LightGBM-Based Classification of Heart Failure Phenotypes Using Morpho-Energy Features from High-Resolution ECG.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Heart failure (HF) is a significant global health issue requiring accessible diagnostic methods.
• Current classification relies on left ventricular ejection fraction (LVEF) assessed via resource-intensive imaging.
• Electrocardiogram (ECG) offers a low-cost, non-invasive alternative for cardiac assessment.

Purpose of the Study:

• To develop and validate a multi-algorithm AI framework for automated HF phenotype classification using high-resolution ECG signals.
• To assess the efficacy of ECG-derived morpho-energy features in discriminating between HFpEF, HFmrEF, and HFrEF phenotypes.
• To establish AI-driven ECG analysis as a reliable screening tool for HF in resource-limited settings.

Main Methods:

• A hybrid AI approach combined Pan-Tompkins and NeuroKit2 for ECG signal preprocessing and feature extraction.
• Extracted temporal, morphological, and energy-based features from segmented ECG beats.
• Trained and evaluated ensemble machine learning models (LightGBM, XGBoost, etc.) using a 70-15-15 split and 5-fold cross-validation on data from 303 chronic HF patients.

Main Results:

• The LightGBM model demonstrated superior performance, achieving 98.45% test accuracy, 0.9989 AUC, and 0.9804 macro F1-score.
• The AI framework effectively outperformed other ensemble models and a stacking classifier.
• ECG-derived morpho-energy features proved highly effective for HF phenotype discrimination.

Conclusions:

• AI-driven analysis of ECG morpho-energy features provides a reliable, non-invasive method for early HF phenotype discrimination.
• This approach can support clinical decision-making and enhance patient management, particularly in resource-limited environments.
• Automated ECG analysis holds significant potential for improving the accessibility and efficiency of HF diagnosis.