https://scholars.lib.ntu.edu.tw/handle/123456789/83074
標題: | 嵌入先驗之三維區域單元競爭演算法:多切面電腦斷層掃描冠狀動脈粥樣硬化影像分割與分析 Prior-Embedded Three-Dimensional Cell Competition Algorithm for Atherosclerotic Coronary Segmentation and Analysis in Multislice Computed Tomography |
作者: | 李健豪 Lee, Chien-Hao |
關鍵字: | 影像分割;血管追蹤;分水嶺轉換;區域單元競爭;冠心病;多切面電腦斷層掃描;Image Segmentation;Vessel Tracking;Watershed Transformation;Cell Competition;Coronary Artery Disease;Multislice Computed Tomography | 公開日期: | 2008 | 摘要: | 無論以學術或醫療為觀點,許多研究都顯示多切面電腦斷層掃描 (MSCT) 對於診斷心血管相關疾病而言具有偉大的貢獻。然而,過往在此類疾病的診斷上,大都是根據醫師們多年的臨床經驗,以視覺直接判斷來對MSCT影像進行大量的手動圈選工作,由此定義血管邊界、斑塊種類甚至鈣化的體積進而研判病因,這樣一系列流程不僅耗時費力,醫師們所判斷的結果也經常因人而異。因此,本研究的目的在於提出一套自動化的電腦輔助診斷系統,用以自動偵測粥樣硬化之冠狀動脈血管區段,期望能改善由人為操作變因所造成的不客觀因素。 本研究所提出的方法是一個複合型的影像分割演算法,稱為嵌入先驗之三維區域單元競爭演算法。首先我們採取模型化血管追蹤將影像裡的血管、背景等等特徵擷取出來,再將臨床上感興趣的區域打碎成許多性質相近的區塊,並同時嵌入血管追蹤所得之特徵作為先驗,稱之為「區域單元」,是進行區域競爭的基本元素。競爭機制初始當下,每個「區域單元」正巧唯一決定一個「區域集合」,基於所設計之統計能量函數,這些「區域集合」會反覆地以「區域單元」為單位相互競爭,每回合主動搶奪或被動讓出一個「區域單元」而成為較大或較小之「區域集合」,一旦能量函數收斂,所有「區域集合」便會各自成為一塊特徵區域,而我們所欲擷取之目標物則會突顯於一個或多個「區域集合」當中。換句話說,基於所嵌入之不同特徵的先驗,便能自動偵測出涵蓋斑塊的血管內腔,區隔影像的背景部份。同時,我們也設計了一些好用的工具與方便的介面以提供各項有利診斷的量化數據與統計資訊,這將使醫師在臨床上更容易判讀影像的表現,進而提高診斷的準確率。 本研究採用的影像資料來自12位在台大醫院進行MSCT冠狀動脈血管造影的病人,其中一位同時進行了血管內超音波 (IVUS) 的檢查,作為驗證本研究所提出之演算法的方式之ㄧ。而本研究所提出的方法,嵌入先驗之三維區域單元競爭演算法,在這12組MSCT影像當中都能成功地擷取出涵蓋斑塊的粥樣硬化冠狀動脈區段與其分支,並且斑塊部分是與冠狀動脈分離而自成一個區域,正說明了本方法分割的結果相當良好。至於所剩之1組IVUS影像正由台大醫院影像醫學部驗證本演算法的成效中。 Elaborate studies have reported that multislice computed tomography (MSCT) appeared to make tremendous contributions and may be a standard imaging modality for the diagnosis of cardiovascular diseases. However, clinical practice for dealing with MSCT coronary angiography placed major importance on experiences, subjectively extracting coronaries and quantifying atherosclerotic plaques in MSCT datasets, which is an extremely tedious task if done manually and may cause different diagnoses. Therefore, this study aims to propose a computer aided diagnosis system, which enables to detect the atherosclerotic coronary segments and plaques automatically without any operator dependent factors. hybrid approach, called prior-embedded three-dimensional (3D) cell competition algorithm, is proposed in this study. It first decomposes the region-of-interest (ROI) into small homogeneous areas which are called cells, with features extracted by the model-based vessel tracking algorithm in it. Each cell uniquely defines a region at the very beginning. Based on a statistical energy function, all adjacent regions will compete against each other, merge or split simultaneously in a cell-by-cell fashion until a steady state is reached. Meantime, each region defines a prominent component and our object of interest is characterized by at least one of them, so that the plaques and the lumens throughout the coronary can be auto-identified by the prior information. Furthermore, this algorithm is designed with several valuable tools and a friendly user interface. All beneficial information such as statistical analyses and quantitative measurements will be provided to improve diagnostic accuracy.he proposed method was validated by 12 patients who were scheduled for MSCT coronary angiography in National Taiwan University Hospital (NTUH). One of them also underwent intravascular ultrasound (IVUS) as a part of research protocol. Totally 12 MSCT datasets were examined in this study and the promising results were achieved. To make a comparative study and to strictly verify the proposed algorithm, the IVUS datum is in the process of making quantitative measurements in the department of medical imaging, NTUH. |
URI: | http://ntur.lib.ntu.edu.tw//handle/246246/184647 |
顯示於: | 醫學工程學研究所 |
檔案 | 描述 | 大小 | 格式 | |
---|---|---|---|---|
ntu-97-R95548058-1.pdf | 23.53 kB | Adobe PDF | 檢視/開啟 |
在 IR 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。