A deep learning framework for multi-lead ECG arrhythmia classification with CAM interpretability and LLM-driven...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Cardiac arrhythmias are a significant cause of illness and death.
• Accurate and interpretable automated diagnosis of arrhythmias is crucial.
• Current diagnostic methods may lack efficiency and interpretability.

Purpose of the Study:

• To develop a lightweight deep learning framework for multi-lead ECG arrhythmia classification.
• To enhance classification accuracy and provide interpretable diagnostic insights.
• To generate patient-specific clinical recommendations.

Main Methods:

• Utilized a Winograd-scaled 1D MobileNet for feature extraction.
• Implemented Lead-aware Skip Weighting Residual ConvNeXt Attention with Builder Optimizer (LSW-RCABO) for enhanced classification.
• Employed Grad-CAM for visual explanations and BioGPT for clinical recommendations.

Main Results:

• Achieved superior classification performance on benchmark datasets.
• Demonstrated high classification accuracies: 99.98% (MIT-BIH), 99.98% (12-lead ECG), and 99.97% (PhysioNet 2020).
• Provided interpretable visual explanations and personalized clinical insights.

Conclusions:

• The proposed lightweight deep learning framework offers highly accurate and interpretable ECG arrhythmia classification.
• The system has the potential to significantly improve automated arrhythmia diagnosis.
• Integration of visual explanations and clinical recommendations enhances clinical utility.