肺部電腦斷層掃描之肺葉分割研究

Abstract

近年來國人之十大死因之中，肺部疾病高居第五名及第四名，且在國人十大死因之中的第一名---惡性腫瘤之中，肺部是最容易發病的位置之一，故胸腔疾病的診斷及治療成為近年重要的課題之一。多切片電腦斷層掃描(MSCT)是目前非侵入性的胸腔病影像學檢查中，最為重要且有效的影像方法之一，利用此方法，我們建立起電腦自動化胸腔診斷（Computer-aided Diagnosis, CAD）系統，幫助醫師做術前的規劃。在慢性阻塞性肺病(Chronic obstructive pulmonary disease, COPD)當中，肺葉分割的資訊對肺氣腫的評估能提供有效幫助，而在肺部疾病的分類上也是相當重要的分類依據，故本研究將開發一個新且優良的自動化的肺裂分割演算法，並藉由肺裂資訊來達到切割肺葉的結果，幫助CAD系統更加完善。 本篇演算法主要分成兩大步驟:第一步驟為肺葉擷取，演算法是利用海森矩陣(Hessian Matrix)之特徵向量資訊，配合區域成長法(Region Growing)的概念，將具有板狀特徵的結構給擷取出來。藉由成長多個大板塊後，將屬於肺裂的板塊合併、不屬於肺裂之雜訊排除，達成肺裂擷取的成果。第二步驟為肺葉分割，由於肺葉普遍具有肺裂不全(斷裂、破碎)的現象，我們將利用模型似合(Model Fitting)工具將此問題排除，本研究是採用薄板坊樣演算法(TPS)來達成此目的。 本研究之影像皆由台大醫院影像醫學部提供之多切面胸腔電腦斷層影像(MSCT)，並採用三十二組含有不同品質與肺氣腫病例之電腦斷層掃描影像來進行實驗，將結果與正確答案比較過後，本演算法在平均準確率達到89%，此結果顯示本演算法可以在不同影像品質與肺氣腫之影像上成功且準確地分割肺葉邊界，在與商用軟體VIDA比較後也顯示本演算法具有較平滑、正確、且連續的肺葉邊界。

Lung diseases has been highly ranked as forth and fifth major cause of death in Taiwan recently, and in the most lethal disease-nodule, lung is the most easiest position to fall ill, that is, the treatment and diagnosis of pulmonary diseases become one of the most important issues today. Multi-slice computed tomography(MSCT) is one of the most important and efficient methods in the field of noninvasive imaging examination of pulmonary diseases. We can establish Computer-aided Diagnosis system to help doctors planning before surgery through using this method. Lobe-based visual assessment of volumetric CT in COPD may be used to evaluate the severity of emphysema, meanwhile, it also plays an important role in the classification of lung diseases. Therefore, this research is devoted to develop a new, superior and automatic segmentation of fissure algorithm, and make CAD system more perfect. The algorithm can be separated into two steps. First step is extraction of lobe. The extraction algorithm is based on the eigenvector of Hessian Matrix and the concept of Region Growing, so it can extract the structure with platy feature. Then, the size of plate becomes more and bigger, so it is easier to combine the plate which belongs to fissure and eliminate the noise which doesn’t belong to fissure. Finally, extraction of fissure can accomplish. The second step is segmentation of lobe. Because incomplete fissure can be generally find in lobe, we have to get rid of this problem by using Model Fitting Method, and this research chose thin plate spline (TPS) to reach this goal. All the images of this research came from MSCT which were provided by National Taiwan University Hospital Department of Medical Imaging. This research took 32 images of pulmonary computed tomography within different qualities to experiment. Compare to the right answer, the rate of accuracy of this algorithm reach 89% on average. It show that this algorithm can successfully and accurately segmentation of boundary of lobe in the different qualities of images and the images of emphysema. Besides, compare to the business software VIDA, it also show that this algorithm can calculate more smooth, correct and continuous boundary of lobe.