雙層組織模型對定量子宮頸黏膜組織內在螢光特徵之影響

Effects of Tissue Model Layers on the Quantification of Intrinsic Fluorescence Characteristics of Cervical Mucosa

林國聖(Guo-Sheng Lin)

指導教授 ： 宋孔彬

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202201673

摘要

本論文研究的最終目的是測量活體子宮頸黏膜光譜，以數值模型推算組織內螢光物質含量，研究其是否有偵測癌前病變之能力。依據先前研究基於兩段式曲線擬合流程，建立單層均質組織模型並與雙層模型比較確立組織層數對於定量子宮頸黏膜組織內在螢光特徵之影響，以找出降低擬合活體光譜誤差的方法。其中內在螢光代表不受組織散射吸收影響的內在螢光光譜，可由螢光效率及螢光波型組成，可以確實反映出組織內部螢光物質的情況。而整套流程是基於蒙地卡羅法(Monte Carlo method, MC)作為描述分析光譜之模型。輸入組織光學參數通過數值模型計算出不同波長下的反射率會形成光譜為正向模型，透過調整輸入參數使輸出的光譜與目標活體光譜有最小誤差為光譜擬合為逆向模型，此時的輸入參數一定程度象徵光譜的組成。在漫反射光譜的分析上使用人工神經網路(artificial neuron network, ANN)取代MC作為產生光譜的正向模型，因為在逆向模型中需要多次使用正向模型輸出光譜，透過ANN可以大幅降低正向模型時間增加擬合效率。最終利用三個實驗完成整個研究，首先以100組隨機參數組合對應的模擬光譜驗證單層模型的可用性，並證明單層與雙層模型皆可以順利使用。接著使用活體光譜擬合結果的37組參數組合以雙層模型MC模擬出目標光譜後分別以單層模型和雙層模型擬合。定量內在螢光效率的結果，使用單層模型進行擬合，誤差至少是使用雙層模型擬合的七倍。而內在螢光波形在兩種組織模型假設下結果差異不大。此外調整了光源偵測器距離為0.22 mm光纖的光譜中螢光物強度貢獻比。菸鹼醯胺腺嘌呤二核苷酸磷酸(nicotinamide adenine dinucleotide, NADH)和膠原蛋白的比值分別為0.25和1代表正常組織和癌化組織，在雙層模型下結合多通道系統相較單層組織模型可以更準確定量此兩種螢光物質效率比例。最後招募懷疑有子宮頸癌前病變並接受陰道鏡檢查的受試者，進行活體實驗，測量並分析每位受試者組織切片部位光譜和正常部位做為對照組至少兩個部位。在37組活體光譜的螢光光譜分析中單層模型和雙層模型各自平均光譜誤差分別使用學生t檢定和F檢定皆有顯著差異，代表使用更貼近真實組織的雙層組織模型萃取活體光譜內在螢光可以有效降低擬合光譜誤差。進一步分析具有完整分析條件的12位受試者， 計算NADH與膠原蛋白的螢光效率比值，並使用對照組比值正規化切片部位比值，以解決個體差異的問題，結果顯示癌前病變部位約會是正常部位2倍以上。

關鍵字

內在螢光 ； 蒙地卡羅法 ； 漫反射光譜 ； 螢光光譜 ； 子宮頸上皮內瘤樣病變 ； 人工神經網路 ； 基因演算法

並列摘要

The final purpose of this study is to measure the spectrum of the cervical mucosa in vivo, to estimate the fluorescent substance content in the tissue with a numerical model, and to study whether it can detect precancerous lesions. According to previous studies, based on a two-step curve fitting process, a single-layer homogeneous tissue model is established and compared with the two-layer model to determine the effect of the number of tissue layers on quantifying the intrinsic fluorescence characteristics of cervical mucosal tissue, to find out and reduce the fitting error of the in vivo spectrum methods. The intrinsic fluorescence that is not affected by tissue scattering and absorption, which can be composed of intensity or intrinsic fluorescence efficiency, and intrinsic fluorescence waveform, which can truly reflect the situation of fluorescent substances in the tissue. The whole process is based on the Monte Carlo method (MC) as a model to describe the analytical spectrum. The input optical parameters are calculated by the numerical model and the reflectance at different wavelengths will form a spectrum as a forward model. The input parameters are continuously adjusted so that the output has the smallest error with the target spectrum as a fitting inverse model, and the input parameters will represent the composition of the spectrum. In the single-layer model, an artificial neural network (ANN) is used to replace the MC as the forward model for generating the diffuse reflectance spectrum, because the forward model needs to be used multiple times to output the spectrum in the reverse model, and the forward model can be greatly reduced through ANN. To increase the fitting efficiency to the model time, the forward model of the two-layer model is also replaced by an ANN. Finally, three experiments were used to complete the whole research. First, the availability of the single-layer model was verified by the simulation spectra corresponding to 100 sets of random parameter sets, and it was proved that both the single-layer and two-layer models could be used. Then, using the 37 sets of parameter sets of the in vivo spectrum fitting results, the target spectrum was simulated with the two-layer model MC, and then the single-layer model and the two-layer model were respectively fitted. It was found that the quantitative intrinsic fluorescence efficiency, fitted using a single-layer model, was at least seven times more error than fitting using a two-layer model. The shape of the emission fluorescence spectrum was not significantly different under the assumptions of the two tissue models. In addition, the intensity contribution rate of the fluorescent substance in the optical fiber spectrum with a photodetector distance of 0.22 mm was adjusted. The ratio of nicotinamide adenine dinucleotide (NADH) to collagen was 0.25 and 1, respectively, for normal and cancerous tissues. In a two-layer model combined with a multi-channel system, the efficiency ratio of fluorescent species can be quantified more accurately than in single-layer tissue. Finally, subjects with suspected precancerous cervical lesions who underwent vaginal examination were recruited for in vivo experiments to measure and analyze the spectrum of each subject's tissue sections, both normal and biopsy sites. In the fluorescence spectrum analysis of 37 sets of in vivo spectra, the average spectral errors of the single-layer model and the two-layer model were significantly different using the Student's t-test and the F-test, respectively, which means that the two-layer tissue model that is closer to the real tissue is used to extract the intrinsic properties of the in vivo spectrum. the 12 subjects with full conditions were further analyzed. The ratio of fluorescence efficiency of NADH to collagen was calculated, and the ratio of the normal site was used to normalize the ratio of the biopsied site to solve the problem the ratio of fluorescence efficiency of NADH to collagen individual differences. The ratio can be used as a reliable indicator for detecting precancerous lesions and it are found to be more than double the ratio of normal site.

並列關鍵字

intrinsic fluorescence ； Monte Carlo method ； diffuse reflectance spectroscopy ； fluorescence spectroscopy ； cervical intraepithelial neoplasia ； Artificial neuron network ； Genetic algorithm

參考文獻

[1] 中華民國衛生福利部. (2021). 109年死因統計結果分析.

[2] X. J. G. o. Castellsagué, "Natural history and epidemiology of HPV infection and cervical cancer," vol. 110, no. 3, pp. S4-S7, 2008.

[3] R. Drezek et al., "Understanding the contributions of NADH and collagen to cervical tissue fluorescence spectra: modeling, measurements, and implications," J Biomed Opt, vol. 6, no. 4, pp. 385-96, Oct 2001.

[4] G. D. Birkmayer and J. Zhang, "NADH in Cancer Prevention and Therapy," 2004, pp. 541-554.

[5] J. D. Boone, B. K. Erickson, and W. K. J. J. o. g. o. Huh, "New insights into cervical cancer screening," vol. 23, no. 4, pp. 282-287, 2012.