利用紋理特徵於電腦斷層影像預測急性缺血性腦中風中大腦動脈出血轉化之輔助診斷系統

Texture Analysis for Aided Diagnosis of Hemorrhage Transformation of Acute Middle Ischemic Stroke in CT Images

胡學成(Shiau-Cheng Hu)

指導教授 ： 蘇振隆

中原大學/工學院/生物醫學工程研究所/碩士(2011年)

https://doi.org/10.6840/cycu201101092

摘要

急性缺血性腦中風注射血栓溶解劑後，造成出血性轉化併發症之發生，這在臨床上將對病患造成了重大的危害。核磁共振造影(MRI)，在臨床上具備高效能的偵測能力，無法及時診斷。故利用第一線工具電腦斷層攝影(CT)及應用影像處理技術，以預測出血轉化之病患，輔助醫師快速獲得正確的診斷資訊，做出正確的處置減少出血的危險。
本研究透過影像處理技術與紋理量化分析，建構一套電腦輔助診斷系統(CAD)。將中風序列之切片CT影像輸入至系統，經影像前處理後得到腦組織影像，並用中大腦區域分割方法獲得左、右中大腦半球影像。分別利用鄰近灰階相關矩陣(NGLDM)、灰階共生矩陣(GLCM)及結合兩者來擷取紋理特徵。
由醫師診斷結果之10位病患序列切片CT影像，依據相同部位的影像分類，可獲得近端、中端、遠端之中大腦半球影像各20張，進行紋理特徵擷取並用Leave-one-out模式來訓練支持向量機(SVM)的分類。系統評估後發現，NGLDM在近端、中端、遠端之影像分類結果優於其他兩種方法，其Accuracy分別為0.9、1、0.95，Sensitivity分別為1、1、1，Specificity分別為0.8、1、0.9，Kappa值分別為0.8、1、0.9，故中端腦部切片影像之效果為最優。
整體言之，本研究以未施打血栓溶解劑之腦部電腦斷層影像，以紋理量化分析及影像處理應用，發展出一套預測急性缺血性中風出血性轉化之預測系統。本系統將可有效提升CT診斷能力，提供醫師預測出血性轉化的參考依據，對於深受急性缺血性腦中風為害的病患將有重大幫助。

關鍵字

急性缺血性腦中風 ； 出血性轉化 ； 電腦斷層掃描 ； 鄰近灰階相關矩陣(NGLDM) ； 電腦輔助診斷系統(CAD)

並列摘要

In clinical, the hemorrhagic transformation of acute ischemic stroke may occur after the injection of tissue plasminogen activator (t-PA) for acute ischemic stroke patient. Magnetic resonance imaging (MRI) has the best detection ability among imaging modalities to prevent this risk, however, MRI can’t done within short time. Thus, processed computer tomography (CT) images may provide a solution to assist physician to predict hemorrhagic transformation of acute ischemic stroke patients, and make the right disposition to reduce the risk of hemorrhagic transformation.
Image processing and texture quantification analysis methods were used to construct a computer-aided diagnosis system (CAD) in this study. The CT images of head which been diagnosis as stroke by physicians were pre-processed to obtain brain images, and then left and right of cerebral hemispheres image were segmented by applied middle cerebral artery (MCA) regions segmentation method. Finally, three difference methods which include neighboring grey co-occurrence matrix (NGLDM), grey level co-occurrence matrix (GLCM), and the combination of both were used to extract texture features in the CT images. Finally, texture feature extractions and the CAD system has trained by Leave-one-out method of support vector machine (SVM) for image classification.
A series of CT images of 10 patients were served as training dataset. The proximal, middle, and distal of cerebral hemispheres images (totally 20 images for each) were picked according to image classification of the same partial brain. The evaluation among three different methods, we find that the classified results of NGLDM for the proximal, middle, and distal of cerebral hemispheres were performance better than other two methods. In proximal, its accuracy was 0.9, sensitivity was 1, specificity was 0.8, and Kappa values was 0.8; in middle, its accuracy was 1, sensitivity was 1, specificity was 1, and Kappa values was 1; and in distal, its accuracy was 0.95, sensitivity was 1, specificity was 0.9, and Kappa values was 0.9, respectively. Therefore, the best performance for image classification of the CAD system was in middle of cerebral hemispheres.
In conclusion, this study used quantitative analysis of texture features and image processing by no injection of the t-PA of the brain CT images, to develop a set of prediction of hemorrhagic transformation of acute ischemic stroke system. This system will be able to improve the interpretation capabilities in CT images, and provide a significant reference for physicians to predict with hemorrhagic transformation. For the patients with acute ischemic stroke, this CAD system will be useful.

並列關鍵字

computer tomography (CT) ； hemorrhagic transformation ； acute ischemic stroke ； computer-aided diagnosis system (CAD) ； neighboring grey co-occurrence matrix (NGLDM)

參考文獻

[15] 黃綠宣，“結合紋理分析與影像增強技術於電腦斷層影像中輔助診斷急性缺血性腦中風”，中原大學醫學工程研究所碩士論文，民國97年。

[25] 胡祐莊，“以膠囊內視鏡影像的色彩與紋理特徵應用於潰瘍之偵測”，中原大學醫學工程研究所碩士論文，民國99年。

[4] Knight RA, Barker PB, Fagan SC, Li Y, Jacobs MA, and Welch KM, “Prediction of Impending Hemorrhagic Transformation in Ischemic Stroke Using Magnetic Resonance Imaging in Rats” American Heart Association, Vol. 29, pp.144-51, 1998

[7] Andrea Kassner, Fang Liu, Rebecca E. Thornhill, George Tomlinson, and David J. Mikulis, “Prediction of Hemorrhagic Transformation in Acute Ischemic Stroke Using Texture Analysis of Postcontrast T1-Weighted MR Images” Journal of Magnetic Resonance Imaging, Vol. 30(5) , pp. 933–941, 2009.

[9] Sun Changming, “Symmetry detection using gradient information” Pattern Recognition Letters, Vol. 16(9), pp.987-996, September 1995.