Abstract:
Cardiovascular diseases (CVDs) remain the leading cause of morbidity and mortality globally,
underscoring the critical need for timely, accurate classification to improve patient outcomes.
Although electrocardiograms (ECGs) are widely employed for cardiac assessments, their manual
interpretation is time-consuming and prone to human error. Many previous studies have been
carried out to diagnosis and classify cardiovascular diseases using image processing, deep
learning, and machine learning techniques. However, the studies relied on limited, imbalanced,
lack of diverse dataset, low performance, high computational resource due to this their model poor
ability to generalizability when applied to unseen patient data. To address these gaps in this study,
we proposed a convolutional neural network (CNN) model that targets multi-class classification
problem of cardiovascular disease and trained on a publicly available ECG dataset containing five
classes: normal, myocardial infarction, covid19, abnormal heartbeat, and history of MI. The main
objective of this research is to design and develop hybrid model capable of extracting both spatial
and temporal features from ECG images. The scope is limited to classification based on ECG
images only, but other diagnostic modalities or grading of disease severity are not considered. The
proposed CNN model providing classification accuracies of testing is 98.76%. Receiver operating
characteristic (ROC) curves demonstrated near perfect discrimination, with values of AUC ≥ 0.99
across all classes, and confusion matrices confirmed minimal misclassification. Hence,
comparative analysis between the proposed model and state-of-the-art models (VGG19,
EfficientNetB0, MobileNetV3, and Squeeze Net) indicates that the proposed model outperforms
better performance. From the result of this study, we understand that the proposed CNN model
