Development of an artificial neural network for recovering the optical properties of superficial volume of biological tissues in the non-diffusion regime

• Main Authors: Yu-WenChen, 陳昱文
• Other Authors: Sheng-Hao Tseng
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/896839

碩士 === 國立成功大學 === 光電科學與工程學系 === 102 === Typical diffuse reflectance systems can work with photon diffusion models to accurately determine the absorption coefficient (μ_a) and reduced scattering coefficient (μ_s') of tissues in the wavelength range from 650 to 1000 nm, where tissues have high-albedo so that the diffusion approximation is satisfied. In this thesis, we used a steady state diffuse reflectance system and a novel algorithm to determine the physiological parameters of superficial biological tissues at wavelengths ranging from 500 to 1350 nm. We combined the Monte Carlo method (which is accepted as the gold standard approach for photon migration modeling) with Compute Unified Device Architecture, in which a parallel computing platform and programming model were implemented by the graphics processing units to establish the reflectance database with high speed. We further utilize the database to establish a connection between the optical properties and diffuse reflectance spectra with Artificial Neural Network. With this novel model, we can accurately and immediately simulate every conditions of photon migrating in the tissue without any limitation. In this study, we employ this new algorithm to reveal the optical properties of different liquid phantoms and the performance of this model will be surveyed. We also measured different positions of human skin and recovered the absorption and reduces scattering spectra. The derived absorption spectra can be fit linearly with the known chromophore absorption spectra to obtain the concentration of chromophores including oxygenated hemoglobin, deoxygenated hemoglobin, water, melanin, and collagen. We found that the chromophore fitting performance by taking longer wavelength (1000 to 1350 nm) into account is more reasonable than that obtained by analyzing the short wavelength spectra (500 to 1000 nm) only.