ECG Image Classification Using Hybrid Machine Learning and Deep Learning Techniques

Abstract

Electrocardiogram (ECG) interpretation is one of the most important diagnostic procedures in cardiovascular medicine. It is routinely used to detect myocardial infarction, rhythm disorders, conduction abnormalities, ischemic changes, and previous cardiac injury. However, accurate interpretation of ECG reports still depends greatly on physician experience, especially when ECG records are stored as scanned documents, printed sheets, or image captures rather than raw digital waveform signals. This creates a practical need for automated image-based ECG classification systems that can support physicians in diagnosis, screening, triage, and telemedicine applications. This paper presents a hybrid framework for four-class ECG image classification using enhanced grayscale preprocessing, advanced handcrafted feature extraction, hyperparameter-optimized machine learning, and deep learning comparison. The dataset contains four diagnostic categories: myocardial infarction (MI), abnormal heartbeat (AHB), history of myocardial infarction (HMI), and normal ECG. The proposed system automatically detects the ECG chart region, crops the useful waveform area, removes background artifacts, resizes all images to a fixed resolution, balances the training data through augmentation, extracts discriminative image features, and evaluates multiple classifiers. Experimental results show that the Support Vector Machine (SVM) model achieved the highest classification accuracy of 96.76%, followed closely by the Convolutional Neural Network (CNN) model with 96.40%. These results demonstrate that optimized classical machine learning can remain highly competitive with deep learning when supported by strong preprocessing and informative feature engineering. The proposed framework can serve as a reliable foundation for intelligent ECG screening systems and computer-aided cardiac diagnosis.