Files in this item

FilesDescriptionFormat

application/pdf

application/pdfSP20-ECE499-Thesis-Gupta, Arjun.pdf (1MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Deep learning for cardiologist-level myocardial infarction detection in electrocadiograms
Author(s):Gupta, Arjun
Contributor(s):Zhao, Zhizhen
Subject(s):Machine Learning
Signal Processing
Cardiology
Abstract:Heart disease is the leading cause of death worldwide. Among patients with cardiovascular diseases, myocardial infarction is the main cause of death. In order to provide adequate healthcare support to patients who may experience this clinical event, it is essential to gather supportive evidence in a timely manner to help secure a correct diagnosis. In this thesis we design domain-inspired neural network models, trained, tested and validated with the Physikalisch-Technische Bundesanstalt (PTB) data set, to conduct a series of studies. First, acknowledging that the identification of suggestive electrocardiographic (ECG) changes may help in the classification of heart conditions, we adapt the ConvNetQuake neural network model---originally designed to identify earthquakes---to train, validate and test neural network models that take in from one to several ECG leads. This systematic analysis, first of its kind in the literature, indicates that out of 15 ECG leads, data from the v6, vz, and ii leads are critical to correctly identify myocardial infarction. Second, we show that using two independent train-validation-test data splits, namely, record-wise and patient-wise, does not change the finding that the combination of the leads v6, vz, and ii provides the best classification results for myocardial infarction, achieving 99.43% classification accuracy on a record-wise split, and 97.83% classification accuracy on a patient-wise split. These two results represent cardiologist-level performance for myocardial infarction detection after feeding only 10 seconds of raw ECG data into our multi-ECG-channel (v6-vz-ii) neural network model. Third, we show that our multi-ECG-channel neural network model achieves cardiologist-level performance without the need of any kind of manual feature extraction or data pre-processing.
Issue Date:2020-05
Genre:Other
Type:Other
Language:English
URI:http://hdl.handle.net/2142/107253
Date Available in IDEALS:2020-06-11


This item appears in the following Collection(s)

Item Statistics