الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 230 |  |  |
| | | from | 230 | | |
Abstract
Dealing with cardiovascular Diseases (CVDs) and processing the main heart signals to achieve efficient diagnosis performance is vital. Machine learning (ML) and Deep Learning (DL) approaches have been very salient in obtaining meaningful knowledge and providing accurate performance in several fields. Therefore, it is important to develop heart diagnosis systems based on machine and deep learning techniques at all stages. These systems will assist the doctors and the health care experts in the detection of the type of heart disease these patients suffer from. Also, these systems provide a second decision or a second option for the physicians to obtain a first impression of a specific patient’s case.
Electrocardiogram (ECG) is the main electrical impulse that shows a great variance when the patient is affected by heart arrhythmias or diseases. The ECG can be represented in the form of signals or images. In addition to this, it was discovered recently the coronavirus (COVID-19) does not only affect the lungs but also influences the heart of the individual. Therefore, the detection of (COVID-19) relying on ECG is seen to be an important issue. The thesis is related to developing three heart diagnosis systems that can process ECG signals and ECG image reports in diagnosing and detecting (COVID-19) and other heart arrhythmias.
The first system is developed to detect four types of ECG classes (i.e., normal, ST changes, supraventricular arrhythmia, and myocardial infarction (MI)). It is important to mention that the first system depends on digital ECG signals. The diagnosis system provides three main stages: pre-processing, feature extraction, and classification. Discrete wavelets transform (DWT), band stop, low pass, and smoothing filters are provided on the ECG signals as a de-noising step. Also, the features are extracted based on two deep-learning models. The first model is based on the combination of the continuous wavelet transform (CWT) with the Resnet50 pre-trained model. The second model is based on a proposed 1D CNN model. The last stage, classification, is based on three main classifiers: Softmax, random forest (RF), and XG-Boost classifier.
The second and the third systems are developed to diagnose (COVID-19), in addition, to three ECG classes (diseases) (i.e., MI, patients with a history of MI (PMI), and abnormalities). These two systems mainly depend on ECG image reports.
Concerning the second system, consists of three main stages: pre-processing and augmentation, feature extraction, and classification. The pre-processing and the augmentation stage in the second system are based on cropping, masking, median, and sharpening filters, while the augmentation is based on geometric transformations such as rotation, shearing, and reflection. The feature extraction stage is based on five main deep learning models namely the Vgg16, Vgg19, Resnet101, Xception, and ECGConvnet models. The proposed ECGConvnet model is based on the ensembling of the Xception model with the temporary convolutional network (TCN). Finally, the classification is based on four classifiers: Softmax, RF, multi-layer perception (MLP), and support vector machine (SVM).
The third heart diagnosis system also consists of three main stages: de-noising, augmenting, digitizing the ECG image reports, feature extraction, and classification. The de-noising and the augmentation steps are the same as the second system. Meanwhile, the digitization stage converts the ECG leads in the ECG image reports to (1D-ECG) signals. Then, the (1D-ECG) signals are converted to 2D ECG images (scalogram) based on CWT. Afterward, the scalogram images are fed to three main deep-learning models for feature extraction. These models are based on the combination of the convolutional neural networks (CNN) and recurrent neural networks (RNN) models. The first model is a combination of the Resnet50 model and LSTM layers, while the second model is a coordination of the Xception model and GRU layers, and the third model is an ensembling of the InceptionV3, TCN, and 1D CNN modules. Finally, the classification was based on the XG-Boost classifier.
Several qualitative and quantitative statistical measurements are used to present the final performance of the heart diagnosis systems.
The results emphasize the efficiency and reliability of the three proposed heart diagnosis systems. The comparison between the proposed systems with other related studies revealed that the performance of the presented diagnosis systems proved their consistency in the detection of COVID and the classification of other heart diseases based on ECG signals and reports.