Deep Learning and Simulation for the Estimation of Red Blood Cell Flux With Optical Coherence Tomography

Abstract

We present a deep learning and simulation-based method to measure cortical capillary red blood cell (RBC) flux using Optical Coherence Tomography (OCT). This method is more accurate than the traditional peak-counting method and avoids any user parametrization, such as a threshold choice. We used data that was simultaneously acquired using OCT and two-photon microscopy to uncover the distribution of parameters governing the height, width, and inter-peak time of peaks in OCT intensity associated with the passage of RBCs. This allowed us to simulate thousands of time-series examples for different flux values and signal-to-noise ratios, which we then used to train a 1D convolutional neural network (CNN). The trained CNN enabled robust measurement of RBC flux across the entire network of hundreds of capillaries.