Hybrid Deep Learning Model for Scalogram-Based ECG Classification of Cardiovascular Diseases

Cardiovascular diseases (CVDs), particularly arrhythmias, remain a leading global health challenge, underscoring the need for accurate and timely detection. Accurate and robust classification of electrocardiogram (ECG) signals for arrhythmia detection remains challenging due to noise, data imbalance, and the complexity of integrating spatial and temporal features. This study proposes a novel hybrid deep learning model for automated ECG-based heartbeat classification, achieving an outstanding classification accuracy of 99.20%. The architecture integrates Residual Network (ResNet)-50 with Squeeze-and-Excitation (SE) blocks and Long Short-Term Memory (LSTM) networks to effectively capture spatial and temporal features from ECG signals. Unlike most existing methods that treat spatial and temporal components separately, this hybrid integration represents a key novelty, enabling joint learning of hierarchical spatial features and sequential temporal dependencies. Raw ECG waveforms are transformed into two-dimensional scalograms using Continuous Wavelet Transform (CWT), enabling the extraction of rich time-frequency patterns. ResNet-50, enhanced with SE blocks, prioritizes diagnostically relevant spatial features, while LSTM models sequential dependencies critical for identifying cardiac irregularities. The model was trained and validated on a balanced dataset comprising three classes—Arrhythmia (ARR), Congestive Heart Failure (CHF), and Normal Sinus Rhythm (NSR)—using preprocessing techniques such as denoising, segmentation, and data augmentation. With precision, recall, and F1-scores exceeding 0.99, the model demonstrates robust generalization and significant potential for integration into clinical diagnostic tools, offering a scalable solution for improved cardiovascular outcomes.