三維電腦斷層影像冠狀動脈追蹤與重建演算法

Coronary Artery Tracking and Extraction Algorithm in Multi-slice Computed Tomography Image

鍾政宏(Cheng-Hong Chung)

指導教授 ： 陳中明

國立臺灣大學/工學院/醫學工程學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.10356

摘要

根據衛生署統計，心臟疾病為國人死亡率排名第二名。而心臟疾病在美國根據CDC統計排名為死亡率第一，而冠狀動脈疾病又為心臟疾病之首，因此冠狀動脈疾病的診斷與治療變成為維護國人健康最重要的課題。而利用電腦斷層掃描影像偵測冠狀疾病為目前最方便有效的工具，發展快速簡易的診斷工具實為重要。冠狀動脈疾病在電腦輔助診斷過程中第一步最重要的就是針對冠狀動脈進行追蹤取萃取的動作，當描繪出冠狀動脈後在對血管內部進行分割並辨識斑塊的種類及危險性分析。本研究著重於冠狀動脈的追蹤與萃取，並以全自動化為目標降低使用者操作的過程，提高冠狀動脈分支的辨識率。

本研究中主要目的在於發展較為自動化的工具並提升總體冠狀動脈分支的準確率。其演算法大略分為三個步驟，首先由使用者點選主動脈任一區域，利用區域成長法(region growing)對每一切面進行主動脈的成長，並由主動脈的位置與半徑決定心臟區域大略位置。再來採用管狀結構偵測法(Tube-like object detection approaches)中之Hessian matrix進行分析，分析其特徵值與特徵向量(eigenvalue, eigenvector)擷取出大量疑似管狀結構像素。接下來利用主動脈的輪廓以及主動脈經由平滑化的輪廓作exclusive or運算以偵測冠狀動脈的起始切面與起始點，作為區域成長的種子點。最後改良區域成長法讓其可以適應局部區域的變化，並利用不同成長條件來達到取得最佳化的冠狀動脈血管束。

本研究採用八組電腦斷層掃瞄影像來進行實驗，實驗結果與現行醫師利用商用軟體手動描繪出之冠狀動脈血管束輪廓作比較。結果顯示本研究可以追蹤並萃取出各種不同管徑的冠狀動脈分支，對於末梢細微的血管粹取也有很好的效果。

關鍵字

冠狀動脈疾病 ； 冠狀動脈追蹤 ； 冠狀動脈粹取 ； 區域成長法

並列摘要

According to statistics from the Department of Health, heart disease has been the second major cause of death in Taiwan. And heart disease in the United States is ranked first according to CDC mortality statistics. Coronary artery disease is the leading cause of heart disease; the diagnosis and treatment of coronary artery disease to safeguard the health of citizens have become the most important issue. CT imaging is one of the most convenient and effective tools for detecting coronary disease. The simplicity of diagnostic tools is crucial. In the first step of Computer-aided diagnosis of coronary artery disease, the most important step is to track and extract all the coronary arteries branches, find plaques in arteries, identify types of plaque, risk analysis finally. This research focuses on tracking and extraction of coronary arteries, reducing interactions between user and system, and improving the precision of coronary artery branches.
The main contribution of this study is to develop an automated tool and improve the overall precision of coronary artery branches. This algorithm is roughly divided into three steps including clicking any aortic region by the user using region growing for each slice of aortic growth, and setting heart zone by the location and radius of aortic region. By using the Hessian matrix, tube-like object detection approaches and used to extract a large number of suspected tubular structure voxels. Next, making exclusive-or operation to detect coronary artery starting slices and points as the initial seeding points in the next region growing method. At last, adaptive local region growing adapts local variation and use different growth conditions to optimize the coronary arteries
In this study, eight sets of CT scan images are used as testing data. The experimental results are compared with the commercial off-the-shelf software. The result shows that this algorithm detect and extract various branches of coronary arteries, also has very good results in small lumen of coronary arteries.

並列關鍵字

coronary artery disease ； coronary artery tracking ； coronary artery extraction ； hessian matrix ； region growing

參考文獻

[2] C.D.C Leading Causes of Death (Data are for the U.S.)

[5] Nikolaou K, Flohr T, Knez A, et al. Advances in cardiac CT imaging: 64-slice scanner. Int J Cardiovasc Imaging 2004; 20:535-540.

[6] Miller JM, Rochitte CE, Dewey M, et al. Diagnostic performance of coronary angiography by 64-row CT. N Engl J Med 2008; 359:2324-2336

[7] Schmermund A, Nowak B, Voigtlander T. Non-invasive CT coronary angiography: can high diagnostic image quality be achieved with less radiation exposure? Eur Heart J 2008; 29:2955-2956.

[8] Mollet NR, Cademartiri F, de Feyter PJ. Non-invasive Multislice CT Coronary Imaging. Heart 2005;91:401-407.