SpO2 Estimation Using Deep Neural Networks: A Comparative Study

Recently, there has been a growing interest in using smartphones as non-invasive medical devices for estimating vital signs. One of the most important parameters for assessing a patient's respiratory and circulatory function is the blood oxygen saturation level, commonly known as SpO2. This paper presents a novel approach for SpO2 estimation using smartphone cameras and neural networks. Our proposed method uses the RGB camera of the smartphone to capture images of the fingertip and extract the photoplethysmography (PPG) signal. The PPG signal is then analyzed using a convolutional neural network to estimate the SpO2 value. Specifically, a machine learning algorithm was developed for estimating blood oxygenation levels by analysing and implementing models from the literature and comparing their performance to select the best model. Model validation is performed by creating a smartphone application to capture fingertip images, extract the PPG signal, and use the chosen machine learning model for estimating blood oxygenation levels.