OxiTidy: motion artifact detection-reduction in photoplethysmographic signals using artificial neural networks

Abstract

Nowadays, technological evolution has allowed advances in several areas, especially in healthcare. Digital transformation in health has brought benefits to both professionals and patients. What was possible to do only with high-cost and unwieldy biomedical equipment, has been popularized with the emergence of wearable devices. This technology allows clinical monitoring beyond medical offices, being able to be incorporated into the daily life of patients and working as another tool for prevention and promotion of health and well-being. Among the various features present in wearables is pulse oximetry. Through this non-invasive technique, it is possible to measure physiological parameters, such as oxygen saturation (SpO2) and heart rate (HR). However, the way pulse oximeters are developed and used directly influences the quality of the information provided to the user. Photoplethysmographic (PPG) signals from pulse oximeters are susceptible to noise, which is largely caused by user movement during monitoring. These motion artifacts can cause measurement errors and false alarms. In order to mitigate these issues, this work proposes an algorithm based on artificial neural networks (ANNs) capable of detecting and reducing the undesirable effects produced by noise in PPG signals. The performance of this algorithm, called OxiTidy, was compared with three other approaches — raw, discrete Fourier transform (DFT) and simple moving average (SMA) —, using data from 17 healthy volunteers. OxiTidy identified the intervals where the measurements were incorrect and estimate new SpO2 values with a good approximation to the readings performed by a pulse oximeter certified by the Brazilian Health Regulatory Agency (Anvisa).