Home // ICCGI 2013, The Eighth International Multi-Conference on Computing in the Global Information Technology // View article

Finding an Optimal Model for Prediction of Shock Outcomes through Machine Learning

Authors:
Sharad Shandilya
Xuguang Qi
Kayvan Najarian
Kevin Ward
Michael Kurz
Rosalyn Hargraves

Keywords: predictive model; overfitting; machine learning; defibrillation success; parameter search.

Abstract:
Predicting defibrillation success is of paramount importance to resuscitating a victim of cardiac arrest. Several studies have attempted to extract/discover predictive features from electrocardiogram signals. Till date, no method has been accepted or implemented in the field, primarily due to low accuracy and/or specificity. We process a relatively large database of signals and report performance of an integrative Machine Learning model through multiple measures. 358 signals, with 140 leading to return of spontaneous circulation through defibrillation attempts and the rest, 218 signals, leading to unsuccessful defibrillations were used to train and test the model on non-overlapping sample sets. Techniques from machine learning, non-linear dynamics and signal processing were applied to extract features and subsequently classify them. In this study, we identify opportunities for reducing variance in the predictive model and propose a method for searching the optimal model. The accuracy and Receiver Operating Characteristic area of the proposed model are 78.8% and 83.2%, respectively. These compare with 74% and 69.2% accuracy and Receiver Operating Characteristic area for the leading 'Amplitude Spectrum Area' measure. The performance of the model will further improve with addition of other physiologic signals, as previously shown in a study by our research group. The model shows great potential to be viable in the clinical setting.

Pages: 214 to 218

Copyright: Copyright (c) IARIA, 2013

Publication date: July 21, 2013

Published in: conference

ISSN: 2308-4529

ISBN: 978-1-61208-283-7

Location: Nice, France

Dates: from July 21, 2013 to July 26, 2013