Investigation of ischemic changes in electrocardiogram signal using machine learning and deep learning algorithms

Objective: Ischemic heart disease (IHD) is one of the main causes of mortality worldwide. The similarity of electrocardiographic (ECG) findings in IHD and other cardiac diseases, particularly left bundle branch block (LBBB), poses limitations for first intervention and monitoring. Therefore, in this study, we aimed to develop a model that helps clinical diagnosis by analyzing the ECG signal using artificial intelligence (AI) techniques and distinguishing between healthy, IHD, and LBBB individuals. Method: In this study, various feature extraction methods were employed to identify healthy, IHD, and LBBB characteristics from the ECG signal by using different data sets. These features were categorized as morphological, non-linear, frequency-based, raw ECG signal, and ECG-derived scalogram. Subsequently, these features were utilized as input for machine learning and deep learning algorithms to develop an AI model for diagnosis. The performance of the developed model in diagnosis was assessed using various metrics. Results: The results indicate that nonlinear and frequency-based features are more effective in diagnosing groups compared to morphological features. The support vector machines method achieved the highest success rate among the machine learning algorithms tested with these features (84%). Deep learning algorithms outperformed all machine learning algorithms in images created with raw ECG signals. The model trained with V5–V6 precordial leads achieved the highest success rate (89%). However, deep learning algorithms trained with scalograms generally achieved the highest success rate (94%). Conclusion: In the study, an AI model was developed with deep learning algorithms and scalograms obtained from the ECG signal, which has the potential to assist in diagnosing healthy people, IHD, and LBBB with a very high success rate.