Document Details
|
Document Type |
: |
Thesis |
|
Document Title |
: |
ENHANCED COMPUTER-AIDED DIAGNOSIS SYSTEM FOR AUTOMATED CLASSIFICATION OF CARDIAC ECG SIGNALS نظام تشخيص آلي معزز بمساعدة الكمبيوتر لتصنيف إشارات تخطيط القلب الكهربائية |
|
Subject |
: |
Faculty of Engineering |
|
Document Language |
: |
Arabic |
|
Abstract |
: |
Classification of ECG signals in manual or traditional way is an area which could be improved by having such automated classification system for ECG signals. In this work, enhanced Computer-Aided Diagnosis (CAD) software system is introduced for automated classification of cardiac ECG signals. Total of 480 ECG signals were taken as dataset for the purpose of this study from MIT-BIH Arrhythmia Database; those dataset signals included 96 Normal ECG signals, as well as 384 Abnormal ECG signals belonging to four types of cardiac abnormalities which are Ventricular Couplet, Ventricular Tachycardia, Ventricular Bigeminy, and Ventricular Fibrillation, where each one of those types has 96 ECG signals as well. Then, re-sampling has been done for all given signals at 360 samples per second, except for VF signals, which have been re-sampled at 250 samples per second. After that, iterative feature extraction process has been applied with the help of Classification Learner App existed in MATLAB, resulted in 94 features including basic first order statistical features, transform domain features, as well as advanced first order statistical features and morphological features based on temporal and spectral analysis. Following to that, classification has been done, of course, with Classification Learner App, where 32 classifiers have been tried to reach best possible accuracy. Proposed system has been tried for Normal/Abnormal ECG signal classification firstly, followed by trying the system for mentioned Five-Class classification for given ECG signal. For Normal/Abnormal classification, Wide Neural Network Classifier has recorded best possible accuracy of 98.3% also, for Five-Class classification, the same model, which is Wide Neural Network Classifier gave best accuracy with 89.0% for Classification. Anyway, application of PCA has resulted with lower accuracy results, for Normal/Abnormal classification, best accuracy after application of PCA has been gotten from Weighted KNN classifier (number of neighbors is 1), with accuracy of 87.7%, while application of PCA has registered best accuracy of 69.2% for Five-Class Classification, which has been reached via Cubic SVM Classifier By Classification Learner App. In fact, those results were evaluated using performance assessments techniques which are, in addition to given Accuracy by the App, Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), Error Rate, and Area Under the Curve (AUC). Application of feature addition and extraction has been done in an iterative way as mentioned. Also, using 5-fold cross validation technique built in Classification Learner App, classification has been done. Results of proposed system could help in generalization of classification system to be used for other classes, cardiac abnormalities, or Arrhythmia types which are not included in proposed system. |
|
Supervisor |
: |
Prof. Yasser M. Kadah |
|
Thesis Type |
: |
Master Thesis |
|
Publishing Year |
: |
1445 AH
2023 AD |
|
Added Date |
: |
Sunday, October 29, 2023 |
|
Researchers
| شادي محمد عبيد | Obaid, Shadi Mohammed | Researcher | Master | |
|
Back To Researches Page
|