Fast Monte Carlo fitting of two-layered tissue structures for short source-detector distances

Abstract

New small-scale optical probes designed for endoscopic near-infrared imaging, can have source-detector distances considerably less than 1 cm, pushing the limits for which the diffusion approximation is valid for quantifying optical properties. Scalable Monte Carlo algorithms have been successful in recovering optical properties in this realm for homogeneous tissue; however, these techniques are unsuitable for heterogeneous media. The purpose of this study was to implement a fast Monte Carlo algorithm to retrieve optical properties in media comprised of two distinct layers (a geometry similar to that which would be seen in endoscopic prostate imaging where the rectal wall is superficial to the prostate). A two-layer phantom was constructed from two materials with known optical properties (one 5-mm thick slab was placed on a 3-cm thick slab). The phantom was raster-scanned using a reflectance-based system with a 3-mm source-detector separation. Data was analyzed using Monte Carlo eXtreme as a forward model in an iterative fitting procedure. On average, 125 iterations per pixel were required for convergence (3.8 minutes of computational time on a single nVidia GTX480 GPU card). The errors in recovered absorption coefficients were 0.58% and 0.39% in the top and bottom layers, respectively. This work demonstrates the promise of ultrafast Monte Carlo algorithms for applications within an iterative fitting routine for geometries where the diffusion approximation fails.