A non-contact oxygen saturation estimation using Video Magnification and a Deep Learning method

In this paper, we present a new non-contact strategy to estimate the Peripheral Oxygen Saturation (SpO(2)) based on the Eulerian motion video magnification technique and a Machine Learning method. The magnification procedure was carried out using the Hermite decomposition on the blue and red channels of the image frame. The SpO2 is estimated from a 3D Convolutional Neural Network from a ROI detected using transfer learning models on the Machine Learning Yolov5. In this proposal it is not needed a calibration step per subject for the estimation. We have tested the method on 18 healthy subjects. Each video includes the subject and the data of the contact pulse oximeter device that was detected automatically. To compare the performance of the methods, we compute the Mean Absolute Error, The Root Mean Squared Error and metrics issues from the Bland and Altman analysis to investigate the agreement of the methods with respect to a contact pulse oximeter device as reference. The proposed solution shows an agreement with respect to the reference of 98%.