使用共光路斷層繞射顯微術之單細胞幾何與光學體積研究

Abstract

無標記定量相位顯微術以光束穿透細胞後，獲取細胞內部組成差異所造成的二維相位差分布，該量測到的分布是由穿透光沿光軸通過樣本後的折射率線積分。依據繞射斷層掃描術原理，可藉由獲取不同角度下的二維相位影像進行重建而得到樣本的實際厚度與三維折射率分布影像。為了達到高通量掃描的之目的，本論文開發二維相位影像回復與三維折射率分布影像重建之平行運算來提升成像效率。 由於紅血球型態變異資訊可作為辨別血液相關疾病的診斷依據，本論文採用共光路式斷層繞射顯微術獲取紅血球的三維折射率分布影像，藉由定量健康紅血球與海洋性貧血紅血球的平均折射率、相位分布、光學體積與三維形態特徵來區分健康受試者與具有海洋性貧血之患者。其結果顯示由海洋性貧血紅血球較正常紅血球具有較小的光學體積與表面積與體積比值、球型指數和表面積四個特徵所建立的判別模型可獲得優異的分類結果，證實斷層繞射顯微術在區分健康與海洋性貧血紅血球上相當具有潛力。 雖然透過斷層繞射顯微術可以準確地區分出輕度海洋性貧血患者與正常受檢者，但其光學系統架構的複雜度和資料擷取與處理效率導致該系統難以直接應用於臨床上。為了解決此問題，本論文亦採用數位全像顯微術獲取紅血球的二維相位影像並使用遮罩區域卷積神經網路技術建立點對點分類模型獲得比傳統影像處理自動偵測更高的效率，並由該模型分割出的紅血球區域進行定量相位影像相關特徵計算，分析其與三維折射率影像特徵之相關性。 為了充分利用斷層繞射顯微術可獲得定量影像之優勢，本論文採用時序性掃描獲取一系列的視網膜色素細胞行有絲分裂的三維折射率分布影像。由於有絲分裂是一動態過程，因此藉由量化視網膜色素細胞的相位差統計、幾何型態變化與運動向量場來建立偵測與預測有絲分裂發生之模型。本實驗所建立的有絲分裂偵測與預測模型在分類準確度上皆高達100%的辨別結果，因此藉由模型的判辨能力將有助於分析單一細胞在微環境改變下的反應。

Label-free quantitative phase imaging (QPI) is capable of mapping in two dimensions the phase shift caused by cellular constituents when a source light been transmits though a transparent cell. The measured phase shift represents the line integral of the refractive index (RI) contrast between the specimen and its environment along the light path which is parallel to the optical axis and corresponds to physical thickness of the specimen. The physical thickness and three-dimensional (3-D) RI maps of specimen can be reconstructed from multiple two-dimensional (2-D) phase images acquired at various illumination directions using diffraction tomography techniques. To achieve high-throughput screening, a self-programmed software was developed to enhanced the efficiency by parallelizing the computation of 2-D phase retrieval and 3-D RI tomogram reconstruction. Since altered red blood cell (RBC) morphology is an important feature in distinguishing a variety of blood-related diseases, tomographic diffractive microscopy (TDM) was used to acquire 3-D RI tomograms of RBCs. The mean of refractive index, distribution of phase shift, optical volume and 3-D morphological features of healthy RBCs and thalassemic RBCs were measured to distinguish healthy subjects and patients with thalassemia. A multi-indices prediction model achieved perfect accuracy of diagnosing thalassemia using four features including the optical volume, surface-area-to-volume ratio, sphericity index and surface area. The results demonstrate abilities of TDM to provide quantitative, hematologic measurements and assess morphological features of erythrocytes to distinguish healthy and thalassemic erythrocytes. Although excellent accuracy of distinguishing thalassemia-minor patients from healthy subjects has been demonstrated, the complexity of the instrument, and inefficiencies in both data acquisition and data processing prevent the technique to become a clinical tool. To address the issues, digital holographic microscopy (DHM) was used to acquire 2-D QPI data of RBCs, and the Mask region-based convolutional neural network (R-CNN) technique was implemented to perform end-to-end classification with higher detection efficiency than conventional image processing-based methods. In addition, features extracted from quantitative phase images of RBCs segmented automatically by the Mask R-CNN model to characterize QPI features of thalassemia, and analyzed correlations between the 2-D QPI features and those extracted from 3-D RI tomograms of the same RBCs. To fully exploit TDM, a sequence of time-lapse RI tomograms of in vitro retinal pigment epithelial (RPE) cell was acquired to record the evolution of mitosis. Since mitosis is a dynamic process, the optical statistics, cell morphology and motion vector fields of RPE cell were characterized to build models for detecting and predicting mitotic events. Both the mitotic classifier and predictor have achieved perfect accuracy (100%) that can aid the single cell analysis by investigating cellular response to changes in the microenvironment.