Fast Blood Flow Reconstruction Study Based On Diffuse Correlation Spectroscopy And Tomography

Diffuse correlation spectroscopy and tomography(DCS/DCT)is a novel near-infrared diffuse light technique used for blood flow reconstruction(Blood flow index extraction and imaging)from light electric field temporal autocorrelation data.In this study,the robust regression algorithm and convolutional neural network are used for diffuse correlation spectroscopy(DCS)and diffuse correlation tomography(DCT),respectively.To to overcome two major bottlenecks in blood flow index extraction and imaging: taking into account the real-time performance and robustness of DCS blood flow extraction;and achieving fast imaging of blood flow.For extraction of blood flow indices,we propose the Nth-order linear algorithm(NL algorithm),which can extract blood flow values in homogeneous and non-homogeneous tissues with arbitrary geometry.The real-time and robustness of the NL algorithm depends on iterative linear regression.In this study,two robust regressions(Huber regression and RANSAC:random sampling consistency)are proposed to be combined with the NL algorithm for the first time.It is also compared with the traditional method(OLS:Ordinary Least Square)through computer simulations and clinical trials.The results show that Huber and RANSAC have much higher accuracy in terms of extracting blood flow values than that of OLS(smaller MAPE values)while ensuring real-time performance.For blood flow reconstruction imaging,the traditional reconstruction algorithm is based on the physical model of optical field transmission,The system equations are seriously ill-conditioned,which limits the speed and stability of blood flow imaging.For the first time,this study uses deep learning methods to explore the mapping relationships from optical data to image domains through computer simulation.The powerful learning capability of the encoding-decoding structure of neural networks is exploited to achieve fast blood flow imaging.Compared to traditional algorithms,the method proposed in this paper will make full use of the a priori information of blood flow images and overcome limitations such as the sparsity of optical measurement data,thus improving the speed and stability of blood flow imaging.The results of this paper will contribute to the early detection and critical care of various diseases such as brain and tumour,promote the clinical application of DCS/DCT technology,and provide a reference for other functional imaging or detection techniques that extract physiological parameters from optical data.