Study of the Few-Shot Learning for ECG Classification Based on the PTB-XL Dataset

Area of Science:

• Cardiology and Artificial Intelligence
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing heart diseases.
• Accurate R-peak detection in the QRS complex is the initial step for ECG feature isolation.
• Existing classification methods may have limitations in recognizing diverse cardiac conditions.

Purpose of the Study:

• To evaluate the applicability of Few-Shot Learning (FSL) for proximity-based ECG signal classification.
• To compare the performance of FSL with traditional softmax-based classification for heart disease detection.
• To develop and implement an effective method for R-peak labeling and QRS complex extraction.

Main Methods:

• Utilized the PTB-XL database, analyzing signals from Leads I-XII.
• Trained Deep Convolutional Neural Networks using FSL to classify 2, 5, and 20 heart disease classes.
• Implemented R-peak detection and k-mean clustering for QRS complex extraction from 12-lead ECG signals.

Main Results:

• FSL achieved higher accuracy (93.2%-89.2%) in classifying healthy/sick patients than softmax (90.5%-89.2%).
• FSL outperformed softmax (80.2%-77.9% vs. 77.1%-75.1%) in classifying five distinct disease classes.
• A robust R-peak labeling and QRS complex extraction procedure was successfully implemented.

Conclusions:

• Few-Shot Learning offers a more accurate approach for ECG-based heart disease classification.
• The proposed FSL network demonstrates enhanced diagnostic performance for various cardiac conditions.
• The developed R-peak and QRS extraction method is effective for ECG signal processing.