A Non-Invasive Approach for Fetal Arrhythmia Detection and Classification from ECG Signals

This paper aims to present an intelligent system for autonomous diagnosis of fetal arrhythmia based on fetal ECG recordings. The present scheme uses one dimensional (1D) convolution with a wavelet kernel to extract time domain features from subjects possessing normal fetal ECG and fetal arrhythmia ECG. Time- domain features obtained from the convoluted signals are fed to a trained artificial neural network (ANN) with gradient descent learning to identify and classify fetal ECG signals. The experimental evaluation of the proposed scheme has been tested with a six- channel fetal ECG signal, available in the NIFEADB database. An overall accuracy of 96% is obtained by evaluating standard performance metrics. The use of 1D convolution not only reduces the computational burden but also helps to specify the feature space to develop an intelligent system for portable embedded system applications.