A Comparative Performance Analysis of Activation Functions for Cardiovascular Disease Detection Using ECG Images

ABSTRACT In recent years, artificial intelligence (AI) has become an automated tool for detecting cardiovascular diseases using ECG images. Activation functions are the core of neural network models, ranging from shallow to deep convolutional neural networks (CNN). In ECG image‐based cardiovascular disease detection, activation functions enable the network to capture non‐linear patterns like irregular heartbeats and subtle anomalies. The proposed CNN architecture in this paper comprised convolutional layers for feature extraction, followed by custom activation functions to introduce non‐linearity and enhanced learning. These features are downsampled using max pooling and aggregated through global average pooling. Fully connected layers, with a suitable dropout regularization, map the features to the final classification output, which is probabilistically determined using a softmax activation function. This paper used a public dataset of ECG images of cardiac patients to analyze the significance of activation functions in predicting the four main cardiac abnormalities: irregular heartbeat, myocardial infarction, history of myocardial infarction, and normal person classes. We have analyzed 19 different activation functions for their detection performance on the same dataset. The detection performance is compared with the existing state‐of‐the‐art studies. A set of activation functions is suggested for robust and accurate detection of cardiovascular disease using ECG images.