Prediction of Heart Disease Using New Proposed CNN Model Architecture

• Published in: 2023 3rd International Conference on Electronic Engineering (ICEEM) pp. 1 - 8
• Main Authors: Mahmoud, Shaimaa, Gaber, Mohamed, Farouk, Gamal, Keshk, Arabi
• Format: Conference Proceeding
• Language: English
• Published: IEEE 07.10.2023
• Subjects: CNNs; Convolution; Convolutional neural networks; Diseases; ECG; Electrocardiography; Heart; Heart disease; Myocardium; Prediction; Predictive models
• DOI: 10.1109/ICEEM58740.2023.10319589

The classification of electrical cardiac signals is a crucial technique in this field because it is significantly dependent on the early identification of patients with heart disease. Many researchers have dedicated their efforts to addressing this issue, aiming to reduce the average mortality rate associated with heart conditions. However, the methods used thus far have not been accurate enough and require further improvement to enhance their performance. This paper introduces a new proposed model for prediction the heart disease patients using the concept of convolutional neural networks (CNNs). The prediction process is enhanced by incorporating a well-structured model comprising six convolutional layers, three MaxPooling layers for downsampling, and three fully connected layers. This design allows for increased capacity, enabling the model to handle complex patterns and make accurate predictions. For this study, a dataset of 928 ECG images was utilized, categorized into four groups: individuals classified as normal, individuals identified as abnormal, individuals with a past record of myocardial infarction, and individuals with a current infarction diagnosis. The newly introduced CNN architecture adeptly captures the key characteristics present in the collection of ECG images. When compared to other CNN models such as VGG-19, LeNet-5, and VGG-16, the novel model demonstrated superior performance. According to the experimental findings, the recently suggested CNN model, trained on the ECG image dataset, attained an impressive accuracy of 98%.