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

The electrical activity of the human heart, recorded via an electrocardiogram (ECG), is characterized by distinct waveforms such as the P wave, QRS complex, and T wave. By analyzing the duration, morphology, and intervals between these waveforms, various cardiac disorders can be identified. This study aims to develop a deep learning-based approach for the accurate classification of congenital heart disease (CHD) using ECG data. We employed convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to analyze ECG signals, leveraging their ability to detect multiple features in time-series data. A deep learning model was developed and trained using features such as estimated peak locations, inter-peak intervals, and other ECG parameters. To address class imbalance, we applied the synthetic minority oversampling technique (SMOTE), which generates synthetic samples to balance each class. The analysis was conducted using the MIT-BIH Arrhythmia Database, enabling CHD classification based on ECG patterns. The proposed method improved classification accuracy by effectively balancing the dataset with SMOTE. Compared to conventional methods, the deep learning algorithms demonstrated robust performance in analyzing ECG data and detecting disease-related patterns, achieving superior results. This study highlights the potential of CNNs and RNNs for classifying CHD from ECG signals. By mitigating data imbalance with SMOTE, the approach enhances both accuracy and reliability. Future work will focus on validating the model with additional datasets and addressing real-world challenges such as noise handling and external validation.