Adaptive wavelet base selection for deep learning-based ECG diagnosis: A reinforcement learning approach.

Area of Science:

• Cardiology
• Signal Processing
• Artificial Intelligence

Background:

• Electrocardiogram (ECG) signals are vital for diagnosing cardiovascular diseases (CVDs).
• Wavelet-based feature extraction combined with deep learning (DL) shows promise for ECG diagnosis.
• Optimal wavelet base selection is a critical challenge impacting feature quality and diagnostic accuracy.

Purpose of the Study:

• To propose a reinforcement learning-based wavelet base selection (RLWBS) framework for dynamic, signal-specific wavelet base customization.
• To address the limitations of traditional, fixed wavelet bases in ECG analysis.
• To enhance the accuracy of DL-based ECG diagnosis by optimizing feature extraction.

Main Methods:

• Developed a reinforcement learning (RL) agent to iteratively optimize wavelet base selection (WBS) strategies.
• The RL agent receives feedback on classification performance to refine its WBS strategy.
• The framework dynamically customizes wavelet bases for individual ECG signals.

Main Results:

• The RLWBS framework achieved more detailed time-frequency representations of ECG signals.
• Experimental results on the PTB-XL dataset demonstrated enhanced diagnostic performance for ECG abnormality classification.
• The proposed method outperformed traditional WBS approaches.

Conclusions:

• Dynamic wavelet base selection using reinforcement learning offers a superior approach to traditional methods for ECG analysis.
• The RLWBS framework can significantly improve the accuracy of deep learning models in diagnosing cardiovascular diseases.
• This adaptive strategy enhances the extraction of relevant features from ECG signals, leading to better diagnostic outcomes.