- 學位論文
結合圖像對位演算法與多視角光學同調斷層掃描術應用於牙齒成像開發
Development of Multi-view Optical Coherence Tomography (OCT) and Image Registration Algorithm for Tooth Imaging
摘要
近年來隨著牙科數位化的浪潮，有越來越多的新型口腔內成像技術被提出，用以更精確地判讀口內疾病狀況，而其中光學同調斷層掃描術(Optical Coherence Tomography, OCT)能提供高解析度的軟硬組織斷層影像，且不會像傳統電腦斷層影像(Computed tomography, CT)有著游離輻射的危害風險，因此是極具有潛力的新型口內成像工具。由於OCT是基於收集背向散射光的成像原理，而受限於牙齒硬組織的光學性質，以及入射光主要還是沿著直線傳播，常使成像上具有垂直方向的偽影(Vertical line artifact)存在，造成在有限成像深度之下的資訊缺漏，也限制了OCT在實際臨床使用上的可用性。 而在本篇論文之中，對於上述的問題，我們期望藉由多視向的取像方式對牙齒進行成像，避開可能造成資訊缺漏的因素，並藉由影像拼接的演算法將不同視向的影像進行融合，用以補齊缺漏的影像資訊，並且抑制影像的散斑雜訊(Speckle noise)，改善牙齒OCT影像的成像品質。而為了探索多視向成像對於增進牙齒影像品質的可能性，我們利用實驗室自行開發的掃頻式光學同調斷層掃描(Swept-source OCT, SS-OCT)系統，結合可調整旋轉角度之雙軸測角台，進行不同旋轉角度之下的牙齒樣本取像，用以模擬實際不同視角下取得之多視向影像。而所使用系統其中心波長為1310 nm，光源頻寬為100 nm，掃描頻率為100 kHz，其在空氣下的軸向解析度為15.6 μm，橫向解析度為9.84 μm，偵測到的平均靈敏度為99.1 dB。 另外，對於多視向的牙齒OCT影像，我們開發了一套基於迭代最近點演算法(Iterative closest point, ICP)的多視角牙齒影像自動拼接演算法，藉由計算影像梯度最大值的方式，來提取出牙齒的表面型態點雲，並利用ICP演算法計算出不同點雲檔之間所產生的位移量，將多視向牙齒OCT影像對齊至相同位置，用以達到多視向拼接後的最終影像。為了驗證影像拼接演算法的穩定性及通用性，我們挑選了6顆不同適應症的牙齒樣本進行拼接融合，並觀察其橫截面影像的變化，而關於演算法對於影像散斑雜訊抑制的成效評估，我們使用信噪比(Signal-to-noise ratio, SNR)及對比雜訊比(Contrast-to-noise ratio, CNR)指標，來量化評估本套演算法改善程度的趨勢，最後，並透過與實際牙齒切片影像相互比較，來確認演算法對於牙齒解剖構造的重建效果。 而在未來的研究方向中，我們期望基於本論文的結果，整合至可攜式之微型多視向口內掃描探頭，達到實時性多視向牙齒影像拼接，以助於OCT應用於臨床牙科的實行性。
並列摘要
With the trend of digitization in dentistry in recent years, more and more new intraoral imaging technologies have been proposed to accurately diagnose oral diseases, among which optical coherence tomography (OCT) can provide high-resolution cross-sectional images of soft and hard tissue without the risk of ionizing radiation as computed tomography (CT). Thus, it is a promising new intraoral imaging tool. Owing to OCT is based on the imaging principle of collecting backscattered light, and due to the optical nature of dental hard tissues and the fact that the incident light mainly propagates along the straight line, there is often a vertical line artifact on the image, resulting in information loss under the limited imaging depth, which also limits the usability of OCT in actual clinical use. In this study, we hope to acquire teeth images using a multi-view imaging setup to avoid those factors leading to information loss. Using the image-stitching algorithm to combine images from different perspectives, the fused images could reconstruct the missing information. Moreover, it can suppress the speckle noise present in the image, thus improving the OCT imaging quality. In order to explore the possibility of multi-view imaging to improve teeth image quality, we used the in-house swept-source optical coherence tomography (Swept-source OCT, SS-OCT) system, combined with a dual-axis goniometer stage to adjust rotation angle, to acquire OCT images of teeth samples under different rotation angles to simulate the images obtained under the different viewing angles. The central wavelength of the system used is 1310 nm; the light source bandwidth is 100 nm; and an A-scan rate of 100 kHz. The axial and lateral resolutions under the air of the OCT system were measured to be 15.6 μm and 9.84 μm, respectively, and the average sensitivity is 99.1 dB. In addition, in order to generate the multi-view teeth OCT images, we have also developed a multi-view automatic stitching algorithm based on the iterative closest point (ICP) algorithm. By calculating the maximum value of the image gradient, we can extract the surface shape of the teeth using point cloud format, use the ICP algorithm to calculate the displacement between different point cloud files, and align multi-view teeth OCT images to the same position to achieve multi-view image stitching. As for validating the stability and generality of the image-stitching algorithm, we selected six tooth samples with different indications to conduct and observe the change of its cross-sectional images. As for the evaluation of the effect of the suppression of speckle noise, we use signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) to assess the degree of improvement. Finally, we also confirmed the reconstruction of the tooth anatomy by comparing it with the actual tooth slice images. In the future, we hope to integrate the results of this work into a portable miniature multi-view intraoral scanning probe to achieve real-time multi-view tooth image stitching, to help OCT be practically applied to dental clinics.