折射畸變修正演算法與長距離光學同調斷層掃描術於眼科應用之整合開發

Abstract

全球近視盛行率快速增長，特別是在亞太地區和台灣，近視的診斷與預防成為重要議題。光學同調斷層掃描術(Optical coherence tomography, OCT)是現代眼科診斷的重要技術，具有非接觸式檢測和高解析度影像的優勢。然而，由於角膜表面的高曲率特徵以及眼球內部折射率的非均勻分布，傳統OCT系統受到光學畸變的影響，導致影像無法準確反映眼球內部的真實結構。 在本論文研究中，開發了一種折射畸變修正演算法，用於修正OCT全眼影像中的光學畸變，從而重建眼球內部組織的真實結構。透過逐段修正的方式，依序針對角膜、前房、水晶體、玻璃體與視網膜，分別應用對應的相折射率(Phase index)與群折射率(Group index)進行影像修正。採用一種基於圖論(Graph theory)的層分割演算法，來搜尋並分割眼球結構組織的界面。並運用反向光線追跡(Inverse ray tracing)的概念，進行影像扭轉(Image warping)以修正畸變。演算法同時會計算眼球結構特徵參數，包括角膜曲率、眼角膜中央厚度與光軸位置等。在系統方面，本研究中使用低掃頻速率的高折射率差光柵垂直共振腔面射型雷射(High-contrast grating vertical-cavity surface-emitting laser, HCG-VCSEL)作為掃頻式光學同調斷層掃描術(Swept-source OCT, SS-OCT)的光源。透過額外架設馬赫-曾德爾干涉儀(Mach-Zehnder interferometer)收集校準訊號，並利用雙通道擷取重採樣技術校正干涉頻譜圖線性度，以進一步增加成像深度，實現全眼的OCT成像。 研究驗證的部分，演算法經自製假體的影像測試，證實其修正準確性。實驗結果展示了小鼠全眼影像的修正效果，以及眼球結構參數的量測結果。最後，將演算法整合至實驗室現有以C++程式語言撰寫的MFC操作介面，實現實時影像修正。應用於長距離OCT系統對眼球模型成像，利用演算法輔助調整眼球模型傾角並量測眼軸長。最後量化分析量測數值的精確度。

The global prevalence of myopia is increasing, particularly in the Asia-Pacific region and Taiwan. Accurate diagnosis and prevention of myopia have become critical issues. Optical coherence tomography (OCT), a key diagnostic tool in modern ophthalmology, offers non-contact measurement and high-resolution imaging. However, the cornea's high curvature and the eye's non-uniform refractive index cause optical distortion, hindering accurate imaging of internal ocular structures. This study develops a refractive distortion correction algorithm to address optical distortion in whole-eye OCT imaging, thereby reconstructing ocular structures. By employing a sequential correction approach, the algorithm applies phase indices and group indices for each ocular region. A graph-theory-based layer segmentation algorithm is used to delineate the interfaces of ocular structures, followed by inverse ray tracing to perform image warping. Additionally, the algorithm computes ocular structural parameters such as corneal curvature, central corneal thickness, and optical axis position. A high-contrast grating vertical-cavity surface-emitting laser (HCG-VCSEL) with a low sweep rate was used as the light source for the swept-source OCT (SS-OCT) system. A Mach-Zehnder interferometer was set up to collect calibration signals, and dual-channel acquisition resampling techniques were employed to correct the frequency-domain linearity of the interference spectrum. This configuration enhanced the imaging depth of the OCT system, enabling whole-eye OCT imaging. The algorithm was validated using phantom images, confirming its correction accuracy. The corrected whole-eye images of mice were demonstrated, and ocular structural parameters were measured. The algorithm was integrated into an existing MFC-based interface developed in C++ to achieve real-time image correction. It was applied to image a phantom eye, assisting in alignment adjustment and axial length measurement. Finally, the accuracy of the measurements was quantitatively analyzed.