含顯影劑與不含顯影劑三維電腦斷層影像之 冠狀動脈斑塊比較分析：血管重建與斑塊對位

Abstract

近年來，心血管疾病在國人十大死因中高居第二名，而冠狀動脈疾病又為心血管疾病之首，故冠狀動脈疾病的診斷及治療成為重要的課題。而多切片電腦斷層掃描(MSCT)是目前非侵入性的冠狀動脈疾病影像學檢查中，最為重要且有效的影像方法之一。然而，目前市面上雖有商用分析軟體處理冠狀動脈血管樹(Vessel tree)以及斑塊的定量分析與危險性分析，但提供的資訊相對有限。本研究著重於發展一套自動化的工具降低使用者操作過程，強化整體冠狀動脈分支的準確率，並分析血管中斑塊的資訊，提供完善的資訊幫助醫師診斷及治療。 本篇方法主要分成三大步驟:第一步驟為血管分割，是利用加入機率值概念之區域成長法(region growing)對每一切面進行主動脈的成長，最後得到初步的冠狀動脈走向當做血管的初始輪廓(initial contour)，再套用本論文提供的小波轉換以及Neutrosophic set方法偵測細部的血管分支，得到最終多切片電腦斷層掃描冠狀動脈影像之冠狀動脈輪廓，以便進行斑塊分割與斑塊危險性之評估與預測。 第二步驟為鈣化斑塊分割與對位，由於顯影劑會造成鈣化斑塊在影像上的誤判，所以醫師常以不含顯影劑影像之鈣化斑塊當做判斷的依據。故本步驟之目的是希望藉由對位的方式，找出含顯影劑影像之鈣化斑塊在不含顯影劑影像中相對應的鈣化斑塊，進而得知正確的鈣化斑塊位置及大小，以便進行後續之危險性評估。在分割含顯影劑影像之鈣化斑塊方面，由於上述之問題，故使用門檻值(threshold)大略分割含顯影劑影像中之鈣化斑塊。另一方面，因為不含顯影劑影像之對比度高，所以利用門檻值方法即可輕易地分割出鈣化斑塊，最後計算兩組影像之斑塊的質心，並求解斑塊與斑塊間的距離比，利用此資訊進行對位。 第三步驟為斑塊定量分析，此步驟依照斑塊種類可分成兩種評估方式。首先利用Agatston score判讀鈣化斑塊之危險程度。另一方面，為了計算非鈣化斑塊所造成的狹窄程度，首先必須求得血管橫切面(Cross section)，再從中分割出非鈣化斑塊，最後在用臨床上常用的方式，也就是以非鈣化斑塊最長直徑與血管直徑之比值當作依據。 本研究之影像皆由台大醫院影像醫學部提供之多切面胸腔電腦斷層影像(MSCT)，並採用三十組電腦斷層掃描影像來進行實驗，結果顯示本研究可以重建出各種不同管徑的冠狀動脈分支，且對位結果也相當準確，而斑塊定量分析結果也與醫生之結果符合。

Atherosclerosis, the leading cause of heart disease, has been highly ranked as the second major cause of death in Taiwan for the past decades. The diagnosis and treatment of coronary artery disease have become the most important issue. Multi-slice computed tomography (MSCT) is the conventional strategy for the diagnosis of atherosclerosis. However, commercial software provides very limited information for quantitative analysis of both calcified and non-calcified plaques. This research focuses on developing automated system, improving the precision of coronary artery branch detection, and providing comprehensive information of plaques. The investigation consists of the following steps. First of all, region growing and discrete wavelet transform are applied to adequately segment coronary arteries. The modified region growing is used to initialize the segmentation of coronary arteries. According to this initial segmentation, discrete wavelet transform with neutrosophic set detects tiny branches of coronary arteries. The outcome of this procedure will contribute to further analysis of plaques. Segmentation and registration of calcified plaques are the second step in our investigation. For the disturbance of contrast agent by the calcified plaques, physicians are used to observe the calcified plaques in computed tomography without contrast agent. Therefore, the registration scheme is utilized to find the corresponding calcified plaques between two types of computed tomography. It will help obtain the characteristics of calcified plaques. Different thresholds are set to detect the plaques in images with and without contrast agent. The distance between two centroids of neighboring plaques are analyzed during the registration. The third step is the plaque quantification and plaques with different ingredients are individually quantified. Agatston score is utilized to analyze calcified plaques. On the other hand, cross sections are evaluated to segment con-calcified plaques and investigate the stenosis caused by non-calcified plaques. Accordingly, we compute the ratio of plaque diameters to vessel diameter and stenosis area to area of normal lumen. Our raw MSCT images were collected by National Taiwan University Hospital Department of Medical Imaging. We developed a fast automated algorithm for the segmentation of coronary arteries, registration of CP and quantification of NCP and CP from MSCT. The experimental results demonstrate that the proposed approach has good performance.