- 學位論文
電腦斷層造影於篩檢冠狀動脈血管粥樣硬化的應用
Application of MDCT to identify Atherosclerosis in Coronary Arteries
摘要
摘要： 動脈粥狀硬化(atherosclerosis)及其併發症，近年來已成為台灣地區常見的疾病；因而造成急性心肌梗塞及缺血性腦中風更是其中重要的死亡原因。因此，醫界除了應該瞭解動脈硬化形成的原因外，如何觀察動脈硬化血管壁特徵變化才能達到早期治療早期診斷的效果。本研究目的是想利用組織間(鈣化、纖維化或脂肪)不同的衰減係數，設計一個最佳化的矩陣及閾值將冠狀動脈電腦斷層影像做斑塊型態分類。 研究中利用雙能量電腦斷層(DSCT)產生良好的空間解析度與時間解析度之結構性影像來評估急性冠狀動脈疾病。本研究排除遠端冠狀動脈及其分支有狹窄的樣本，避免因為遠端冠狀動脈及其分支血管直徑過小而造成程式錯誤判斷的結果。從PACS上取得影像後，運用四種不同的矩陣大小及四種不同的閾值區間，以減少其他因素所而造成組織影像在視覺上的差異。利用顯影劑在主動脈底部(Aortic root)的濃度做參考點，與冠狀動脈可能狹窄處的血管濃度值(HU)相比，用來分類該病患的冠狀動脈的斑塊種類。利用不同閾值之運算結果與放射診斷科醫師的報告所得之準確率、敏感度與專一性及其ROC曲線來進行的系統評估。參數訓練上先取樣23位近端或中段有冠狀動脈狹窄的病患，計有33個不同型態的斑塊，再取樣21個無冠狀動脈狹窄的影像作為對照組；最後再用12個隨機取樣的樣本作系統之測試。 由研究結果發現當使用3X3矩陣，閾值為1.15~0.95可得到最大的AUC。當設定比值大於1.15的ROI區域定義為混合型或鈣化的斑塊型態，比值介於1.15~0.95的ROI區域定義為無明顯斑塊，比值小於0.95的ROI區域定義為脂肪或纖維化的斑塊型態進行訓練時，系統之準確率、靈敏度、專一性分別為90.7%、90.9%、90.5%；測試時則分別為91.6%、87.5%、100%。 結果顯示，此一方法可成功將斑塊型態分類；但是此一方法仍無法提供斑塊的活性(發炎反應)資訊，尙不易找出不穩定脆弱的斑塊加以預防。未來如能把正子造影與電腦斷層結合，提供結構性與功能性的資訊，再結合影像處理技術，應可得到更好的診斷靈敏度、準確率之結果。如此將可造福更多急性冠狀動脈疾病的病患，對於冠狀動脈疾病的預防也有助益。
並列摘要
Abstract: Atherosclerosis and its complications in recent years have been one of the most common diseases in Taiwan. Atherosclerosis can cause acute myocardial infarction (AMI) and ischemic stroke, which are among the most common causes of death. Thus, it is important for physicians to understand the causes and the methodology for observing the changes of characteristics of arterial wall, to achieve early diagnosis and early treatment of these diseases. The purpose of this study is plaque classification on coronary artery computed tomography (CTA) images, by using different attenuation coefficients of different tissues (calcification, fibrotic, and fat tissues), and designing optimal matrix sizes and optimal thresholds. This study utilizes dual-source computed tomography (DSCT) with the advantages of better spatial and temporal resolution of structural images, to evaluate acute coronary artery diseases. The study excludes the image samples of stenosis at distal coronary arteries and their branches, to avoid the possibility of error resulting from too small caliber or arteries. We acquire the images from picture archiving and communication system (PACS) and process the images with four different sizes of matrix and four sets of different thresholds, by which to reduce other possible errors due to tissue characteristics on the images. Using the concentration of contrast medium at aortic root as reference points, and comparing with the suspected stenotic coronary artery, we calculate the ratios of concentration by Hounsfield Units (HU), to classify patients’ coronary artery plaques. The image processing data with different thresholds were compared with the reports interpreted by radiologists to evaluate the accuracy, sensitivity, and specificity. We utilized receiver operating characteristic (ROC) curve statistics for systemic evaluation as well. We used 23 patients with stenosis at proximal or mid coronary arteries as training data, which had total 33 different plaques, and 21 patients without coronary artery stenosis were used as control group. Finally, 12 random samples were used as test data set for evaluated this system. The results showed that by using 3x3 matrix and the thresholds between 1.15 ~ 0.95, maximal area under curve (AUC) in ROC test can be achieved. The ratios of region of interest (ROI) greater than 1.15, between 1.15 and 0.95, and less than 0.95 were defined as mixed or calcified plaques, no obvious plaque, and fat or fibrotic plaques, respectively. Using the definitions for training process, the systemic accuracy, sensitivity, and specificity are 90.7%, 90.9%, and 90.5%, respectively. For the testing process, the systemic accuracy, sensitivity, and specificity are 91.6%, 87.5%, and 100%, respectively. The study elucidates that the methodology can classify coronary artery plaques successfully. But the methodology can not yet provides enough information about plaque activity, such as inflammatory activity, to identify the unstable fragile plaques for prevention. In the future, we plan to integrate positron emission tomography (PET) and DSCT to provide both structural and functional information, combining with image processing techniques to achieve better diagnostic sensitivity, specificity, and accuracy. We hope that the further study can be beneficial to more patients with coronary artery diseases and to facilitate the prevention of coronary artery diseases.