Unveiling Hearts: Deep Learning-Based Electrocardiogram Classification for Congenital Heart Disease Detection

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Electrocardiograms (ECG) record the heart's electrical activity, featuring distinct waveforms like P, QRS, and T waves.
• Analysis of ECG waveform duration, morphology, and intervals aids in identifying cardiac disorders.
• Congenital heart disease (CHD) requires accurate and timely diagnosis for effective management.

Purpose of the Study:

• To develop a deep learning-based approach for accurate classification of congenital heart disease (CHD) using ECG signals.
• To leverage Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for analyzing complex ECG time-series data.
• To enhance classification performance by addressing class imbalance in ECG datasets.

Main Methods:

• Developed a deep learning model integrating CNNs and RNNs to analyze ECG features, including peak locations and intervals.
• Applied the Synthetic Minority Oversampling Technique (SMOTE) to balance the dataset and mitigate class imbalance issues.
• Utilized the MIT-BIH Arrhythmia Database for training and evaluating the CHD classification model.

Main Results:

• The deep learning model demonstrated robust performance in analyzing ECG data and identifying disease-specific patterns.
• SMOTE application significantly improved classification accuracy by effectively balancing the dataset.
• The proposed method achieved superior results compared to conventional approaches in CHD classification.

Conclusions:

• CNNs and RNNs show significant potential for accurate CHD classification from ECG signals.
• The integration of SMOTE enhances the reliability and accuracy of deep learning models for cardiac disorder detection.
• Future research should focus on external validation and addressing real-world noise in ECG data.