A Hybrid Deep Learning Approach for Coronary Artery Disease Detection and Severity Analysis Using ECG Signals

Cardiovascular diseases such as Coronary Artery Disease (CAD) are life-threatening. CAD can be detected in the early stages using Electrocardiogram (ECG) signals. However, ECG disturbances have a wide range of manifestations with varying clinical significance. This paper presents a deep learning-based technique for the detection of CAD using ECG signals. The proposed method addresses common signal quality issues by employing advanced preprocessing techniques, including elliptic (Cauer) filtering for baseline wander elimination, Chebyshev Type I filtering for power line interference removal, and adaptive comb filtering for electrode motion artifacts exclusion. The preprocessed signals are then standardized using z-score normalization. During signal segmentation (one beat in each segment), the proposed model eliminates false peaks based on an adaptive threshold with morphological and statistical evaluation. The proposed model integrates Bidirectional Long Short-Term Memory (Bi-LSTM) and Neural Basis Expansion Analysis for Time Series (N-BEATS) for feature extraction, dimensionality reduction, and classification. A clustering approach is used to not only enhance the accuracy of cluster formation but also improve the overall efficiency of the K-means algorithm. By integrating the dynamic characteristics of horse herd behavior, the method adapts to varying data distributions, leading to more robust clustering results. After CAD detection, its severity is categorized using the Minnesota Code (MC) by analyzing QRS voltage, ST elevation, ST depression, and T-wave inversion patterns on different ECG leads. The proposed system was evaluated using accuracy, specificity, sensitivity, and F1-score, showing that it has the potential to aid clinicians in the detection of CAD in the initial stages using ECG signals.