Signal Processing for Diffuse Correlation Spectroscopy with Recurrent Neural Network of Deep Learning

Diffuse correlation spectroscopy (DCS) is a newly developed near-infrared technology for tissue blood flow measurement. We have proposed an Nth-order linear (NL) algorithm to extract the blood flow index (BFI) in heterogeneous tissue with arbitrary geometry. A critical procedure included in NL algorithm is the linear regression, through which the DCS data are denoised. In this study, we proposed a recurrent neural network (RNN) regression model of deep learning to perform linear regression. For model evaluation, the DCS data at different source-detector separations (i.e., 1.5, 2.0, 2.5 and 3.0 cm) were generated by computer simulations and processed by RNN and the support vector regression (SVR) that was previously proposed by us. The results show that the accuracy of RNN regression model is at the same level as SVR approach at 3.0 cm source-detector separation. However, the RNN model exhibits better robustness than SVR method. The outcomes derived from this study will improve the signal processing for DCS blood flow measurements.