電腦輔助PC-ASPECTS評估系統建置

The Development of Computer Aids Evaluation System for PC-ASPECTS

徐以琳(Yi-Lin Hsu)

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(2018年)

https://doi.org/10.6840/cycu201800122

摘要

腦中風是我國第四大死因，其發生率80%是由於腦缺血。後腦循環缺血中風的致死率和殘疾風險高於前腦循環，加上臨床醫師判斷中風程度較為主觀，因此本論文針對缺血性腦中風之後腦循環的部分，開發一套電腦輔助PC-ASPECTS評估系統，協助醫生客觀判斷嚴重程度，及時給予治療。
本研究主要針對缺血性腦中風CT影像為主。系統讀入非增強CT影像首先會以區域成長法取得有效CT影像資訊、中值濾波器用於去除雜訊，之後手動ROI圈選有、無缺血區域以9個紋理參數分析，針對PC-ASPECTS各別腦區進行訓練與測試來篩選紋理參數。篩選後的紋理參數與其支持向量機(SVM)分類計算出的向量相乘加上常數，再依據閥值判別是否缺血，手動圈選PC-ASPECTS 8個區域計算分數。除了以假體測試驗證系統，並使用排除腫瘤、水腫、出血之CT及對應MRI各20組3切面之影像分別進行系統之訓練與測試。系統實用評估由MRI影像為標準、醫師以肉眼判讀之PC-ASPECTS值與本系統之結果比對。
開發之系統具有BMP、JPG影像讀取(舊缺血影像需去除)、手動圈選ROI之缺血辨別與計分等功能。結果發現:以紋理參數及常數運算所得之臨界值可用以判讀是否缺血及不同部位用及之參數及臨界值不同。其中橋腦為以4個紋理參數(相關性、對比度、能量、最大機率值)運算< 0.1345 小腦以4個紋理參數(相關性、能量、熵值、變異數)運算> -0.6220；中腦以8個紋理參數(相關性、對比度、能量、熵值、最大機率值、變異數、群集顯著性、同質度) 運算< -0.6885；丘腦以4個紋理參數(相關性、能量、熵值、變異數) 運算<-0.8282；PCA區域以4個紋理參數(相關性、能量、熵值、最大機率值、變異數) 運算< 0.2735 為缺血區域。單獨以紋理參數辨別缺血準確率為77.5%，但以向量運算判別缺血準確率提升至90%。
系統整體統計PC-ASPECTS區域評估準確度、敏感度、特異性及Kappa值分別為95.6%、85.3%、98.4%、0.86；PC-ASPECTS值準確率為70%； PC-ASPECTS值≧7評估之準確度、敏感度、特異性為95%、80%、100%，系統評估結果皆優於3位臨床醫師。系統單一切面圈選評分秒數為31.52秒較臨床醫師費時但有較佳的準確率。
本研究系統給予醫師客觀及更加準確的病理判斷與評分，但系統在舊的缺血區域上使用較有限制，左、右腦皆缺血或缺血區域淺且小的CT影像會降低判別的準確率，期望未來能夠針對舊缺血區域篩選更佳的參數加設為系統判讀之項目，提升淺且小缺血區域之準確率。

關鍵字

急性缺血性腦中風 ； PC-ASPECTS ； 紋理特徵參數 ； 電腦輔助系統 ； 影像處理

並列摘要

Stroke is one of the mortality factors in Taiwan. Posterior circulation ischemic stroke has higher mortality rate than anterior circulation stroke in brain, and the estimation of stroke level by physician is subjective. In this thesis, a computer-aided detection system was developed for ischemic stroke diagnosis to help physician objectively determine the seriousness level of ischemic and give treatment in limit time.
This study focus on CT image with ischemic stroke and exclude cases with tumor, edema and hemorrhage. This system was developed based on image processing technology. For preprocessing, a bi-level, regional growth methods and medium filter were used to de-noise and obtain effective image information for brain CT image, respectively. And then the normal and abnormal areas are analyzed by 9 texture parameters. Totally 8 data set were used to train and the others 8 data set were used to test this system in PC-ASPECTS 5 regions, respectively. The ischemic region can be obtained by using the selection of texture parameters. Finally, the support the vector machine (SVM) was used as classification tool and to obtain the corresponding vector equation to judge region of area (ROI) belongs to normal or abnormal, and then PC-ASPECTS score was calculated. For the usability test for this system was test based on the interpretation results (PC-ASPECTS) from CT and corresponding MRI by physician.
The functions of this developed system including of the BMP and JPG images reading, selective ROI, stroke definition and PC-ASPECTS scoring. The result showed that the index which created from parameters in SVM to identify as a stroke area for pons, cerebellum, midbrain, thalamus and posterior cerebral artery (PCA) are different and the index estimated by 4 texture parameters (autocorrelation(AUT), contrast(CON), energy(ENE) and maximum probability(MAXP) in pons; AUT, ENE, entropy(ENT), and variance(VAR) in cerebellum and thalamus; AUT, ENE, ENT, MAXP, and VAR in posterior cerebral artery, PCA) are < 0.1345; > -0.6220; <-0.8282; < 0.2735, respectively. The index for midbrain is estimated by 8 texture parameters (AUT, CON, ENE, ENT, MAXP, VAR, cluster prominence and homogeneity) and < -0.6885. There was 77.5% accuracy to identify the ischemic stroke by texture parameters only, and the accuracy was growing up to 90% after SVM was used. Moreover, the estimation of accuracy, sensitivity, specificity and kappa value by PC-ASPECTS region are 95.6%, 85.3%, 98.4% and 0.86. However, the accuracy, sensitivity and specificity of stroke scoring (PC-ASPECTS ≥7) are 95%, 80% and 100%. In addition, the accuracy of PC-ASPECTS score is 70%. According to these results, the estimation of stroke by this system is more proper than the interpretation of physician inspect that this system takes more time to score the single image.
Results show that this developed system can be used in clinic, and some other parameters may need to add in to improve the performance of this system in the future.

並列關鍵字

acute ischemic stroke ； PC-ASPECTS ； texture parameters ； computer aids evaluation system ； image processing

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," 中原大學生物醫學工程系碩士論文, 民國97年7月.

[17] 張嘉麟, "以影像處理技術評估腦內微出血於電腦斷層影像之研究," 中原大學醫學工程系碩士論文, 民國102年7月.

[18] 黃馨儀, "電腦輔助ASPECTS系統於急性缺血性腦中風病人之研究," 中原大學醫學工程系碩士論文, 民國106年7月.

[4] J. Broderick, S. Connolly, E. Feldmann, D. Hanley, C. Kase, D. Krieger, et al., "Guidelines for the management of spontaneous intracerebral hemorrhage in adults: 2007 update: a guideline from the American Heart Association/American Stroke Association Stroke Council, High Blood Pressure Research Council, and the Quality of Care and Outcomes in Research Interdisciplinary Working Group," Stroke, vol. 38, pp. 2001-23, Jun 2007.

[5] C. S. K. Julio A Chalela, Lauren M Nentwich, Marie Luby, John A Butman, AndrewM Demchuk, Michael D Hill, Nicholas Patronas, Lawrence Latour, and Steven Warach, "Magnetic resonance imaging and computed tomography in emergency assessment of patients with suspected acute stroke: a prospective comparison," Lancet, vol. 369, pp. 293–298, January 27 2007.