高通量免標定光學繞射斷層掃描術於紅血球三維型態之分析

Abstract

作為人體氣體與調控循環系統的核心，紅血球的三維圓盤形狀、兩側表面凹陷程度、血紅素狀態甚至細胞膜表面的黏彈性，都與其物質攜帶交換的能力息息相關。因此可以推論出當其中出現變化勢必會伴隨著相關疾病的存在，像是貧血、糖尿病等患者都有異常血球的出現。臨床上針對紅血球的型態檢驗方式主要是血液抹片，但此方式在判定上較為緩慢且主觀，而在黏彈性等資訊上則需利用滲透壓梯度細胞計數儀等專業器材獲得。因此需多相關研究都提出不同的三維血球分析方式，包含在更高通量下獲取資訊，或是以旋轉樣本或旋轉光源方式拍攝紅血球三維資訊。本文整合並提升這些方式中的技術，達到高通量又能定量分析的三維紅血球量測技術。 定量相位顯微術作為一種免標記技術，可以動態的將樣本厚度及內部物質折射率資訊定量的記錄在干涉影像中，而這樣的技術去除了人為觀察，並可將其中所有參數以數值方式定量記錄。在先前文獻中即提出利用此技術結合微流道建立出高通量斷層繞射顯微系統以提升拍攝通量，並實現三維影像的擷取。建立在此技術之上，本研究將其最佳化並建立高通量斷層繞射顯微系統。在拍攝通量上可達到每分鐘43顆紅血球的拍攝速度，而為使處理速度同樣提升，利用深度學習模型進行影像分割並以圖形處理器進行後續運算上的加速。在三維重建流程中，本實驗提升澤爾尼克多項式擬合角度的正確性，並修正三維重建中資訊缺失的問題。文中將此技術應用在加入不同濃度戊二醛，來模擬血球型態異常之紅血球與一般情況之對比分析。在結果上平均物質質量、光學體積、細胞膜表面黏彈性、大小、體積、血球表面凹陷程度、球型率等變化，皆符合假設與文獻結果且都具有顯著差異。由此可以預期此技術在紅血球相關疾病或是觀察藥物影響的應用上，能以定量且快速的優勢輔助臨床檢驗。

As the central cell of the gas and circulatory system regulation, the characteristic of red blood cells includes disk shape, biconcave geometry, hemoglobin concentration, or even membrane fluctuation all related to the ability to carry and exchange. Therefore, we can speculate that the variation of that property will accompany the disease. For instance, in anemia, diabetes patients have been found to have abnormal RBCs. In clinical, the analysis of RBCs 3D features mainly uses blood smear, but the process is slow and subjective and the viscoelasticity of red blood cells requires professional equipment to measure, such as osmotic gradient ektacytometry. Hence, many studies have proposed different methods to analyze RBC, including higher throughput or captured 3D information by rotating incident light or sample. In this article, we have incorporated some of these techniques to achieve higher throughput while quantitatively measuring RBC. Digital holographic microscopy is a label-free method that can dynamically and quantitatively encode sample thickness and inner substance into an interferogram. This method records all parameters in numerical values to eliminate subjective judgments. In this study, we combined a digital holographic microscope with a microfluidic device to construct optical diffraction tomography to improve capture throughput. The results show that with this setting, 43 RBCs can be captured per minute. As throughput increases, this article uses deep learning and GPU to accelerate processing. For 3D RBC reconstruction, this research optimizes the Zernike polynomial angle estimation algorithm and minimizes the problems caused by the missing angle problem. Finally, this setting was applied to observe red blood cells induced by different amounts of glutaraldehyde to simulate the comparison between abnormal red blood cells and normal red blood cells. The result shows that mean mass density, optical volume, membrane fluctuation, size, volume, sphericity, and concavity are all in agreement with the hypothesis and previous research significantly. From the result, this research is expected to use quantitative and rapid advantages to assist clinical testing on other RBC-related diseases or measure the impact of drugs.