Explainable Machine Learning for Predicting Early Gastric Cancer

Abstract Background: Early detection of gastric cancer is crucial for improving patient survival rates. Currently, the primary method for diagnosing early-stage gastric cancer is endoscopy, which has various limitations. Additionally, single laboratory tests continue to fall short of the requirements for early screening. This study aims to develop a machine learning (ML) model using clinical data to predict early-stage gastric cancer and apply SHapley Additive exPlanation (SHAP) values to explain the ML model.

Methods: This study involved patients who provided gastric tissue samples at Wenzhou Central Hospital from 2019 to 2023. The investigators gathered various laboratory test results from these patients. The investigators constructed and evaluated nine ML models to predict early-stage gastric cancer, using the area under the curve (AUC), accuracy, and sensitivity to assess their performance. For the most effective prediction model, The investigators utilized the SHAP method to determine the features' importance and explain the ML model.