Paper Info
Title
A Deep Neural Network and Reconstructed Phase Space Approach to Classifying 12-lead ECGs
Abstract
The aim of this work is to classify 12-lead ECGs into 26 classes, including normal sinus rhythm, atrial fibrillation, left bundle branch block, and ST-segment depression and elevation. This work is team Marquette's submission to the PhysioNet/Computing in Cardiology Challenge 2020. Our approach is to apply two modelling techniques: a reconstructed phase space - Gaussian mixture model (RPS-GMM) method and a one-dimensional convolutional neural network. The one-dimensional convolutional neural network consists of 11 layers consisting of both convolutional and fully connected layers. It takes inputs of varying lengths to output a single diagnosis and is trained from scratch within the competition time limits. Our second method, the RPS-GMM approach, embeds each ECG lead into an l11-dimensional space and classifies using a maximum likelihood classifier. While we propose and discuss two methods only the deep convolutional neural network was used in our submissions. The RPS-GMM approach was not scored as it exceeded the competition training time limit. We achieved a score of 0.492 on the test data, but were not ranked due to omissions in the submission. Next steps include reducing the training time of the RPS-GMM approach and ensembling the two methods.
Year
DOI
Venue
2020
10.22489/CinC.2020.074
2020 Computing in Cardiology
Keywords
DocType
ISSN
reconstructed phase space approach,atrial fibrillation,left bundle branch block,one-dimensional convolutional neural network,ECG,maximum likelihood classifier,deep convolutional neural network,RPS-GMM,reconstructed phase space-Gaussian mixture model,normal sinus rhythm,ST-segment depression
Conference
2325-8861
ISBN
Citations 
PageRank 
978-1-7281-1105-6
0
0.34
References 
Authors
0
2
Name
Order
Citations
PageRank
David Kaftan100.34
Richard J. Povinelli222520.40