Explainable AI Methods for Early Screening and Prognosis Prediction for Liver Cancer Using Electronic Health Records

Objectives: To develop and validate explainable artificial intelligence (AI) models for early risk prediction and prognosis prediction for liver cancer using routinely collected electronic health records (EHR). Hypothesis to be Tested: The explainable AI models proposed can improve the accuracy of early liver cancer detection and prognosis prediction compared to current statistical methods. Design and Subjects: This is a retrospective study using adult patients with liver cancer risk factors (e.g., chronic viral hepatitis, cirrhosis, and non-alcoholic fatty liver disease) under the public healthcare system in Hong Kong. Study Instruments: Clinical data that are routinely collected are used. Interventions: There is no intervention done for patients. Main Outcome Measures: The primary endpoint is the diagnosis of liver cancer. Data Analysis: Advanced AI models will be developed, trained, and validated to make predictions of early risk and prognosis. Cross-validation and external validation will be carried out and metrics, including accuracy, sensitivity, specificity, F1-score, receiver operating characteristic (ROC), and precision-recall curves, will be used to evaluate the performance. Expected Results: We anticipate that the AI models will demonstrate superior performance in early liver cancer detection and prognosis prediction compared to existing methods, with high accuracy and explainability.