電腦斷層掃描冠狀動脈影像分割與斑塊體積估計演算法之研究

Abstract

隨著種種醫療儀器成像技術的快速進步，在臨床診斷上，專業的醫護人員可以藉由針對不同組織所呈現的數位醫學影像，達到更準確的初步病情診斷；重要的是，藉由此種非侵入式的診斷方式，除了快速、容易獲得，更可免除病人在非必要的情形，就直接進行侵入式切片檢查、導管手術等具有一定風險且伴隨相當壓力的診斷方式。 本研究所使用的是縱向連續切面的電腦斷層掃瞄影像，目的在於提出一套完整的自動分割演算法，適用於即時的三維醫學影像平台上，快速且客觀地將空間中感興趣的冠狀動脈分割出來，並且對於其中具關鍵性的斑塊、血管壁體積，依照影像特徵客觀地計算與分析，作為臨床研究統計的輔助工具。 我們所提出的方法，主要可分為兩步驟：冠狀動脈血管影像分割與斑塊體積估計演算法。前半段將會先對影像去除雜訊，再配合所提出的連通路徑演算法與主成分分析演算法將空間中感興趣的血管進行座標轉換的直立動作，接著使用分水嶺演算法產生初始的區域單元，從使用者給定的血管起始層根據一系列所提出來的血管成長準則，將三維空間中的血管自動化分割出來。 後者根據已分割出來的冠狀動脈，如果管壁中存在硬化的鈣化組織，則先以形變模型的概念將其分割出來，其餘部分或是只有初期軟性斑塊的冠狀動脈，則使用統計學上的模糊分群演算法，將斑塊、血管壁、與血液部分分類。 在過去的臨床診斷上，往往在計算斑塊、血管壁體積的部分，都需要大量且費時的人工圈選，這不僅需要多年的臨床經驗累積，並且常常缺乏客觀且具說服力的標準，故在判讀可能的病因時，較不易獲得廣泛認同。而本研究提出的演算法，可以最少的人為操作，達到自動且客觀的目的。

With the rapid improvement of medical imaging technologies, medical doctors are able to make the first diagnosis more accurately based on the different tissue information revealed by various medical imaging modalities. More importantly, its noninvasiveness not only promises a fast and easy diagnosis, but also greatly reduced the needs of invasive surgery, e.g., biopsy, cardiac catheterization, etc.. This study aims to develop an automatic segmentation algorithm for segmentation of the coronary artery of interest in a volume of CT images, which is composed of a set of 2D axial slices. The ultimate goal is to characterize the plaques and calcifications in the coronary artery using such indicators as volumes, intensity distribution, and so on, to assist diagnosis of coronary diseases. The proposed method is comprised of two major points. One is an automatic image segmentation algorithm for coronary artery based on the series CT images. The other is volume estimation for the plaque and calcification in a vessel segment. The main steps in the vessel segmentation algorithm include de-noising, end-marker connection, using principal component analysis to re-align the vessel along the vertical direction, using watershed transformation to generate cells, and segmenting the vessel using cell-based region growing. Based on the segmented coronary artery, the calcification is first extracted by a 3D deformable model if it exists. Following that, the plaques and vessel walls are identified by using a fuzzy C-mean clustering algorithm. Conventionally, it is a tedious task to estimate the volumes of the plaques and vessel walls. It not only takes an enormous amount of time to delineate the objects of interest manually, but also requires tremendous clinical experience. Moreover, it often lacks of objective end convincing standard so that the diagnosis may vary with medical doctors. Although in a preliminary development stage, the promising results achieved in this study shed the light of an automatic and objective analysis of coronary diseases using a volume of CT images.