折射畸變修正與多視向影像配準於光學同調斷層掃描術之演算法開發

Abstract

牙科診斷中，醫師常依賴各種影像技術協助病變判斷與治療規劃。隨著臨床數位化的發展，光學同調斷層掃描術(Optical coherence tomography, OCT)因具備高速、非接觸、非破壞等特性，逐漸被應用於牙齒組織的高解析度三維成像。相較於電腦斷層掃描(Computer tomography, CT)，OCT不僅無輻射風險，亦能觀測CT難以辨識的牙釉質早期脫礦病變。然而，由於牙齒組織的高散射性質，易導致病變下方影像資訊缺失，產生陰影效應(Shadowing effect)；同時，未考慮光學之折射率差異也可能造成光路偏移與幾何畸變，影響成像精度與診斷判讀。 本研究聚焦於整合一套折射畸變修正暨多視向影像配準之自動化影像處理流程，以提升OCT成像品質與臨床應用可靠性。系統部分，以實驗室自行建構之掃頻式光學同調斷層掃描術系統(Swept-source OCT, SS-OCT)搭配光學時脈頻率倍增模組(Optical clock frequency doubling circuit module, OCFD)與雙軸旋轉平台，實現多視像牙齒影像擷取。演算法設計上，首先透過影像梯度提取牙齒表面輪廓，並進行光路反向追蹤與幾何補償，修正因折射率差異造成的深度扭曲與結構誤差；接著，根據輪廓建立三維表面模型之點雲，作為多視角影像配準基準，採用最近點迭代法(Iterative closest point, ICP)對齊各視角影像，以消除視角遮蔽造成的資訊缺失，並強化結構連續性與幾何一致性。 最終透過自製瓊脂糖凝膠樣本與牙齒切片進行比對，確認修正後影像的結構準確性與厚度一致性。實驗結果顯示，本方法能顯著提升影像品質與診斷可靠度，展現結合折射率修正與多視向OCT影像的整合方法於高精度口內影像應用的潛力。

In dental diagnostics, clinicians often rely on various imaging techniques to evaluate lesions and develop treatment plans. With the advancement of clinical digitization, Optical Coherence Tomography (OCT) has gained prominence for its high-resolution, non-contact, and non-destructive three-dimensional imaging capabilities of dental tissues. Compared to Computed Tomography (CT), OCT offers the advantages of being radiation-free and capable of detecting early enamel demineralization that is often indiscernible on CT. However, the highly scattering nature of dental tissues frequently leads to information loss beneath lesions, manifesting as shadowing effects. Additionally, neglecting refractive index differences can result in optical path deviations and geometric distortions, thereby compromising imaging accuracy and diagnostic interpretation. This study proposes an integrated and automated image processing framework that combines refractive distortion correction with multi-view image fusion to enhance OCT image quality and clinical reliability. On the hardware side, we employ a self-developed swept-source OCT (SS-OCT) system equipped with an optical clock frequency doubling circuit module (OCFD) and a dual-axis rotational stage to acquire multi-angle dental images. The algorithmic approach begins with gradient-based tooth surface boundary extraction, followed by reverse ray tracing and geometric compensation to correct depth distortions and structural inaccuracies resulting from refractive index mismatches. A three-dimensional surface point cloud model is then constructed from the detected contours, serving as the reference for multi-view image registration using the Iterative Closest Point (ICP) algorithm. This process mitigates data loss caused by angular occlusion and enhances both structural continuity and geometric coherence. Validation experiments using custom agarose gel phantoms and dental slices demonstrate the structural accuracy and thickness consistency of the corrected images. The results confirm that the proposed method significantly improves image quality and diagnostic reliability, underscoring its potential for high-precision intraoral imaging through the integration of refractive correction and multi-view OCT fusion.