由心臟電腦斷層掃描自動擷取分析冠狀動脈

Abstract

近年來，由於多切面電腦斷層掃描（Multi-Detector Computed Tomography，簡稱 MDCT）的快速發展，電腦斷層影像變的更精確、解析度更高，而被廣泛的使用。因此 近年來許多學者希望能夠發展出一套自動預診系統，來輔助人工診斷。然而預診要成功 的前提，是要能夠自動化的找出我們要診斷的部分，也就是冠狀動脈。所以我們的論文 就是希望能夠自動化的找出冠狀動脈，提供給後續預診系統來使用。首先我們運用影像 處理的方式，來分割出心臟部位並找出屬於冠狀動脈的部分，再使用距離轉換（distance transform）找出冠狀動脈的中心當作代表點，接著再根據冠狀動脈分布的方向、密度來 判斷代表點之間的連通性，來找出大略的冠狀動脈樹狀結構。接著我們藉由找出主動脈 來求出冠狀動脈的起點，並根據這個起點來執行一系列的圖形演算法（graph theory）來 取出正確的冠狀動脈樹狀結構。

In recent years, due to the rapid development and expanded use of MDCT (multi-detector computed tomography), it becomes much easier than before to obtain more accurate and high-resolution image data of the heart. Many scholars have attempted to develop algorithms for the automatic extraction to find the coronary arteries from the MDCT images to facilitate the screening and diagnosis. This is also the purpose of this thesis. First we use morphological image processing to find the heart region and coronary artery candidate points. Next we use distance transform to find the centers of the coronary artery candidate points as prototypes. We then utilize the distribution and density of candidate points around each prototype to link the prototypes into coronary artery trees. The starting points of the coronary arteries are identified by finding the location of aorta. Finally we use some algorithms of the graph theory to select the correct coronary artery trees.