Leveraging discriminative dictionary learning algorithms for single lead ECG classification
Date
2015
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Detecting and classifying cardiovascular diseases and their underlying etiology are necessary in critical-care patient monitoring. In this thesis, we explore the effectiveness of discriminative dictionary learning algorithms for electrocardiogram (ECG) classification task and exhibit that they can achieve very competitive performance compared to traditional methods with lower computational cost. We demonstrate dictionary learning and classification processes simultaneously following the detection of supraventricular and ventricular heartbeats using a single-lead ECG. Label information for each dictionary atom is incorporated to enforce discriminability in sparse codes during the dictionary-learning process. Such a discriminative label-consistent learning procedure for adapting both dictionaries and classier to a specified ECG signal, rather than employing pre-defined dictionaries is novel. The effectiveness of the proposed algorithms is demonstrated on real ECG signals from the MIT-BIH arrhythmia database. The performance of the algorithm is evaluated in terms of classification accuracy, sensitivity, positive predictive value and false positive ratio. The results demonstrate a classification accuracy of 94.61% for Supra Ventricular Ectopic Beats (SVEB) class and 97.18% for Ventricular Ectopic Beats (VEB) class at sampling rate of 114 Hz on MIT-BIH database. Therefore, a sampling rate of 114 Hz provided enough discriminatory power for the classification task. Results illustrated that our approach gave emulous results as compared to the state of the art models at a lower sampling rate and a set of simple features.