傅立葉疊層顯微術系統與先進影像技術結合應用於高解析度成像

Abstract

本論文對傅立葉疊層顯微術（Fourier ptychography microscopy，FP）的設計、自動化和應用進行了全面的研究。該研究重點通過整合先進技術，包括深度學習和超穎透鏡，來提高本顯微術的性能和功能。 傅立葉疊層顯微術為一種定量相位成像免標記技術，可以在觀測細胞或組織的同時避免因染色傷害細胞造成細胞毒性，最主要的目的是為了解決一個影像系統容易遇到的問題：當在使用的傳統的顯微鏡系統時，為了得到大範圍視野的影像會使用較小放大倍率的物鏡，但此時的影像解析度較低。如果想得到較高的影像解析度，可以提高物鏡的放大倍率，但是高解析度的較大倍率物鏡所能得到的視野較小。傅立葉疊層顯微術的目的即是保持傳統顯微鏡使用較低方大倍率物鏡、低數值孔徑的物鏡來得到較大視野且同時能夠讓影像有高解析度。傅立葉疊層顯微術在低數值孔徑的系統架構下，使用多個不同角度的入射光得到多張影像輸入，因此有較高的視野，再藉由在經過合成孔徑以及影像重建過程後，藉由影像之間的相位差異來重建出原本樣本的影像，在這一個過程中會影像會在空間域與頻率域之間不斷的進行傅立葉轉換，可以在原本的系統數值孔徑再加上一個合成數值孔徑來提升FP系統來達到提升解析度的目的。 傅立葉疊層顯微術目前最需要改進的層面在於其大量的影像輸入，因此在整個傅立葉疊層顯微術的過程中，在影像拍攝與重建的同時，會因為大量的影像輸入，因此便需要耗費大量時間，而其中拍攝的過程所耗費時間又更長，因此將整個傅立葉疊層顯微術系統的拍攝快速自動化，能夠減少因為手動所耗費大量的時間。而此快速的系統化配置也能應用於其他的顯微系統當中，例如定量微分相位差顯微術，此種顯微術亦需要多張的影像輸入。在硬體系統上，將LED矩陣替換為薄膜電晶體液晶顯示器，可以自由調整光源的位置，我們設計了有別於傳統的拍攝光源位置，對於光源的使用可以提高效率並且有更高的自由度。隨後，探討了 FP 顯微鏡系統的設計和自動化，利用LabVIEW 面整合軟體達到系統自動化，應於生物醫學影像分析上，可以使用於透明的薄層細胞，並且在加快效率並且自動化的同時能夠有利於長時間的細胞觀察。 為了進一步提高傅立葉疊層顯微術的分辨率和圖像質量，應用深度學習技術，利用是殘差卷積神經網絡（RCNNs）。開發的 RCNNs-FP 模型在圖像重建和分析方面取得了顯著進展。此外，本研究將超穎透鏡整合到 FP 顯微鏡中，稱為 Meta-FP 顯微鏡。超表面能夠在納米尺度上操縱光波，從而增強成像能力。通過實驗評估了 Meta-FP 顯微鏡的性能，展示了其在高分辨率成像和定量相位成像方面的潛力。本論文通過探索創新方法來提高 FP 顯微鏡的性能和擴展其應用，為 FP 顯微鏡的發展做出了貢獻。本研究中的發現有可能影響包括生物醫學研究、材料科學和光學工程在內的各種領域。

This thesis presents a comprehensive investigation into the design, automation, and application of Fourier Ptychography (FP) microscopy. The work focuses on enhancing the performance and capabilities of FP microscopy through the integration of advanced technologies, including deep learning and metasurfaces. FP microscopy is a label-free imaging technique of quantitative phase imaging. It allows observing cells or tissues without inducing cell toxicity due to staining. Traditional microscopes, when used with lower magnification objectives to achieve a wider field of view, often sacrifice image resolution. Conversely, higher magnification objectives yield better resolution but offer a narrower field of view. In FP microscopy these issues can be overcome. The goal of FP microscopy is to use lower magnification objectives with lower numerical apertures to maintain a wider field of view while ensuring high-resolution images. In this technique, multiple images are captured from different illumination angles under a low numerical aperture system architecture to achieve a broad field of view. By employing synthetic aperture and image reconstruction techniques, the phase differences between images are utilized to recreate the image of the original sample. This involves iterative Fourier transformations between spatial and frequency domains, potentially enhancing resolution by adding a synthetic numerical aperture to the original system. One major area for improvement in FP microscopy lies in managing the substantial amount of image inputs. The process of capturing and reconstructing images is time-consuming, particularly the image capture phase. Automating the entire system's capture process can significantly reduce manual time investment. This rapid system configuration is also applied to other microscopy systems, such as quantitative differential phase contrast microscopy, requiring multiple image inputs. On the hardware front, replacing LED matrices with thin-film transistor liquid crystal displays allows for flexible adjustment of light source positions. By proposing a novel approach to light source utilization, diverging from traditional methods, to enhance efficiency and provide greater flexibility. In biomedical image analysis, this could be applied to transparent thin-layered cells, facilitating efficient automation and prolonged cell observation. LabVIEW automation system with GUI is designed, developed, and implemented for FP microscopy. Subsequently, the design and automation of FP microscopy systems are explored, emphasizing the importance of precise control and efficient data processing. Deep learning techniques, specifically Residual Convolutional Neural Networks (RCNNs), are applied to further enhance the resolution and image fidelity of FP microscopy. The developed RCNNs-FP model demonstrates significant advancements in image reconstruction and analysis. Additionally, the integration of metasurfaces into FP microscopy, referred to as Meta-FP microscopy, is investigated. Metasurfaces enable the manipulation of light waves at the nanoscale, leading to enhanced imaging capabilities. The performance of Meta-FP microscopy is evaluated through experiments, showcasing its potential for high-resolution imaging and quantitative phase imaging. In conclusion, this thesis contributes to the advancement of FP microscopy by exploring innovative approaches to enhance its performance and expand its applications. The findings presented in this work have the potential to impact various fields, including biomedical research, materials science, and optical engineering.