A pooling convolution model for multi-classification of ECG and PCG signals.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cardiovascular diseases necessitate efficient detection methods.
• Electrocardiogram (ECG) and phonocardiogram (PCG) signals offer vital diagnostic information.
• Deep learning presents a promising avenue for analyzing complex physiological signals.

Purpose of the Study:

• To develop and evaluate novel deep learning models for multi-classification of ECG and PCG signals.
• To enhance the accuracy and efficiency of cardiovascular disease detection using signal analysis.
• To introduce straightforward yet effective pooling convolutional models for physiological signal recognition.

Main Methods:

• Preprocessing of ECG and PCG signals.
• Design of structural blocks: stacked block (MCM) with convolutional and max-pooling layers, and residual block (REC).
• Adjustment of structural block numbers to accommodate varying signal sampling rates.

Main Results:

• MCM block models achieved 98.70% accuracy on ECG and 92.58% on PCG fusion datasets.
• MCM models outperformed variations and REC block models, with accuracy improvements of 0.02% and 4.30% respectively.
• Highest accuracy was also achieved on a synchronized ECG-PCG dataset classifying fatigue levels.

Conclusions:

• The proposed MCM structural block offers superior performance for ECG and PCG signal classification.
• The developed deep learning models are effective and generalizable for cardiovascular signal analysis.
• This approach significantly enhances the potential for early and accurate cardiovascular disease detection.