Neural Network based indicative ECG classification

Abstract

The Electrocardiogram (ECG) is undoubtedly the most used biological signal in the clinical world and it is a means for detection of several cardiac abnormalities. Pattern recognition, diagnostic classification of ECGs constitutes an interesting application of Artificial Neural Networks (ANNs). This paper illustrates the ability of a feed-forward back propagation using Neural Network for classify unknown ECG waveforms keen on one of the 4 discrete class. Out of the 4 classes, 3 of them correspond to abnormal ECG signals and 1 represents the healthy group. In addition, the Neural Network model developed has the option to categorize unknown ECG input signals as unclassified, since it represents an unknown pathology. Preliminary results are obtained using data from 4 different Physiobank ECG database.

Ecg Signal Analysis

“…Ashutosh Gupta et al [21] have presented a non-linear neural network model for ECG classification. They have used two layered feed-forward back-propagation network with sigmoid output neurons in order to grade the input ECG dataset into four different signal classes: Myocardial Ischemia Qibin Zhao and Liqing Zhang [22] have used support vector machine (SVM) with Gaussian kernel to classify different ECG heart rhythm.…”

Untitled

“…For convenience, Person_01/rec_1 record of MIT-BIH ECG ID database has been renamed as P1(1). The 72 extracted features are subjected to a two layered feed forward network, with Sigmoid hidden and Softmax output neurons, which can classify vectors arbitrarily well, given enough neurons in its hidden layers [24]. From Figure 9, the 72 generated features were taken as input.…”

“…At the primary stage, the Neural Network has to be trained with ECG data of different persons, and then the neural network generated from the training is used for biometric identification of the individuals [23][24]. MIT-BIH ECG ID Database was used for training and testing purposes.…”