以逆向白蒙地卡羅法分析模擬資料及人體大腦實驗尋找功能性近紅外線光譜技術之最佳化波長組合

Abstract

功能性近紅外線光譜技術(fNIR, Functional near-infrared spectroscopy)是近年來新興的非侵入式大腦功能性影像方法，除了具有造價低、時間解析度高的優點外，還可在開放空間中使用。隨著其發展，除了常見的定量帶氧血紅素及去氧血紅素濃度變化外，有些研究開始偵測細胞色素C氧化酶的氧化還原狀態。由於後者的濃度變化相對於前兩者低很多，使用寬頻系統量測可獲得較佳的靈敏度以及抗雜訊能力。然而寬頻系統具有較高的複雜度且較為笨重，故本研究試著找出在近紅外光譜技術中，要如何選擇使用的波長數量以及組合，以少數幾個波長的光源以及能量偵測，取代寬頻光源與光譜的偵測，達到僅犧牲些許精準度而大幅降低系統成本及提高其可攜性。本研究透過分析模擬資料以及人體大腦量測分析，並使用較為精準的逆向white monte carlo法，取代常見的modified beer lambert law方法，從漫反射光譜萃取大腦中帶氧血紅素、去氧血紅素以及細胞色素C氧化酶的濃度變化。再以基因演算法找出在3~15個波長下的最佳波長組合，以及其對應的誤差。結果得到隨著使用的波長數量增加，定量濃度變化的誤差在7個波長時為5.2%，與使用寬頻(43個波長)得到4.81 %的誤差已非常接近。因此若要以最少波長數量取代全波長的量測，7個波長數量會是較佳的選項。有了上述的結果，往後使用功能性近紅外線光譜技術量測細胞色素C氧化酶，便可以少數波長組合得到與寬頻接近的結果。

Functional near-infrared spectroscopy (fNIR) is a novel non-invasive brain functional imagin. Recently, in addition to quantifying changes in oxy and deoxy-hemoglobin concentration, some studies have begun to measure changes in the redox state of cytochrome C oxidase (CCO). Information regarding changes in CCO concentration has great potential for clinically measuring the metabolic response of the brain. Since changes in the concentration of CCO during brain activation is often much lower than that of hemoglobin, a broadband system is necessary to overcome noise and obtain better sensitivity. However, broadband systems are highly complex and cumbersome. Therefore, this study attempts to identify a method to choose the number of wavelengths and combinations used in near-infrared spectroscopy. Replacing broadband with a few discrete wavelengths can minimize system cost and increase portability with only a small sacrifice in accuracy. This study analyzes simulated as well as in vivo human brain measurement data and, by employing the more accurate inverse white monte carlo method to replace the common modified beer lambert law method, extract the oxy and deoxy-hemoglobin along with CCO concentration changes from diffuse reflectance spectroscopic measurments. Genetic algorithm is used to find the optimal wavelength combination from a range of 3~15 wavelengths and calculate its corresponding error. Results show that using only seven wavelengths achieved an root-mean-square error of 5.2% in estimated concentration chages of oxCCO, oxy- and deoxy-hemoglobin, which is close to the error of 4.8% using the full spectrum of 43 wavelengths. Therefore, 7 wavelengths is an optimal number of wavelengths to replace full wavelength measurement with only minimal losses in accuracy.