利用多輸入神經網路及蒙地卡羅組織模型定量中央靜脈血氧飽和度

Abstract

本研究將重點集中在以非侵入性的光學方法，透過內頸靜脈(internal jugular vein, IJV)定量中央靜脈血氧飽和濃度(central venous oxygen saturation，ScvO2)。在實驗中，我們透過近紅外光作為光源將光束入射受試者之頸部組織，可以量測受試者組織之漫反射光譜(diffuse reflectance spectrum, DRS)。 另一方面，為了將量測之光譜予以分析，本研究基於蒙地卡羅演算法，開發IJV組織之模型，並以活體光譜佐以驗證其正確性，同時決定組織模型中數個重要參數，包括其幾何結構、光學係數、以及吸收物質組成。以組織模型為核心，產生大量蒙地卡羅的模擬資料以訓練類神經網路，加速模擬之速度，並再最後利用基因演算法來達成光譜的逆向擬合。以此架構開發出一套功能完整的分析模型。利用該模型有效地分析光譜，定量出組織的血氧飽和度。最後，測試該模型在不同組織預測之表現，並以組織仿體來驗證及估計其誤差及正確性。

In this study, the goal is to noninvasively quantify the oxygen saturation of central venous on the site of internal jugular vein. A near-infrared light source is employed in order to acquire the diffuse reflectance spectrum from the neck of healthy volunteers.

For the sake of interpret the spectrum correctly, an IJV tissue model is needed. According to that, a Monte Carlo based tissue model is built and validated by comparing it with in vivo spectrum. The geometry parameters, optical parameters, and the compositions of the tissue model are determined in this process.

Once the tissue model is validated and feasible for depicting the natural of IJV tissue, it is utilized to generate simulated data for training an artificial neural network, which can accelerate the simulation process dramatically. Finally, genetic algorithm act as the last component of the pipeline, the inverse model, optimizing the similarity between the simulated spectrum and the measured one’s, extracting the physiological parameters such as oxygen saturation. The full process is then tested on different volunteer, and a tissue phantom is used to estimate the precision of the inverse model.