Using Neural Network with Random Weights and Mutual Information for Systolic Peaks Classification of PPG Signals

Abstract

The detection of peaks in photoplethysmogram (PPG) signals is vital to ensure the information gathered from the peaks in an accurate manner. The false peaks will interrupt the accuracy for future classification of any related events. This study presents the implementation of feature enhancement method for systolic peaks classification of PPG signals using mutual information and neural network with random weights (MI-NNRW). MI-NNRW method is proposed to improve the accuracy performance of the conventional NNRW method. MI method implements at sixteen time-domain features and then the NNRW classifier predicts between false and true systolic peaks point of PPG signals. The results indicate that by using sigmoid as activation function, the accuracy of sensitivity (Se) for Intracranial Pressure (ICP) signals can increase up to 81.71 %. Overall, the MI-NNRW method improves the accuracy performance compared to the conventional NNRW method which leads to the improvement of accuracy for the detection of systolic peaks.