https://scholars.lib.ntu.edu.tw/handle/123456789/82987
標題: | 以區域單元為基礎之二維/三維乳房超音波影像圖形切割演算法 Cell-Based Graph Cut for Segmentation of 2D/3D Sonographic Breast Images |
作者: | 江昕鴻 Chiang, Hsin-Hong |
關鍵字: | 超音波影像;影像分割;區域單元競爭演算法;圖形切割演算法;以區域為基底之圖形理論;Ultrasound Image;Image Segmentation;Cell Competition;Graph Cut;Region Based Graph Theory | 公開日期: | 2009 | 摘要: | 在乳房超音波影像上,腫瘤的圈選不但費時耗力並且需要具有相當經驗之醫師或專業人員來進行,三維乳房超音波影像上進行腫瘤的圈選難度更高程序更繁複。因此藉由電腦輔助系統進行乳房超音波影像的腫瘤圈選不僅可以簡化醫師圈選的工作爭取寶貴的治療時間,更可以藉由此系統所提供之腫瘤邊界資訊使診斷更準確。 以圖形理論進行醫學影像之分割已被廣泛的發展應用,在醫學影像上更普遍應用在各種影像格式上如超音波影像,X光影像及核磁共正影像等。但其龐大的計算空間及冗長的計算時間在實務上卻會衍生出許多問題。使用區域作為圖形之節點為解決此問題的方法之一。因此本研究希望藉由在超音波影像中可以產生良好區域結構的區域單元競爭演算法所作為產生區域的方法。將所產生的區域單元視為節點使用Graph Cut對圖形進行劃分。一個良好的以區域為節點的圖形劃分方法除了需要具有良好區域結構及劃分方法外,相似權重函數也扮演著核心要角。因此本研究設計一相似權重函數可藉由預估腫瘤的統計資訊來避免發生超音波影像上灰階值變化平緩的弱邊界及Graph Cut劃分出較獨立點的問題。以區域單元競爭演算法及Graph Cut的搭配作為以區域為節點的圖形劃分方法除了可以節省計算時間及空間外在三維影像上更可以直接應用,因此本文中展示了在三維影像上的應用及部分研究成果,此外也使用二維影像評斷本研究方法之優劣並展現幾個二維影像的範例。 最後本研究使用之影像係由台北榮民總醫院所提供的160張二維乳房超音波影像其中包括60張惡性腫瘤及100張的良性腫瘤,評斷本研究之方法所圈選之邊界與手繪邊界的相似程度。在三維乳房超音波影像上則使用台大醫院所提供本之影像進行影像分割研究,並且展示分割結果。 In sonographic breast lesions image, circling the lesion part is the complicated and time consuming work, moreover it should be done by experienced doctors or experts. In three dimensional sonographic breast lesions image, circling the lesion part becomes more complicated and more procedure. Therefore, circling the lesion part by computer aided diagnosis(CAD) is not only simplified the circling work and saving the doctor’s precious time, but also providing more medical information by computer.egmenting the medical image by graph theory has been used and developed in couple of years. It is general used in various kinds of medical image format such that ultrasound image, X-ray, MRI and etc. In implementation, it derived lots of problem cause of enormous computation space and long computation time. Using region based method is one of solving method. Our research use Cell Competition Algorithm as producing region structure, because it has good result in ultrasound image. After producing the regions, we use Graph Cut to divide the region based graph. A good dividing approach in region based graph theory must have a good method in producing the regions and a good approach in dividing the graph, in addition to the similar weight function also plays an important role. Therefore our research designs a good similar weight function which can be used to prevent the weak edge problem and the problem of prefer cutting the isolated node in Graph Cut according the estimating statistical information of tumor. Another advantage in our research is that it can intuitively implement in three dimensional images.ast part of this paper shows the experiment result, compares with the handmade boundary and evaluates our boundary result. The experiment 2D and 3D sonographic breast lesions images are provided by National Taiwan University Hospital and Taipei Veterans General Hospital. |
URI: | http://ntur.lib.ntu.edu.tw//handle/246246/183700 |
顯示於: | 醫學工程學研究所 |
檔案 | 描述 | 大小 | 格式 | |
---|---|---|---|---|
ntu-98-R96548014-1.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。