Application of Neural Networks for Heart Rate Monitoring Mohammad Reza Askari ∗ Iman Hajizadeh ∗ Mert Sevil ∗∗ Mudassir Rashid ∗ Nichole Hobbs ∗∗ Rachel Brandt ∗∗ Xiaoyu Sun ∗∗ Ali Cinar*,** ∗ Department of Chemical and Biological Engineering, Illinois Institute of Technology, Chicago, IL 60616 USA ∗∗ Department of Biomedical Engineering, Illinois Institute of Technology, Chicago, IL 60616 USA Abstract: This paper addresses the problem of heart rate (HR) monitoring from photo- plethysmography(PPG) sensors, where artifacts caused by body movements drastically affect the quality of the measurement signal. The PPG signal is windowed into consecutive segments, and for each time-windows, a Butterworth bandpass filter is utilized to attenuate high-frequency noises. Then, the PPG signal is processed by using the singular spectrum analysis technique to obtain a smooth PPG signal. In order to remove artifacts caused by the physical activity of the subject, the 3-dimensional accelerometer signal is used as an auxiliary signal to detect the presence of motion artifact (MA). A new spectral subtraction approach is proposed for MA rejection. For the purpose of HR estimation from the PPG signal, a feature extraction method is performed, and neural network binary classifier is used to detect the most probable frequencies corresponding to the actual HR. HR estimations are passed through a Kalman filter to result in smooth and accurate HR estimations. Keywords: Developments in measurement, Filtering and smoothing, Heart rate estimation, Neural networks, Signal processing. 1. INTRODUCTION Artifacts are disturbances in the measured signal not originating from the process itself, and due to intense corruption of the photoplethysmography (PPG) signal, artifact removal necessitates further investigation. One application where artifacts are prevalent is the estimation of heart rate (HR) using the PPG signal. PPG is an optical technology, which is widely used in wristbands and sports watches to provide an inexpensive and noninvasive method for HR estimation. The main drawback of the PPG signal is that the approach is highly susceptible to corruption from motion artifacts (MA) caused by move- ment. Besides, ambient light disrupts the functionality of the sensor (Zhang et al. (2015b)), which makes the PPG- based method less accurate than the electrocardiogram (ECG) recordings. Artifact removal from a noisy signal is a challenging task because artifacts in the signals are caused by several unknown sources. Furthermore, discriminating between MA and the actual variation of the HR in the PPG signal is a challenging issue because of their similar behavior and having overlapping frequencies in their pe- riodogram. General sources of measurement error include ambient light interference, the location of LED sensors at ⋆ Financial support from the National Institutes of Health (NIH) under the grants 1DP3DK101075 and 1DP3DK101077 and Juvenile Diabetes Research Foundation (JDRF) grant 2-SRA-2017- 506-M-B made possible through collaboration between the JDRF and The Leona M. and Harry B. Helmsley Charitable Trust is gratefully acknowledged. which the signal being recorded, skin pigmentation, and poor contact between the measurement skin surface and the photo-sensor (Castaneda et al. (2018); Zhang et al. (2015a); Zhang et al. (2015b)). A PPG cleaning and MA removal algorithm is developed to generate robust and reliable estimates of the HR by using similarity-based spectral subtraction-signal process- ing. The estimate of HR is obtained by using the Fast Fourier Transform (FFT) and solving a binary classifica- tion problem. Kalman filtering (KF) is used to provide a smooth and accurate estimate of HR. The results show the effectiveness of the proposed denoising and estimation method. 1.1 Related Works Several algorithms are proposed to address the problem of PPG signal denoising in the presence of MA disruptions. A framework to remove the disturbance of motion artifact in the spectrogram of the PPG signal is proposed by Zhang et al. (2015b), which employs singular value decomposition (SVD) to cancel the MA and high-frequency noises. To further enhance the robustness of the scheme, temporal difference operation is applied, and in the final step of signal denoising, spectral signal reconstruction is utilized to approximate the PPG signal with the sparsest signal. Finally, the spectrogram of the PPG signal is calculated by FFT to estimate HR values. Another approach uses the ensemble empirical mode decomposition (EEMD) de- noising technique to remove MA proposed by Khan et al. Preprints of the 21st IFAC World Congress (Virtual) Berlin, Germany, July 12-17, 2020 Copyright lies with the authors 16382