自動化冠狀動脈分割與註冊方法

Abstract

心血管疾病在現代人中罹瘓的機率很高，如果能夠提早發現，及早預防，才能避免因冠狀動脈阻塞所造成的悲劇。近年來，多切面電腦斷層掃瞄(multi-detector computed tomography，簡稱MDCT)的進步，提供我們一項非侵入式的診斷方法，但是由醫生來判斷一層層的電腦斷層影像，是很耗時費力的工作。全自動化電腦預診系統是我們研究的主要方向，主要分成冠狀動脈擷取以及冠狀動脈分析兩大部份。由於前人的相關演算法多省略了自動化找出冠狀動脈的過程，而以人工標記的方法取代，因此我們的論文改善了這一點，利用影像處理形態學的方法擷取出可能為冠狀動脈的部份，再以明確均值分類法(hard c-means)搭配樹狀結構分析，以達到完全自動化擷取冠狀動脈的目標。另外，同一病患，在攝影心臟電腦斷層掃瞄圖時通常至少會拍攝有顯影劑和無顯影劑兩組資料，不同的資料會因為心臟跳動的關係，冠狀動脈的形狀會有些微的變化，因此我們發展了相同位置的定位的演算法，利用iterative closest point(ICP)將兩組冠狀動脈資料先做基本的比對，再以冠狀動脈縱向剖面影像的比較，修正兩組資料的位移量，完成不同資料相同位置的註冊，希望可以藉由不同資料相同位置的資訊，提高預診的準確度。

Cardiovascular disease is common for people. The earlier we discover the disease, the more chance for us to prevent the tragedy caused by myocardial infarction. Recently, the progress of multi-detector computed tomography (MDCT) provides us a kind of diagnostic non-invasive method for coronary. But it is tiring and time-consuming for a doctor to analyze MDCT images layer by layer. Our research is about developing a computer assisted diagnostic system. The inputs are datasets from MDCT. Once we have the inputs, we can segment the coronary artery by computers without human interaction. After that, we can analyze the coronary artery to extract the positions that are in high probability of myocardial infarction caused by calcifications or blood clots. In previous work, in order to simplify the process, the segmentation of coronary artery is made by human. We proposed an algorithm to extract the coronary artery automatically. Firstly we use mathematical morphology to segment coronary candidates. Secondly, we use hard c-means algorithm and tree extraction to reach the goal of fully automatic segmentation. In addition, two different datasets from the same patient will not be the same because of acquiring time. The coronary artery is not rigid, and the shape is slight different by heart beating. So we also proposed a registration algorithm using iterative closest point (ICP).This algorithm can increase the accuracy of diagnosis by comparing the same position in two different sources.