Blood Oxygen Saturation Estimation from Fingertip Videos Using Residual Convolutional Neural Network

This paper proposes a novel method for estimating peripheral blood oxygen saturation (SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf>) using RGB signals extracted from fingertip videos captured by a smartphone camera. SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> is a critical indicator for assessing blood oxygen levels, which can significantly decrease in respiratory conditions such as asthma, chronic obstructive pulmonary disease (COPD), and COVID-19. Inspired by the photophysiological model, this method employs a convolutional neural network (CNN) with cascaded residual blocks to extract RGB signals from fingertip videos for SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> estimation. Compared to facial videos, which are susceptible to motion artifacts and ambient light variations, fingertip video signals are more stable. Experimental results on two public datasets demonstrate that the proposed method effectively estimates SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> and outperforms comparable methods. The feasibility and accuracy of SpO<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</inf> estimation based on RGB signals were also validated.