A smart arrhythmia classification system based on wavelet transform and support vector machine techniques

• Main Author: Chia, Nyoke Goon
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: TP Chemical technology
• Online Access: http://eprints.utm.my/id/eprint/78544/1/ChiaNyokeGoonMFBME2017.pdf; http://eprints.utm.my/id/eprint/78544/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:110868

Heart disease is still the most common cause of death and contributes large number of death in this modern world. According to the survey conducted by World Health Organization (WHO), cardiovascular disease (CVD) is the number one cause of death globally in 2016. CVD claimed 801,000 lifes and heart disease killed more than 370,000 asian people which is 23.2% according to David S. Siscovick who is the chair of AHA’s Council on Epidemiology and Prevention. Arrhythmia is defined as an irregular heartbeat which will cause abnormal rhythms of the heart that further lead to serious heart disease like stroke and heart attack.Thus, arrhythmia detection and classification is crucial in clinical cardiology to analyze the extracted features from non-invasive electrocardiogram (ECG) testing. However, the arrhythmia classification accuracy based on the commercial classification software is still a remaining issue and it is an extremely time consuming process for manual visual inspection. This research proposed an arrhythmia classification software based on support vector machine (SVM) algorithm due to its advantage of higher accuracy and solve overfitting problem. The proposed system consists of three stages, namely pre-processing, feature extraction using wavelet coefficient and arrhythmia classification using SVM. All the processing stage and intermediate outputs are displayed in user friendly Graphic User Interface (GUI). The system verification is based on offline MIT-BIH database for classification accuracy, benchmarking with the other related works. The classification result shows that the proposed system is able to detect arrhythmia classification up to accuracy of 91.11%. This research output is used as PC-based ECG classification software which able to run at workstation to perform long duration of ECG data, such as 24 hour holter data. For the future work, it is suggested to add an automated R-peak detection algorithm to the system in order to solve the problem of dependency on the R-peak annotation file.