開發以腦組織萎縮診斷阿茲海默症之電腦輔助偵測系統

Development of a Computer-Aided Detection System for Alzheimer's Disease Using Brain Atrophy

田耀勛(Yao-Hsun Tien)

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/cycu201800820

摘要

由於醫藥技術的進步，人類的平均壽命得以延長，老化所造成的疾病機率也變高了，如: 阿茲海默症。此病症為腦神經退化所造成的疾病是一項需及早發現及用藥的疾病。目前臨床上對於阿茲海默症的診斷，主要是以較為主觀的量表評估，尚未有一個確切的量化標準，且在疾病初期尚無法準確診斷。本論文開發一套系統，利用影像處理快速計算腦組織的萎縮參數，提供客觀數據作為醫師及早診斷的參考。
本研究開發利用CT影像計算大腦灰質厚度及腦組織、特定腦溝、腦室之體積與頭顱容積之比例的電腦輔助偵測系統，透過影像處理技術分析阿茲海默症的嚴重程度。首先以區域成長法及濾波器，去除雜訊並分割出完整的腦組織，藉由設定閥值，區分腦組織與腦脊髓液，並計算腦實質比。再使用區域成長法圈選特定腦溝、腦室，再分別利用影像增強技術及形態學處理分割灰白質及取得灰值區域之圖形骨架以利計算灰質厚度，最終以腦實質、特定腦溝、腦室與頭顱骨之體積比和灰質厚度等參數為基礎之支持向量機(SVM)自動判讀CDR分數與現有CDR結果做相關性分析。使用60組CT影像(40組影像作為訓練組，20組影像為測試組)進行系統驗證，並與病人之CDR及MRI影像進行系統效能評估。
結果顯示:經由40組訓練組的腦實質比例、左、右腦頂葉內溝、左、右腦中央溝、左、右腦外側溝、測腦室、第三腦室、腦池及全腦室等11組參數訓練SVM分類器，並利用20組測試組檢驗分類器效能，所得分類器之準確度、靈敏度、替異性及Kappa值分別為80%、86.6%、84.6%及0.547。若將訓練組中排除腦實質體積比與CDR等級有明顯誤差之病例，再統計其效能，則準確度、靈敏度、特異性及Kappa.值分別為88%、100%、84.6%及0.699。
由結果顯示本研究系統可提供醫師以CT影像取得更加客觀的量化數據，評估阿茲海默症的惡化程度，相較主流研究以MRI為基礎之演算法有相似之系統效能。相對的，由於本系統在圈選腦溝及腦室時需要手動點選種子點，這會有錯誤及判讀時間長的缺點。希望未來可發展自動辨識種子點的系統，以增加系統的效率及圈選的正確率。

關鍵字

阿茲海默症 ； 影像處理 ； 電腦輔助偵測系統 ； CT影像 ； 腦萎縮 ； 腦溝

並列摘要

Due to advances in medical technology, the average life expectancy of human beings has been prolonged and the rupture rate of diseases caused by aging was increase as well. Alzheimer's disease is caused by the deterioration of the brain nerves and need to be detected as early as possible. Currently, clinical diagnosis of Alzheimer's disease is mainly based on a more subjective scale, and there is no exact quantitative standard. It is difficult to accurately diagnose this disease at the early stage. In this study, a system which could calculate of brain tissue atrophy parameters using image processing methods was developed, and could provide an objective reference for physician to make diagnosis in early stage.
This developed computer-aided detection system can calculates the thickness of brain gray matter and the volume ratio of brain parenchyma, specific sulcus, ventricles to skull through image processing methods to analysis of the severity of Alzheimer's disease. Regional growth methods and filters were used to remove noise and segment the complete brain tissue. A threshold was set to distinguish brain tissue from cerebrospinal fluid, and then brain parenchymal ratio was calculated. The regional growth method was used to circle specific sulci and ventricles. Image enhancement technology was used to segment gray matter and then the morphological processing was used to obtain the graphical skeleton of the gray value region to calculate the gray matter thickness. Finally, the correlation between CDR and CDR scores which created with support vector machine (SVM) based on parameters of volume ratio of brain parenchyma, specific sulci, ventricle to skull and gray matter thickness was found. In this study, totally 60 sets of CT images were used, 40 sets of which were used as training groups and 20 sets were used as test groups, to train and test this system. System performance evaluation was performed by using the comparison with CDR and MRI images of patients.
In this system, SVM classifiers were trained in by 11 parameters which including parenchyma ratio, left and right brain parietal sulcus, left and right brain central sulcus, left and right lateral sulcus, cerebral ventricle, third ventricle, cerebral cistern, and whole ventricle in brain image. After using 20 sets of test groups to test the performance of the classifier, the accuracy, sensitivity, specificity, and Kappa value of the obtained classifier are 80%, 86.6%, 84.6%, and 0.547, respectively. The cases with significant difference in brain parenchymal volume ratio and CDR level in the training group were excluded; the statistical classifier performance of accuracy, sensitivity, specificity, and Kappa values are improved to 88%, 100%, 84.6% and 0.699, respectively.
The results show that the developed system can provide physicians with more objective quantitative data from CT images to assess the degrees of deterioration of Alzheimer's disease, which has similar performance compared to mainstream MRI-based system. However, the system needs to manually select seed points when selecting the sulci and ventricles; this will result in errors and long interpretation time. In the future, an automatic identification system shall be developed to increase the efficiency of the system and the correct rate of selection.

並列關鍵字

CDR ； Computer-Aided Detection System ； CT Image ； Brain Atrophy ； Sulcus

參考文獻

[1] 台灣失智症協會. (2017). 認識失智症. Available: http://www.tada2002.org.tw/tada_know_02.html

[2] M. Castleman, 阿茲海默診療室: 天下文化 2001.

[3] K. L. Double, G. M. Halliday, J. J. Krill, J. A. Harasty, K. Cullen, W. S. Brooks, et al., "Topography of brain atrophy during normal aging and alzheimer's disease," Neurobiology of Aging, vol. 17, pp. 513-521, 1996/07/01/ 1996.

[4] S. Kim, Y. C. Youn, G.-Y. R. Hsiung, S.-Y. Ha, K.-Y. Park, H.-W. Shin, et al., "Voxel-based morphometric study of brain volume changes in patients with Alzheimer’s disease assessed according to the Clinical Dementia Rating score," Journal of Clinical Neuroscience, vol. 18, pp. 916-921, 2011/07/01/ 2011.

[5] C. Möller, H. Vrenken, L. Jiskoot, A. Versteeg, F. Barkhof, P. Scheltens, et al., "Different patterns of gray matter atrophy in early- and late-onset Alzheimer’s disease," Neurobiology of Aging, vol. 34, pp. 2014-2022, 2013/08/01/ 2013.