Enhancing cardiac disease prediction with explainable bidirectional LSTM

Area of Science:

• Cardiology
• Machine Learning
• Artificial Intelligence

Background:

• Cardiovascular disorders are the leading global cause of mortality.
• Early detection and classification of heart diseases are crucial for improving survival rates.
• Accurate and explainable predictive models are needed for cardiac disorder diagnosis using machine learning.

Purpose of the Study:

• To propose two novel machine learning models for cardiac disorder detection.
• To implement binary and multi-label classification models for cardiac disease identification.
• To enhance model explainability for better interpretation of ECG reports.

Main Methods:

• Stacking bidirectional long short-term memory (LSTM) with deep learning for feature extraction and classification.
• Training and validation of models on the PTB-XL dataset.
• Utilizing SHAP (SHapley Additive exPlanations) for model explainability.

Main Results:

• The proposed models demonstrated superior performance compared to state-of-the-art methods.
• Achieved high accuracy, precision, f1-score, and recall in cardiac disorder classification.
• Successfully enabled annotation of different diseases on ECG reports through explainability.

Conclusions:

• The developed bidirectional LSTM and deep learning models offer a powerful tool for accurate cardiac disorder detection.
• Explainable AI (SHAP) enhances the clinical utility of predictive models for ECG interpretation.
• This approach holds significant potential for improving cardiovascular disease diagnosis and patient care.