Cerebral blood flow monitoring using a ConvGRU model based on diffuse correlation spectroscopy

Abstract

Continuous monitoring of cerebral blood flow (CBF) provides crucial information for clinical diagnosis and treatment of various cerebral diseases. Diffuse correlation spectroscopy (DCS) uses near-infrared (NIR) coherent point-source illumination to accommodate spectroscopic measurements of CBF variations. In this paper, we investigate and evaluate a deep learning method for CBF quantification based on proposed ConvGRU model. Two in vivo experiments, i.e., deep-breath experiment and breath-holding experiment, were established to measure normalized intensity autocorrelation function data. Compared to conventional methods, promising results for assessing changes of CBF were gained by the developed ConvGRU models. Our results suggest that ConvGRU-based deep learning method can provide an alternative method for continuous monitoring of CBF.