Explainable AI Enables Early Prediction of Intramyocardial Hemorrhage Risk in Acute Myocardial Infarction (MI) Patients

Motivation: Hemorrhagic myocardial infarction (hMI) amplifies the mortality risk in MI patients. Currently, there no methods that can identify MI patients at risk for hMI. Goal(s): To develop an explainable AI model to predict hMI. Approach: Superposable Neural Network (SNN) was trained on data from 264 MI patients. T2* MRI was used to diagnose hMI and identify its key predictors. Results: The model achieved an AUC of 0.92. A point-based scoring system, developed for predicting hMI, showed 84.9% accuracy, 82.3% sensitivity, and 87.3% specificity. The scoring system has the capacity to accurately identify hMI patients and open therapeutic opportunities to intervene to prevent/mitigate hMI. Impact: By providing an accurate and interpretable method to predict hMI risk before reperfusion, this explainable-AI-based tool empowers clinicians to make informed, real-time decisions, potentially reducing complications and improving outcomes in patients mechanically revascularized for MI.