使用斷層掃描影像建立全自動的冠狀動脈分割與斑塊定量演算法

Abstract

動脈粥狀硬化所造成的心血管疾病每年都奪走許多寶貴的性命，已連年佔據台灣十大死因之第二位。電腦斷層掃描影像是主要用來診斷此一疾病之影像工具。為了對這個疾病有更深入的了解，我們透過電腦輔助診斷與全自動的影像分割演算法來重建冠狀動脈結構，並探索血管管腔狹窄所衍伸之問題。 透過優化血管中心線萃取技術並提供血管狹窄程度之整合資訊，本研究致力於建立全自動之電腦輔助診斷於心血管疾病之應用。本研究利用電腦斷層影像發展全自動化之升主動脈分割方法來分割主動脈瓣處的「3顆球」結構，並利用K-mean濾波器進行影像前處理，排除不必要之背景資訊，使本研究中的自適應區域成長演算法得以完整分割冠狀動脈血管。組合所有具血管特徵之像素，並加上血管中心線演算法後，便可對血管管腔中的資訊進行分析。 研究中發現「3顆球」結構之確立有助於降低冠狀動脈起點偵測之難度。而整合灰階值與管狀結構資訊之自適應區域成長演算法能在避免分割影像溢出的情況下得到完整之冠狀動脈網路。藉由血管中心線萃取演算法，我們更得以沿著血管方向分析此一血管路徑上是否發生狹窄、狹窄處之組成成分為何，藉已得到具臨床價值之資訊。 在臨床運用上，本研究之成果將有助於更精準的描述心血管疾病以並提供相關疾病進程之資訊。

Atherosclerosis, the leading cause of heart disease, is still ranked as second major cause of death in Taiwan. Multi-slice computed tomography (MSCT) is the conventional strategy for diagnosis. For the assessment of coronary artery disease, an automated computer-aided diagnosis system is essentially needed to clarify both lumen volume and coronary tree. This research aims to develop automated system, improve the extraction of coronary artery centerline, and provide comprehensive information of stenosis. We collected MSCT images and developed an automated algorithm to extract ascending aorta along with coronary artery from serial computed tomography images. K-means clustering is applied for background exclusion, and the structure of ball-like aortic root is particularly segmented. In self-adjusting region growing scheme, the intensity of each voxel and its neighborhood are both calculated during the establishment of lumen volume. Assembling qualified voxels, the coronary tree is revealed and coronary artery centerline extracted. For further analysis, lumen volume is analyzed after vessel skeletonization is performed. It was demonstrated the delineation of ball-like aortic root would simplify the detection of coronary artery ostium. Integration of neighborhood and vesselness information to the self-adapted region growing scheme was approved to prevent the occurrence of leakage and facilitate the inclusion or exclusion of voxels with similar intensity. With centerline extraction, analysis of vessel profile and quantitation of plaque burden were achieved and diagnostic information acquired. For clinical practice, the automated algorithm and computer-aided tracking system will contribute to consistent and effective assessment for coronary artery disease.