Please use this identifier to cite or link to this item: http://hdl.handle.net/10397/81940
Title: The applications of deep learning in automatic ECG classification
Authors: Xu, Shensheng
Advisors: Mak, M. W. (EIE)
Cheung, C. C. (EIE)
Keywords: Electrocardiography
Machine learning
Issue Date: 2020
Publisher: The Hong Kong Polytechnic University
Abstract: Heart arrhythmias or arrhythmias refer to the irregular heartbeats of patients. Not all arrhythmias are serious or life threatening but some types (e.g., atrial fibrillation, ventricular escape and ventricular fibrillation) may be a sign of heart diseases and could cause sudden cardiac death if prompt treatments are not received. Usually medical doctors use different types of electrocardiography (ECG) (e.g., 12-lead ECG and Holter monitors) to check for a variety of heart conditions and identify arrhythmias through analyzing the ECG. With the growing popularity of wearable technology, a large amount of ECG data are required to be analyzed. Therefore, automatic heartbeat classification from ECG signals is an essential step toward arrhythmias detection in medical practice. This thesis explores the applications of deep learning in automatic heartbeat classification, especially for the detection of occasional arrhythmias during long-term continuous cardiac monitoring. A lot of research efforts have been spent on the classification of heartbeats based on the University of California, Irvine, (UCI) cardiac arrhythmia dataset. Among them, support vector machines (SVMs) and shallow neural networks (NNs) are the most popular classification methods. However, most of the previous studies reported the performance of either the SVMs or the ANNs without in-depth comparisons between these two methods. Also, a large number of handcrafted features have been provided by the UCI dataset, and some may be more relevant to arrhythmias than the others. This thesis is to enhance the performance of heartbeat classification by selecting relevant features from ECG signals, applying dimension reduction on the feature vectors, and applying deep neural networks (DNNs) for classification. A holistic comparison among DNNs, SVMs, and shallow NNs will be provided. Experimental results based on the UCI dataset suggest that DNNs outperform both SVMs and shallow NNs, provided that relevant features have been selected.
To obtain better ECG representation for heartbeat classification, this thesis proposes deep learning methods with signal alignment that facilitate the end-to-end classification of raw ECG signals into heartbeat types, i.e., normal beat or different types of arrhythmias. Time-domain sample points are extracted from raw ECG signals, and consecutive vectors are extracted from a sliding time-window covering these sample points. Each of these vectors comprises the consecutive sample points of a complete heartbeat cycle, which includes not only the QRS complex but also the P and T waves. Unlike existing heartbeat classification methods in which medical doctors extract handcrafted features from raw ECG signals, the proposed end-to­end method leverages a DNN for both feature extraction and classification based on aligned heartbeats. This strategy not only obviates the need to handcraft the features but also produces optimized ECG representation for heartbeat classi.cation. Evaluations on the MIT-BIH arrhythmia database show that at the same specificity, the proposed patient-independent classifier can detect supraventricular-and ventricular­ectopic beats at a sensitivity that is at least 10% higher than current state-of-the-art methods. More importantly, there is a wide range of operating points in which both the sensitivity and specificity of the proposed classifier are higher than those achieved by state-of-the-art classifiers. The proposed classifier can also perform comparable to patient-specific classifiers, but at the same time enjoys the advantage of patient independency. To address the significant variability in waveforms and characteristics of ECG signals among different patients, termed as inter-patient variations, this thesis proposes adapting a patient-independent DNN using the information in the patient-dependent identity vectors (i-vectors). The adapted networks, namely i-vector adapted patient-specific DNNs (iAP-DNNs), are tuned towards the ECG characteristics of individual patients. For each patient, his/her ECG waveforms are compressed into an i-vector using a factor analysis model. Then, this i-vector is concatenated to the middle hidden layer of the patient-independent DNN. Stochastic gradient descent is then applied to fine-tune the whole network to form a patient-specific classifier. As a result, the adaptation makes use of not only the raw ECG waveforms from the specific patient but also the compact representation of his/her ECG characteristics through the i-vector. Analysis on the hidden-layer activations show that by leveraging the information in the i-vectors, the iAP-DNNs are more capable of discriminating normal heartbeats against arrhythmic heartbeats than the networks that use the patient-specific ECG only for the adaptation. Experimental results based on the MIT-BIH arrhythmia database suggest that the iAP-DNNs perform better than existing patient-specific classifiers in terms of various performance measures. In particular, the sensitivity and specificity of the existing methods are all under the receiver operating characteristic curves of iAP-DNNs.
Description: ix, 96 pages : color illustrations
PolyU Library Call No.: [THS] LG51 .H577P EIE 2020 Xu
URI: http://hdl.handle.net/10397/81940
Rights: All rights reserved.
Appears in Collections:Thesis

Files in This Item:
File Description SizeFormat 
991022347053903411_link.htmFor PolyU Users168 BHTMLView/Open
991022347053903411.pdfFor All Users4.14 MBAdobe PDFView/Open
Show full item record
PIRA download icon_1.1View/Download Contents

Page view(s)

18
Citations as of May 6, 2020

Download(s)

6
Citations as of May 6, 2020

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.