Cardiac Dysrhythmia Detection with GPU-Accelerated Neural Networks
Computer Science Department, Stanford University
Stanford University, 2014
@article{haque2014cardiac,
title={Cardiac Dysrhythmia Detection with GPU-Accelerated Neural Networks},
author={Haque, Albert and EDU, STANFORD},
year={2014}
}
Cardiac dysrhythmia is responsible for over half a million deaths in the United States annually. In this work, we evaluate the performance of neural networks on classifying electrocardiogram (ECG) sequences as normal or abnormal (arrhythmia). Using neural networks as our primary learning model, we explain our model’s performance and discuss hyperparameter tuning. Comparing the results of our model to SVMs, random forests, and logistic regression, we find that our neural network outperforms the other three models with a binary classification accuracy of 91.9%. For the multi-class classification task, we achieve an accuracy of 75.7%. The use of GPUs accelerates the neural network training process up to an order of magnitude over CPUs.
January 8, 2015 by hgpu
