الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 114 |  |  |
| | | from | 114 | | |
Abstract
Electrocardiogram (ECG) is an important noninvasive diagnostic method for identification of various kinds of heart diseases. The accurate and efficient classification of ECG data is a key step in detecting cardiac diseases, thus reducing mortality rates. This work proposes a novel fully automated hybrid Deep Learning (DL) Computer Aided Diagnosis (CAD) for cardiovascular diseases early detection and diagnosis with high accuracy and low computational requirements. ECG signals are usually contaminated with several unwanted noise and artifact sources. A novel multi-stage filtering system based on Kernel Recursive Least Squares Tracker (KRLST) and Kernel Recursive Least Squares with Approximate Linear Dependency (ALDKRLS) is proposed for ECG de-noising and artifacts cancellation. Experimental results show that the combined ALDKRLS-KRLST approach is superior in terms of attenuating artifacts components, sensitivity of ECG peak detection, and keeping all essential characteristics of ECG records. This reveals the effectiveness of the proposed multi-stage filter as an effective framework for achieving highresolution ECG from noisy ECG recordings. After removing all artifact and noise sources, a novel hybrid approach based on combining optimized Deep Learning (DL) features with an effective aggregation of ECG features and Heart Rate Variability (HRV) measures using chaos theory and fragmentation analysis is proposed for ECG-based multi-class classification. The Constant-Q Non- Stationary Gabor Transform (CQ-NSGT) algorithm is investigated to transform the 1D ECG signal into 2D time-frequency representation that will be fed to a pre-trained Convolutional Neural Network (CNN) model, called AlexNet. The performance of the proposed CNN with CQ-NSGT is compared versus CNN with Continuous Wavelet Transform (CWT), revealing the effectiveness of the CQ-NSGT algorithm. The Pair Wise Feature Proximity (PWFP) algorithm is employed to select the optimal features from the AlexNet output feature vector to be concatenated with the ECG and HRV measures. The concatenated features are sent to different types of classifiers to distinguish CHF, ARR, and NSR. Results reveal that the Linear Discriminant Analysis (LDA) classifier has the highest accuracy compared to the other classifiers. The performance of the proposed system is examined using several evaluation metrics, including accuracy, sensitivity, precision, specificity, and computational time using 5-fold cross-validation. The proposed CAD is applied to real clinical ECG records, and the experimental results reveal the superiority of the proposed approach compared to other state-of-the-art techniques in terms of accuracy 98.75%,specificity 99.00%, precision 97.11%, sensitivity of 98.18%, and computational time 0.15 sec. This reveals the effectiveness of the proposed system as an effective tool for assisting cardiologists in real-time ECG multi-class classification.