Signal Processing for Diffuse Correlation Spectroscopy with Support Vector Regression

Many diseases would lead to abnormal blood flow in biological tissues. In order to extract the blood flow index (BFI) in tissues with heterogeneity and irregular geometry, we previously proposed an innovative Nth linear algorithm (i.e., NL algorithm) for technology of diffuse correlation spectroscopy (DCS), in which the DCS signals are fully utilized through iterative linear regression. With this approach, the BFI calculation is remarkably affected by the performance of linear regression. In this study, we proposed to use the support vector regression (SVR) method to denoise the DCS data by implementing the iterative linear regression. In addition, two other approaches (least-squared regression, least-absolute regression) were compared for quantitative evaluation. The DCS data generated from computer simulations with varied tissue models (i.e., flat tissue, human head, human limb, and mouse head) and those collected from the phantom experiments were utilized to evaluate the three approaches. Both simulation and phantom experiment results show that the SVR method has the best performance among three methods in extracting the BFI values, regardless of the tissue geometry and size.