ARCHIVES

In settings where computational resources are limited, the ability to extract features efficiently becomes essential for achieving reliable classification and prediction outcomes. This work focuses on employing a fast, DFT-driven version of the one-dimensional Winograd Transform to derive convolutional features from ECG signals. For chest X-ray images, which are two-dimensional in nature, a corresponding 2-D DFT-based convolution approach is applied to obtain meaningful image features. Conventional multi-stage convolution techniques often demand substantial processing time and computational power, making them less suitable for rapid diagnostic applications. To address these limitations and enhance both processing speed and diagnostic accuracy in heart disease detection, this study integrates Winograd-based convolution methods for handling both 1-D ECG data and 2-D CXR images. The extracted features are then used to train a series of artificial intelligence models, including six machine learning algorithms and four deep-learning architectures developed specifically for heart-disease classification. Extensive experiments were carried out using established datasets, and a comprehensive analysis of performance metrics was conducted to compare the models. The study also assessed the time required for feature extraction and model training, providing insights into the overall efficiency gains achieved through the proposed approach.