A novel online method for identifying motion artifact and photoplethysmography signal reconstruction using artificial neural networks and adaptive neuro-fuzzy inference system

Photoplethysmography (PPG) is a noninvasive technique to measure blood volume changes in blood vessels. Despite the wide usage of PPG signal in medical and non-medical applications, this signal can be affected by the motion artifacts leading to data loss. In this paper, we proposed an algorithm to detect motion artifacts and reconstruct the corrupted parts of the signal using real-time modeling based on multilayer perceptron (MLP), radial basis function (RBF) artificial neural networks (ANNs) and adaptive-neuro fuzzy inference system (ANFIS). The developed algorithm was applied to reconstruct the corrupted parts of PPG signals of 23 healthy 25- to 28-year-old volunteers. In the experimental phase, the left- and right-hand PPG signals of the volunteers were simultaneously obtained. While the left hand of the subjects were fixed, they were asked to shake their hands without any predetermined pattern, to simulate the real-life motion accelerations. To statistically and physiologically evaluate the performance of the proposed models, Pearson correlation coefficient (PCC), intraclass correlation coefficient (ICC), Bland-Altman plot (with the 95% limits of agreement), and time-domain feature analysis tests were adopted. The results indicated that the ANFIS with subtractive clustering algorithm shows the best performance in modeling the lost parts of the right-hand signals with an average PCC and ICC of 0.80 and 0.77, respectively, with the reference signal (left-hand signals) over all the tests. Also, the proposed ANFIS-based algorithm had an ability to retrieve the important time-domain PPG signal features, namely mean of inter-beat intervals (NN), standard deviations of NN (SDNN), root mean square of standard deviations (RMSSD) and standard deviation of standard deviations (SDSD) without any significant difference at p < 0.05 level to those of the reference signal (left-hand signal).