以參數光譜性質建立由漫反射光譜提取粘膜組織參數之方法

Abstract

遞迴式曲線擬合是一常用於由漫反射光譜反推組織光學參數的方法。本研究旨在改進使用傳統遞迴式曲線擬合演算法所常出現的萃取參數值不穩定性以及運算時間過長的問題。首先，根據推測過去作法存在此些缺失的原因來自於比較光譜誤差以制定參數調整策略的方式效率不佳。傳統遞迴式擬合以整體光譜的方均根誤差或方均根百分誤差作為調整組織參數的唯一依據，這樣的做法可能某些方面掩蓋了光譜中能反映個別參數量值的局部特性。因此本研究希望能藉由探討漫反射光譜在各種光纖角度、偵測位置以及波長與各種組織參數間的關係，並以此設計出更適用於雙層組織模型的漫反射光譜組織參數萃取方法。 由於組織各種光學與結構參數間交互作用複雜，光譜對單一參數的敏感度易受其它參數量值影響而變動，相異於許多先前研究中將漫反射光譜對各組織參數的靈敏度作精確的分析，本研究對組織參數的量值判斷採取由粗略至精細的策略。在第一階段的研究中，縮放式蒙地卡羅程式被用來模擬各參數個別變動下各種光纖角度及偵測位置漫反射光譜隨之變化的情形，以及分析各參數對漫反射強度的影響幅度，希望觀察出光譜在各種參數組合下反映特定參數量值變化的普遍性行為，同時探討在各偵測光纖及位置的光譜中各參數間的相對關係。兩層組織的散射係數在漫反射量值的變動中佔有主導的地位，其量值不僅直接左右所有光纖偵測位置的漫反射強度，也間接影響了光譜對包括上皮層厚度以及血紅濃度量值變動的敏感程度。上皮層厚度與光譜變動的相關性決定於兩層組織散射係數間的相對關係。兩散射係數的量值差異放大了上皮層厚度對漫反射強度的影響，兩者強弱關係的反轉也在大部分波段造成光譜對上皮層厚度的增加有完全相反的變動方向。基質層組織的吸收來自於血紅素，其影響波長不超過700nm。血氧飽和度影響吸收光譜的形狀，為最具有明顯特色的組織參數，其影響最彰顯於410~440nm間的吸收峰位置以及540~580nm波段的光譜形狀。血紅素濃度的增加與影響波段漫反射強度呈負相關，光譜對其敏感度與組織的散射係數及上皮層厚度亦息息相關。另外，漫反射光譜大部分波段與組織散射係數、上皮層厚度以及血紅素濃度的變動關係之間都有不錯的線性關係。 第二階段的研究中我們根據在第一階段歸納出各組織參數的光譜特性設計出一套萃取組織參數的複合式方法。此方法根據先粗略再精確的策略，以散射係數比對表格與血紅素濃度的縮放擬合與410~440nm波段吸收峰位置大致推測兩層組織散射係數、上皮層厚度、血紅素濃度以及血氧飽和度量值，再以融合線性代數精神的矩陣方法，由總體光譜誤差推測參數組合偏差進行參數量值的修正。這些方法在使用隨機參數模擬光譜的個別試驗中被證明對於參數量值的判斷具有優異的效果，並且與傳統遞迴式擬合方法相比，在效率上大幅提升。未來，我們將對以矩陣方法的參數修正步驟綱領進行更細部的規範，以確保修正過程的效率及穩定性，並在隨後將此套方法實際應用於活體量測漫反射光譜的組織參數判斷。

Iterative curve fitting is a commonly used method to extract optic-related tissue parameters from diffused reflectance spectra. The goal of the study was to solve the two problems of the iterative curve fitting method: ambiguity of extracted parameters and time consumption. We reckoned that the root of the two problems was the index used in measuring spectra similarity, the root-mean-square error or the root-mean-square percentage error, which caused some information shown in spectra to be overlooked. In this study, our aim was to identify the property of each parameter in spectra from different fiber probe geometry and source-detection separation in different wavelengths, and create a new parameter-extraction method with higher efficiency and accuracy. Considering the complexity of spectral sensitivity distribution associated with parameter value combinations, a rough-to-accurate parameter value estimation strategy was adopted. In the first part of the research, the relations between spectra and parameters were investigated by comparing simulated spectra corresponding to different parameter sets. Of all the interested parameters, the scattering coefficients of both epithelium and stroma dominated. Not only were they highly correlated to the intensity of reflectance but they also influenced the spectral sensitivities to the concentration of hemoglobin and the thickness of epithelium. The spectral sensitivity to the epithelial thickness was related to the relative quantities of the two scattering coefficients. The magnitude of the difference of the two coefficients amplified the importance of epithelial thickness, while the significance decided if the influence is positive. The epithelial absorption influenced light with wavelengths shorter than 700nm and was related to two parameters. The oxygen saturation had an unique influence on spectral shape, and the influence was most obvious in the 410~440nm and the 540~580nm regions. The concentration of hemoglobin had a negative influence on reflectance intensities for most wavelengths, and the spectral sensitivity was related to the value of the scattering coefficients and the epithelial thickness. The spectral variation showed high linearity to value changes for most parameters. In the second part, a hybrid parameter extraction method was proposed based on the result of the previous part. Sticking to the rough-to-accurate strategy, values of scattering coefficients, epithelial thickness, oxygen saturation and the concentration of hemoglobin were roughly estimated by scattering-coefficient-estimation table, peak-StO2 relation curve, as well as the Hb-concentration fitting and epithelial thickness classification, and was followed by value correction with the proposed matrix-approach. The proposed methods were tested with simulated spectra and were proven to be effective in enhancing parameter extraction accuracy and efficiency compared to the recursive-curve-fitting method. In the future, the proposed methods will be applied to reflectance spectra measured in vivo after some minor revisions.