高解析度Mirau全域式光學同調斷層掃描儀於組織切片之腫瘤辨識研究

Discrimination of Tumor Boundary Using Ultrahigh-Resolution Mirau-Based Full-Field Optical Coherent Tomography

簡孟庭(Meng-Ting Chien)

指導教授 ： 黃升龍

國立臺灣大學/電機資訊學院/光電工程學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.01329

摘要

臨床上在去除腫瘤的手術中，醫生需要透過組織切片的方法才能得到某組織位置的腫瘤資訊，但過去所使用的組織切片方式需要歷經複雜的製作過程才能於顯微鏡底下觀察病灶[1]，光學同調斷層掃描儀 (Optical Coherent Tomography; OCT) 擁有非侵入式、無需標定與高解析度的優勢，並可以於短時間內得到組織的三維影像，本研究即是要用實驗室所開發的OCT系統，以切下之檢體來做光學組織切片的前期研究，並用演算法進行腫瘤邊界的辨識研究。
文中所使用的OCT系統，其縱向與橫向的空間解析度分別為0.9 μm與0.8 μm，為了要突破OCT系統視野小的限制，運用了影像拼接技術，從原本291 μm × 219 μm視野大小的影像，拓展到可達1 cm × 1 cm的OCT大面積影像，此技術的建立讓本實驗成功的進入臨床組織切片的研究領域。
以OCT的量測優勢，先是掃描從活體切下來的新鮮樣本來得到組織的大面積影像，分別得到了老鼠鱗狀上皮細胞癌與人類黑色素細胞癌的OCT影像，此部分的結果都呈現出鮮明的組織特性，也嘗試得到人類淋巴結的OCT影像，希望由此影像可以評估出癌症轉移的特徵。為了讓OCT影像在不同組織的形貌上可以和HE染色影像做相互印證，嘗試使用了不同的樣本製備方式，最後選擇用組織包埋蠟塊之未染色切片樣本來做大量的影像比對。
而為了提高OCT影像判讀的準確性，本論文引入電腦輔助診斷的概念，藉由HE染色影像，對應出OCT影像內的組織型態，用這些組織區域當作是已知組織類別的訓練樣本，接著同時提取訓練樣本與未知組織類別的測試樣本之特徵參數，用線性與二次判別方法來判定未知組織型態的類別，以達到輔助診斷的目的。
本研究成功撰寫出判別的演算法與程式，分別比較了不同條件下的判讀結果，例如使用不同的訓練樣本與測試樣本，選擇使用線性或二次判別方法，改變空間解析單位或改變提取的特徵參數，判定兩種或三種組織型態，都會有不同的鑑別效果，經過綜合的討論，最後在三種組織型態鑑別中使用線性判別方法，並且解析單位為183 μm時鑑別效果最好，其靈敏度與特異度可高達0.9998與1。

關鍵字

光學同調斷層掃描術 ； 影像拼接 ； 光學組織切片 ； 電腦輔助診斷 ； 線性判別分析 ； 二次判別分析

並列摘要

When doing the cancer surgery to remove the tumor, the only way for physician to get the lesion information is biopsy. But in order to obtain the information from the sample by microscope, the sample preparation is very complicated and time consuming. The advantage of optical coherent tomography (OCT) is that it can get the ultrahigh resolution three-dimensional image in a short time through non-invasive and label-free scanning. In this thesis, we used the OCT system developed by our laboratory to do the research. We used the tissue sample cut from the body to do the previous study of optical biopsy and used the discriminant algorithm to do the research of identification of tumor boundary.
The spatial resolutions of the OCT system we use in axial and lateral directions are 0.9 μm and 0.8 μm respectively. In order to overcome the limitation of smaller field of view in OCT system, we use the technique of image stitching to get the large area image. Now, we can get the size of OCT image from 291 μm × 219 μm up to 1 cm × 1 cm. This size is very close to the clinical application and this achievement makes us successfully enter to the research of biopsy.
By the advantage of our OCT system, we first got the OCT large area tissue images by scanning the fresh samples which were cut from the living body, including the mouse squamous cell carcinoma and human melanoma. The results in this part all have the distinctive organizational characteristics. We also got the OCT image from the human lymph node tissue, trying to assess the characteristics of cancer metastasis.
In order to confirm the credibility of different tissue morphology in OCT images, we have to have the corresponding HE stain images to compare. We tried a lot of sample preparation ways to approach this intent. Finally we chose the unstained paraffin-embedded tissue sections to be the large number of corresponding specimens.
For the purpose to enhance the accuracy of interpretation with OCT images, this experiment introduced the concept of computer-aided diagnosis. By corresponding with HE stain images, we obtained the training set from the known tissue type OCT images and used the unknown tissue type OCT images as the test set. Then we extracted the characteristic parameters from both of them and used the linear or quadratic discriminant analysis to identify the tissue type of unknown tissue type region.
In this thesis, we successfully developed the discriminant algorithm and program, and also compared the discriminant results under different conditions, like choosing the method of linear or quadratic discriminant analysis, changing the spatial resolution unit or the extracted parameters, or discriminating two or three tissue types. There is the best result in the condition of using the method of linear discriminant analysis to discriminate three tissue types with the spatial resolution unit of 183 μm, and the sensitivity and specificity are up to 0.9998 and 1 respectively.

並列關鍵字

optical coherent tomography ； image stitching ； optical biopsy ； computer-aided diagnosis ； linear discriminant analysis ； quadratic discriminant analysis

參考文獻

[12] 許婉儀, " Mirau 全域式光學同調斷層掃描術應用於體外黑色素細胞及黑色素癌細胞之分析與判別," 國立臺灣大學光電工程學研究所碩士論文, (2014).

[20] 林耀聖, "全域式光學同調斷層掃描術之研究," 國立臺灣大學光電工程學研究所碩士論文, (2014).

[3] Fujimoto, J. G., Pitris, C., Boppart, S. A., et al, (2000). Optical coherence tomography: an emerging technology for biomedical imaging and optical biopsy. Neoplasia, 2(1), 9—25.

[4] Jain, M., Narula, N., Salamoon, B., et al, (2013). Full-field optical coherence tomography for the analysis of fresh unstained human lobectomy specimens. Journal of Pathology Informatics, 4, 26—30.

[5] Iftimia, N., Iyer, A. K., Hammer, D. X., et al, (2012). Fluorescence-guided optical coherence tomography imaging for colon cancer screening: a preliminary mouse study. Biomedical Optics Express, 3(1), 178—191.