PPG motion artifact reduction using neural network and spline interpolation

A new method of removing motion artifact from the Photoplethysmogram signal using multilayer feed forward neural network is described in the paper. 3850 number of beats each containing 7 important clinical features collected from BIDMC dataset are used for training the neural network and 770 number of beats are used for testing. The proposed algorithm performs quite well yielding higher levels of accuracy in conserving the systolic peak to systolic peak distance (peak to peak distance) as compared to other existing algorithms. The error between the average peak to peak distance of the clean PPG signal and the processed PPG signal is 6.19%. Another salient contribution of this work is that it preserves other clinical features quite well. Thus the algorithm can be useful for clinical feature based classification of PPG signal in ambulatory monitoring.