DeepECGNet: A Novel Architecture for Aortic Stenosis Detection using optimized Temporal Convolutional Network (TCN) with Attention Mechanism
Main Article Content
Abstract
Aortic stenosis (AS) is a major heart valve condition defined by aortic valve degeneration, which leads to serious cardiac complications. Severe AS symptoms are usually associated with poor prognoses. The diagnosis of AS at an early stage is challenging, largely because of the extended period of unrecognized illness in patients. Typically, conventional screening approaches fail to detect AS during the undiagnosed phase, which stresses the requirement for improved diagnostic tools. This research presents DeepECGNet, a neural network designed for the detection of aortic stenosis. The model uses large ECG datasets as its input signals. Preliminary preprocessing calls for the application of EMD to filter ECG signals, which enhances the quality of data for further investigation. Subsequent to this, Adaptive Filtering and the Hilbert Transform are used to confirm precise segmentation and selection of relevant cardiac beats. This detailed segmentation is important for obtaining important features from the ECG signals in later analytical steps. S-transform analysis of the processed signals produces time-frequency diagrams that emphasize the signal's frequency components as a function of time. Time-frequency representations are essential for the classification process of DeepECGNet. This innovative neural network analyses a range of ECG signal properties including gender significant for cardiac wellness. The inclusion of the Differential Evolution (DE) optimizer increases DeepECGNet potential to detect aortic stenosis. To detect aortic stenosis the DeepECGNet framework is built. A variety of performance metrics are assessed to evaluate the effectiveness of the suggested method for arranging ECG data such as accuracy and precision. The proposed method yields marked superiority over previous approaches used for aortic stenosis recognition. The framework shows that accuracy improves by 14.07%, 25.29%, and 21.12% and precision improves by 22.91%, 11.63%, and 18.90% compared to prominent approaches like DAS-SCGC-CNN, AD-AS-ML and DNN-AF-DAS. DeepECGNet appears to increase the accuracy of early aortic stenosis detection and may assist in enhancing patient outcomes through rapid interventions.