Transabdominal fetal oximetry without explicit fetal signal extraction

Abstract

This disclosure provides a fetal-blood-oxygen-saturation estimation technique using a deep neural network without performing explicit fetal signal extraction from mixed maternal-fetal photoplethysmogram (PPG) signals. In one aspect, the disclosed fetal-blood-oxygen-saturation estimation technique receives multiple channels of PPG signals from two or more photodetectors detecting transabdominal diffused light from two or more light sources emitting two or more distinct wavelengths, wherein the photodetectors and light sources are positioned on a maternal abdomen. Note that each channel of the multiple channels of PPG signals includes mixed maternal-fetal PPG signals. Next, the disclosed estimation technique processes the received PPG signals using a trained deep neural network to directly estimate fetal-blood-oxygen-saturation by: tagging the PPG signals with a set of signal-quality levels; feeding the tagged PPG signals as inputs to the deep neural network; and directly inferring, by the deep neural network, fetal-blood-oxygen-saturation estimations and associated confidence levels based on the tagged PPG signals.