Winograd Transform-based feature extraction for ECG and Chest X-Ray data achieved high classification accuracy (0.94) and AUC (0.98) for heart disease detection while reducing processing time.
Does Winograd Transform-based feature extraction improve processing speed and maintain high accuracy for heart disease detection using ECG and CXR images compared to traditional convolution methods?
Winograd Transform-based feature extraction provides a computationally efficient and highly accurate approach for real-time heart disease detection using ECG and CXR data in resource-constrained environments.
In resource-constrained environments, efficient feature extraction is crucial for applications in classification and prediction tasks. This study investigates a fast, DFT-based, one-dimensional Winograd Transform (WT) to extract convolution-based features from 1-D ECG signals. For two-dimensional (2-D) Chest X-Ray (CXR) images, 2-D DFT-based convolution is employed to generate features. Traditional multi-stage convolution methods for feature extraction can be slow and computationally intensive. Therefore, to improve speed and accuracy in heart disease (HD) detection, WT-based convolution methods for both 1-D and 2-D data are applied to extract features from ECG signals and CXR images. These features serve as inputs for AI-based detection models, with six machine learning (ML) and four deep learning (DL) models developed for HD detection. Using standard datasets, extensive simulations were conducted, yielding various performance metrics that were analyzed and compared. Additionally, the feature extraction and model training times were evaluated and compared. The comparative analysis demonstrates that WT-based feature extraction significantly reduces processing time for both 1-D and 2-D data types. The WT-based method achieved speedups of x times for 1-D and y times for 2-D feature extraction relative to traditional convolution methods. Performance metrics, including classification accuracy (0.94) and AUC scores (0.98), remained consistently high across models, confirming that WT-based circular convolution offers a practical and effective solution for real-time heart disease detection. This approach enhances diagnostic capabilities in healthcare by enabling resource-efficient, real-time feature extraction in medical image and signal processing.
Rath et al. (2025) studied Heart disease. Winograd Transform (WT)-based feature extraction vs. Traditional convolution methods was evaluated on Classification accuracy and AUC scores. Winograd Transform-based feature extraction for ECG and Chest X-Ray data achieved high classification accuracy (0.94) and AUC (0.98) for heart disease detection while reducing processing time.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: