Pixelwise estimation of optical and structural parameters of two-layer tissue using spatiotemporal frequency domain imaging with a shallow neural network

We employed Spatiotemporal Frequency-Domain Imaging (STFDI) to estimate the optical and structural parameters of biological tissue. STFDI uses a time-resolved camera using SPAD or charge modulators with scanned, short-pulse, multi-line illumination. STFDI combines spatial frequency domain imaging (SFDI) and time-resolved spectroscopy (TRS). The spatiotemporal images measured are separated into a spatial modulation transfer function (s-MTF) for SFDI and a phasor trajectory for temporal frequency domain analysis. Based on a two-layer tissue model for each pixel, a shallow neural network (SNN) estimates the optical parameters and thickness of the top layer. We discussed the accuracy of the estimated optical and structural parameters among three types of SNNs: SFDI only, phasor trajectory only, and STFDI. STFDI was the most accurate of the three methods. We also estimated the two-dimensional distribution of optical and structural parameters in an inhomogeneous numerical phantom. The estimated distributions agreed with the ground truth.