method for preoperative PNI prediction in gastric cancer.
• PNI is associated with a significantly worse prognosis in gastric cancer patients.