Interpretable artificial intelligence model for predicting heart failure severity after acute myocardial infarction.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Informatics

Background:

• Heart failure (HF) following acute myocardial infarction (AMI) significantly contributes to global mortality and morbidity.
• Early identification and accurate prediction of HF severity are critical for timely preventive measures and optimized treatment.
• Developing interpretable AI models for HF prediction is essential for clinical decision support.

Purpose of the Study:

• To develop an interpretable artificial intelligence (AI) model for predicting heart failure severity in patients post-acute myocardial infarction (AMI).
• To utilize multidimensional clinical data for accurate and personalized HF risk assessment.
• To enhance clinical application through an interpretable AI model and a user-friendly web platform.

Main Methods:

• A dataset of 1574 AMI patients was analyzed, incorporating medical history, clinical features, physiological parameters, lab results, coronary angiography, and echocardiography.
• Deep learning (TabNet, MLP) and machine learning (Random Forest, XGBoost) models were employed for HF severity prediction.
• The Shapley Additive Explanation (SHAP) method was used to interpret model predictions and identify key clinical factors, with a web platform developed for clinical use.

Main Results:

• The TabNet model achieved the highest performance, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.827 for four-class and 0.831 for binary KILLIP classification.
• Key clinical factors identified as highly correlated with KILLIP classification included GRACE score, NT-pro BNP, and TIMI score.
• Model interpretability was enhanced using SHAP, confirming the clinical relevance of identified factors.

Conclusions:

• The developed AI model accurately predicts HF risk and severity post-AMI using accessible multidimensional clinical data.
• The model supports early clinical intervention, personalized diagnosis, and improved patient outcomes.
• This interpretable AI approach offers significant clinical application value in managing post-AMI heart failure.