AN EFFECTIVE AND ACCURATE PREDICTION OF CARDIOVASCULAR DISEASES USING ECG SIGNAL BY IMPROVISED MULTISCALE ADAPTIVE RESIDUAL DENSENET

To automate the detection of Cardiovascular Disease (CVD) effectively, an adaptive deep-learning method is introduced in this paper. The problem formulation for the designed approach involves implementing an accurate and efficient prediction approach for CVDs utilizing Electrocardiogram (ECG) signals. The challenges in the problem include managing noisy ECG signals, extracting related features, and implementing a reliable prediction approach that can generalize well to new data. Initially, the required ECG signal is collected manually and then subjected to the artifact removal phase. The resultant signal is sent to the spectrogram image conversion. By using the Vision Transformer (ViT), the significant features from the spectrogram image are extracted. Similarly, the spectral features and wave features are extracted. The obtained features are fed into Multi-scale Adaptive Residual DenseNet (MS-ARDNet) for CVD prediction. The parameters in the MS-ARDN are optimized by Modified Random Value-based Sewing Training Optimization (MRV-STO). The novelty of the designed framework lies in its innovative combination of ViT for spectrogram image feature extraction, distinct feature extraction models, and optimization utilizing the MRV-STO algorithm. From the outcomes, the designed MS-ARDNet approach achieves 94.9% accuracy and 95% sensitivity. The designed approach’s potential applications include integration into wearable systems, remote monitoring systems, and clinical decision support systems.