利用三維流體套合技術估量鼠腦磁振影像中風梗塞區特性之研究

Using Volumetric Fluid Registration Techniques to Assess Infarct Regions in Rat Brain MR Images After Stroke

楊婷如(Ting-Ru Yang)

指導教授 ： 張恆華

國立臺灣大學/工學院/工程科學及海洋工程學研究所/碩士(2017年)

https://doi.org/10.6342/NTU201803252

摘要

腦中風為腦血管阻塞或自發性破裂而造成腦部功能失常的疾病，分為缺血性及出血性，急性發病期之後，在慢性期常留下不可逆的後遺症。根據世界衛生組織統計，每年全球約有1500萬人罹患中風，導致其中約500萬人因為腦中風死亡，另500萬人永久失能，因此在臨床研究中，腦中風為關注的議題。在臨床實驗模型中，大多使用齧齒動物之實驗磁振影像作為研究依據。進行腦中風實驗結果分析前需要先擷取出鼠腦梗塞中風區域，再將欲觀察之特定腦部分區依不同比例圈選出來。這些步驟皆需要專家依照每張切片一一以肉眼辨認，再手動圈選梗塞部分與特定腦部分區重疊之區塊，相當耗時。依目前現有之套合技術及工具，受限於要使用與模板影像具相同資訊之影像作為基準影像，且進行套合後，尚需再與特定腦部分區進行疊合以觀察所占比例。因此，本研究根據Y. Lou等人所提出之方法，改良一黏性流體套合技術，以Bhattacharyya距離為基礎的梯度修正物體力方程式，處理腦部磁振影像為主的影像套合系統，延伸並推導出三維流場，發展成一立體影像套合架構。本研究使用差方和、相關係數及一致性、敏感性、識別性以評估此方法。實驗使用不同的模擬影像及多組鼠腦的磁振影像，並且估算鼠腦腦部影像之梗塞區域所占大腦特定分區(M1)之比例。實驗結果顯示，本研究所提出的方法可有效解決此鼠腦影像套合之問題，並能預估大腦分區(M1)之比例，也可使用於多重模式影像套合，其計算時間大多在兩小時內可完成，在差方和及相關係數上亦有良好結果。

關鍵字

三維影像套合 ； 鼠腦套合 ； 非剛性轉換 ； Bhattacharyya距離

並列摘要

Stroke is a disease of brain dysfunction caused by cerebral vascular occlusion or spontaneous rupture, which can be divided into ischemic and hemorrhagic. About 15 million people worldwide suffer from stroke every year. Of these, 5 million die and another 5 million are permanently disabled. Therefore, clinically, stroke is a topic of concern. Mostly magnetic resonance images of rodents are used as the research basis. The infarct area is extracted first and the specific brain region to be observed is then mapped to the atlas. These steps require experts to visually recognize each infarct region in each slice and manually draw the area where the infarct region overlaps with the specific brain functional region, which is quite time consuming. Most existing registration tools are unable to register magnetic resonance images to the atlas map. Additionally, after registration, it is required to compute the overlap area between the infarct region and M1. Therefore, to address these problems, we improve the viscous fluid image registration technique proposed by Y. Lou, and extend it to three-dimension. The proposed methods is based on the Bhattacharyya distance to achieve multimodal registration. We use sum of squared difference, correlation coefficient, conformity, sensitivity, sensibility, and dice to evaluate this method. Different simulated images and numerous T2 rat brain magnetic resonance images were used to evaluate this new method. Finally, we calculated the percentage of the infarct area to the M1 area. The experimental results showed that the method we proposed effectively solved the problem of image to atlas registration, calculated the M1 ratio of the infarct regions, and could be used for multimodal image registration. The calculation time was mostly within two hours. The sum of squared difference and correlation coefficient scores indicated its good performance.

並列關鍵字

three dimensional image registration ； rat brain registration ； non-rigid model ； Bhattacharyya distance

參考文獻

[1] Stroke Statistics. Available: http://www.strokecenter.org/patients/about-stroke/stroke-statistics/

[2] Y. Lou et al., "Multimodal deformable registration of traumatic brain injury MR volumes via the Bhattacharyya distance," IEEE Trans Biomed Eng, vol. 60, no. 9, pp. 2511-20, Sep 2013.

[3] A. Daly, H. Yazid, N. E. B. Amara, and A. Zrig, "Multimodal image registration using multiresolution genetic optimization," in Information and Digital Technologies (IDT), 2017 International Conference on, pp. 557-562: IEEE, 2017.

[4] I. Reducindo et al., "Multimodal non-rigid registration methods based on local variability measures in computed tomography and magnetic resonance brain images," IET Image Processing, vol. 8, no. 12, pp. 699-707, 2014.

[5] J. Woo, M. Stone, and J. L. Prince, "Multimodal registration via mutual information incorporating geometric and spatial context," IEEE Transactions on Image Processing, vol. 24, no. 2, pp. 757-769, 2015.