散射相位函數對子宮頸基質層組織模擬之影響

Abstract

漫反射光譜（Diffuse Reflectance Spectroscopy,DRS）是一種非侵入式技術，常用於偵測組織中的組成成分。利用蒙地卡羅演算法模擬光子在不同散射及吸收特性組織內之行進情形，藉此可以得到蒙地卡羅順向模擬光譜，再透過順向模擬光譜進行組織漫反射的光譜校正，並藉由模擬光譜與量測光譜的比對，最後，逆向擬合定量出組織的光學參數，例如：散射係數、吸收係數、血氧飽和、血紅素濃度、膠原蛋白濃度及上皮層厚度等。 在過去擬合順向模擬光譜與組織漫反射光譜時發現在光源（Light source）跟偵測端（detector）的間距短時在血紅素吸收波鋒經常無法達到較好的擬合，在過去模型中，散射相位函數（Scattering phase function）使用Henyey-Greenstein（HG），在新模型中，我使用Modified Henyey- Greenstein（MHG）作為相位函數，再由不同的測試資料測試過去模型與新模型的表現並比較其結果。最後使用此新模型用於萃取臨床光譜之光學參數，並分析在Normal、LSIL、HSIL情形下光學參數的改變。 本論文致力於改善組織模型及逆向擬合流程，建立出更貼近真實活體組織的模型以及能縮小模擬光譜與實際量測光譜間誤差的逆向擬合方法，最後利用一輸入為光學參數、輸出為模擬光譜的人工神經網路（Artificial neuron network, ANN）來取代原先順向使用的蒙地卡羅，縮短取得光學參數所需時間。並使用此新方法萃取出臨床光譜之光學參數再加以分析。

Diffuse Reflectance Spectroscopy（DRS）is a non-invasive technique for detecting the composition of tissues. We can get the forward Monte Carlo simulated spectrum with simulating the travel situation of photons in different scattering and absorption tissues by Monte Carlo method. Comparing forward Monte Carlo spectrum with tissue spectrum. Finally, use the inverse fitting method to extract the optical properties of tissues. The past fitting results have higher errors in the wavelength where hemoglobin has characteristic absorption peaks when the SDS is too small. In the past model, Henyey- Greenstein phase function（HGpf） was used. Here, I proposed to use Modify Henyey- Greenstein（MHGpf） as phase function, and verified that using MHGpf is more suitable than using HGpf in the simulation. Besides, the main absorption in stroma is hemoglobin. So hemoglobin concentration and oxygen saturation have the most influence toward the shape of spectrum. The shape of spectrum that wavelength between 410~440nm and 540~580nm shows the influence of hemoglobin. Here I use two-step fitting method. First step is to fit spectrum that wavelength between 410~440nm and 540~580nm and then restrict the range of hemoglobin concentration and oxygen saturation to do the second step of fitting. This thesis aim to improve the tissue model and the inverse Monte Carlo fitting method. Besides to build a more actual model that can reduce the error between simulation spectrum and the measure spectrum, I also use Artificial neuron network（ANN） to substitute the forward Monte Carlo. With the use of ANN, we can greatly accelerate the process.