Evaluation of convolutional neural networks as an alternative for the non-linear fitting for multiple exposure speckle imaging of blood flow
Abstract
We evaluate in the present study the implementation of a residual convolutional neural network (CNN) to extract blood flow maps based on MESI synthetic exposure data as an alternative to the non-linear fit method. The neural network has been trained on a MESI speckle contrast images database developed using microfluidic channels of various diameter and controlled flows both representative of the physiology of rodents brains.