時域近紅外光譜術於人體頭部光學參數定量分析及系統驗證

Abstract

光學技術提供了非侵入式的方法來測量組織生理參數和進行治療。例如，功能性近紅外光譜能夠監測大腦的血液動力學變化、經顱紅外光刺激具有提升大腦認知功能的效果等等，使得光學技術被廣泛應用於大腦行為監測和治療研究上。然而，光在組織中的行進特性是由各組織的光學參數描述，這些參數會因頭部結構及物質濃度的不同而有所變化。因此，提升數據準確度及光穿透深度依賴於對頭部組織光學參數的精確定量。 本研究使用波長為800 nm的雷射皮秒光源、混合式光感測器以及計時電路模組建立時域近紅外漫反射光譜系統。使用多個距離在1.5 cm至4.3 cm之間的偵測器，並計算各偵測器之光子在組織中的飛行時間，以獲得光子飛行時間分布。透過測量系統響應函數進行校正，可將實驗值和模擬值比較並進行疊代式曲線擬合，初步以單層仿體驗證系統數據準確性及其定量光學參數的能力。 基於三位受試者之頭部核磁共振影像建立頭皮、頭骨、腦脊髓液、灰質、白質及額竇共六層組織的三維頭部模型，進行蒙地卡羅法模擬光子在頭部的散射及吸收事件，取得光子在各層組織中之行進路徑長，並經由計算取得光子飛行時間分布。為了加快模擬速度，使用白蒙地卡羅法配合平滑化技術去除模擬雜訊，生成大量訓練資料用於訓練神經網路。後續以神經網路取代傳統蒙地卡羅法模擬作為疊代式曲線擬合的順向模型，大量縮短了擬合時間。 在評估最適合的光源-偵測器距離及時間區段後，本研究使用模擬光譜作為擬合目標，透過疊代式調整擬合參數數值，逐步找到與擬合目標接近的光學參數組合，從而獲得光學參數最佳解。以三個頭部模型驗證此流程能夠有效找到合理誤差範圍內之各層光學參數解。為了更進一步定量多波長下之光學參數，本研究同步擬合了寬頻(continuous-wave)近紅外光譜術在波長700-900 nm範圍內的模擬光譜與時域近紅外光譜術之飛行時間分布。結果顯示，結合兩系統數據確實能夠顯著提升光學參數定量的準確性。

Optical technology provides a non-invasive method for measuring physiological parameters of tissues and for therapeutic applications. For example, functional near-infrared spectroscopy can monitor cerebral hemodynamics and transcranial infrared light stimulation can enhance cognitive functions. These advantages have led to the widespread use of optical technology in brain behavior monitoring and therapeutic research. However, the propagation characteristics of light in tissue are described by the optical parameters of each tissue, which vary with head structure and substance concentration. Therefore, improving data accuracy and light penetration depth relies on the precise quantification of the optical parameters of head tissues. This study established a time-resolved near-infrared spectroscopy (TR-NIRS) system using an 800 nm picosecond laser source, a hybrid photodetector, and a timing circuit module. Multiple detectors spaced between 1.5 cm and 4.3 cm were used to calculate the photon travel time in tissues, thereby obtaining the distribution of time of flight (DTOF). By measuring and calibrating the impulse response function, experimental values were compared with simulated values, allowing iterative curve fitting. Initial validation of the system's data accuracy and its capability to quantify optical parameters was performed using single-layer phantoms. Based on MRI images of three subjects, a three-dimensional head model consisting of six layers—scalp, skull, cerebrospinal fluid, gray matter, white matter, and frontal sinus—was constructed. Photon scattering and absorption events in the head were simulated using the Monte Carlo method, and photon travel path lengths in each tissue layer were obtained to calculate the DTOF. To accelerate the simulation, the White Monte Carlo method and smoothing techniques were employed to remove simulation noise, generating a large training dataset for neural network training. The neural network was then used to replace traditional Monte Carlo simulations as the forward model for iterative curve fitting, significantly reducing the fitting time. After evaluating the optimal source-detector distance and time intervals, the study used simulated spectra as the fitting target. By iteratively adjusting the fitting parameter values, the optimal set of optical parameters was identified, matching the fitting target. The process was validated using three head models, demonstrating that it could effectively find the optical parameters within a reasonable error range for each tissue layer. To further quantify the optical parameters at multiple wavelengths, the study simultaneously fitted the simulated spectra of continuous-wave near-infrared spectroscopy within the 700-900 nm wavelength range and the DTOF from TR-NIRS. The results confirmed that combining data from both systems significantly improved the accuracy of optical parameter quantification.