蒙地卡羅於新型逆向擬合模型與活體漫反射光譜研究：口腔黏膜組織

Abstract

漫反射光譜為非侵入式量測系統，目前開發於診斷癌症前病變，目的為診斷發生癌症前的病變狀況，透過非侵入方式早期診斷，使病人能及早治療降低風險。癌前細胞變化於組織結構內部，無法從肉眼觀察，但其變化能反映於組織的光學性質改變。透過蒙地卡羅方法建立組織模型模擬光子於組織中的傳遞與吸收，可計算出不同組織光學參數下的漫反射光譜，再藉由光譜曲線擬合可從活體口腔黏膜量測的光譜定量出組織光學參數。 本研究以區域光譜能提供不同參數資訊為概念，分為兩大部分，先是進行開發一新型擬合模型，改善蒙地卡羅於逆向擬合之缺點，然而分析實際口腔活體光譜時，只是擬合方法的改善不足以修正實際量測與模擬運算上之誤差，因此後半段研究是針對活體光譜的分析與順向模型的調整。 新型擬合模型主要是針對傳統遞迴式光譜擬合方法參數萃取不穩定以及耗費大量時間之缺點，使用新型模型利用漫反射光譜區域波段的敏感度不同，配合事先建立的表格，萃取各參數初估值，降低運算中的不穩定性以求更高精準度；而後將初估參數代入配合表格與矩陣運算，免去曡代過程中針對各變數運算所需的蒙地卡羅，萃取出適當光學參數，提升時間與計算效率。活體光譜分析部分，配合本實驗室硬體系統的架設與量測而得到人體口腔黏膜的漫反射光譜，藉由蒙地卡羅模型分析獲得活體口腔光學特性，然而目前結果不甚理想。藉由修正運算、分析誤差關係、探討真實組織情況等方式，重新檢視順向模擬模型並嘗試改善，以求減少活體光譜計算上的誤差。

Diffuse reflectance spectroscope is a non-invasive method to detect cancer. People use this technique to develop early cancer diagnosis. Patients can accept treatments in time and reduce risk by pre-cancer screening. When carcinoma in situ or even earlier lesions, the tissue pathological changing cannot be detected by visual examination. However, tissue changing under surface corresponds on the different optical properties. The interactions of photon in tissue can be simulated by Monte Carlo method. We can generated the different diffuse reflectance spectra by assigning optical properties. Target spectra are diffuse reflectance spectra which the optical parameters are unknown. Through fitting simulative spectra with target spectra, we can analyze the unknown of target spectra. By detecting diffuse reflectance spectra from oral cavity, we can diagnose early cancer and even quantify optical properties of mucosal tissue. In this thesis, the main concept is different range of diffuse reflectance which can provide different information of optical properties. There are two majored purposes in this thesis with the main concept: One is developing the new inverse fitting model, the other is investigating in vivo diffuse reflectance spectra. Traditional iterative curve fitting is unsteady and time-consuming by Monte Carlo method. We design a new fitting model which use sensitive features of wavelength depending and combine with looked-up table. Through new inverse fitting model, target spectrum is firstly estimated the initial parameters set, then iteratively calculated by matrix. Finally, we can effectively extract optical properties from target spectrum without as many as times of Monte Carlo method. Diffuse reflectance spectra, from human oral mucosa, are acquired by three optical fibers at different source-detector separations. We extract the optical properties from measured spectra by Monte Carlo and fitting processing. However, there are errors, 19-29%, between simulated fitting and measured target. It cannot be improved by only adjusting inverse fitting model. We are going to research forward model because the different, between simulation and reality, should come from the tissue model used in Monte Carlo calculation. In this part of research, there are discussing of spectra different, surveying tissue model, and adjusting calculated method. Try to approach the simulative model to real mucosal tissue.