An enhanced fully automatic ventricular heartbeat classification inspired by cyclic echo state networks / Qurat-Ul-Ain Mastoi

Qurat-Ul-Ain , Mastoi (2022) An enhanced fully automatic ventricular heartbeat classification inspired by cyclic echo state networks / Qurat-Ul-Ain Mastoi. PhD thesis, Universiti Malaya.

PDF (Thesis PhD) Download (2087Kb)

Abstract

Abnormal conduction (arrhythmia) in the lower chamber of the heart (ventricular) can cause cardiac diseases. Premature ventricular contraction (PVC) is a type of ventricular arrhythmia that is quite dangerous due to the frequent occurrence of premature beats in the ECG cycle. Accurate detection of premature ventricular contractions is not easy due to the multiform nature and interpatient variability issues in the heartbeat. The most challenging part of ECG signal analysis is to design an approach that accepts the multiform and interpatient variation in ECG signals for PVC arrhythmia feature extraction. This study develops a fully automatic model for PVC arrhythmia classification that helps to identify the accurate pattern of PVC arrhythmia. In the first part of the experiment, this research conducts extensive experiments to extract the features from ECG signals, such as inverse R-peaks, QRS, P-wave, and T-wave identification, and proposes a template matching technique to verify the abnormality from ECG signals. In the final stage, a cyclic echo state network classification model is proposed to classify abnormal and normal heartbeat conditions. To enhance the efficiency of the proposed model, a classifier is scaled according to the dimension of the proposed feature vector set and tuned accordingly. Three datasets are utilized to conduct this experiment: MIT-BIH-SVDB, AHA (for the experiment), and MIT-BIH-AR (for evaluation). This study follows the standard k-fold cross-validation technique and standard metrics to evaluate the performance of the model. Hence, it is observed that the proposed method achieved remarkable results, which are approximately 99.19% accuracy in SVEB cases and 99.24% accuracy in VEB cases.

Actions (For repository staff only : Login required)