- 學位論文
基於水平集可變形模型之鼠腦影像中風區域分割之研究
Infarct Region Segmentation in Rat Brain Images After Stroke Using Level Set-Based Deformable Models
摘要
腦中風是造成全球人口死亡與失能的主要原因之一，近年來有許多腦中風之相關研究，在臨床實驗模型中，大多使用囓齒類動物之影像作為實驗研究依據。為了將中風區域分割出來，不僅需要專家耗時且費力的進行手動分割，也容易因為各人的評判標準不同而產生不一致之結果。因此本篇論文以大鼠作為實驗動物，利用其腦部磁振影像與經2,3,5-氯化三苯基四氮唑染色之大腦影像開發一自動演算法，將中風區域分割，以利研究者進行分析與研究。本篇論文主要以大腦動脈阻塞之大鼠作為研究對象，針對缺血性中風之鼠腦進行中風區分割。演算法分為以下幾個部分，首先因為缺血性中風位置之像素影像強度較高，利用左右腦之影像進行影像套合後之差值判定為中風區域初始輪廓，接著套用改良的可變形模型得到分割後的中風區影像。改良的可變形模型是基於窄帶區域的無邊緣主動輪廓模型，並將窄帶區域改成更局部的法線方向，並在迭代過程中對時間步長進行調整，使計算效率提升、準確率上升。本篇研究使用擴散權重磁振影像共有67隻大鼠， T2權重影像共有76隻大鼠， 2，3，5－氯化三苯基四氮唑染色影像共有43隻大鼠。結果顯示在上述三種不同鼠腦影像中，本研究演算法可以產生優良的大鼠大腦中風區分割結果。在相同的實驗設定下，本論文提出之方法優於其他可變形模型之分割結果。本研究提出一個全自動的鼠腦中風區分割方法，可以成為良好的輔佐工具，協助進行腦中風相關之研究。
並列摘要
Stroke is one of the leading causes of death and disability in the global population. In clinical experimental models, most stroke studies use rodent images as the research materials. However, manual segmentation of the stroke area is time-consuming and laborious. It may also produce different results because of different judgment standards of different persons. In this thesis, rats are used as experimental animals and their brain magnetic resonance images and 2,3,5-triphenyl tetrazolium chloride stained brain images are the main subject for automatic stroke image segmentation algorithm development. This thesis focuses on ischemic stroke segmentation using rats with cerebral artery occlusion as experimental subjects and examines the images of their brains. The algorithm is divided into the following parts. Firstly, because the pixel intensity of the ischemic stroke region is higher than the normal region, the difference between the images of the left and right hemispheres after image registration is judged as the initial stroke area. Then a modified deformable model is applied to obtain revised stroke area segmentation results. The deformable model adopts a narrow band region-based active contour model without edges and replaces the narrow-band region with local normal direction computation. The time step is adjusted during each iteration, which improves the calculation efficiency and accuracy. Diffusion-weighted magnetic resonance imaging (DWI), T2-weighted MRI, and 2,3,5- triphenyl tetrazolium chloride (TTC) stained brain stroke images were used for experiments. We used 67 rat brain DWI image subjects, 76 rat brain T2 weighted image subjects, and 43 rat brain TTC image subjects. The results show that the proposed algorithm produced good stroke segmentation results in the abovementioned three different rat brain image types. Under the same experimental settings, the proposed algorithm outperformed other deformable model methods. This thesis proposes a fully automatic method for stroke segmentation in rat brain images, which is potential for a good auxiliary tool to facilitate stroke related research using rat brain images.
參考文獻
延伸閱讀
- 林瑋祐（2019）。利用顯著區域檢測與可變形模型從事鼠腦中風梗塞影像之研究〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU201902948
- 張雅棻、魏慈慧、吳菁宜（2006）。侷限誘發療法與雙側動作訓練對中風病患腦重組機制之綜論。臺灣職能治療研究與實務雜誌，2(2)，74-81。https://doi.org/10.6534/jtotrp.2006.2(2).74
- Liu, H. T. (2012). 利用可適性彈性蛇變曲線模型從事核磁共振影像之腦部區域分割 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU.2012.00967
- 劉佩怡、謝富翰、黃百川（2013）。功能性磁振造影應用於中風個案抓握動作常見之影像分析與腦區關聯性分析方式。臺灣職能治療研究與實務雜誌，9(1)，75-83。https://doi.org/10.6534/jtotrp.2013.9(1).75
- 何世馨（2020）。Stroke Lesion Segmentation in Rat Brain MR Images Based on Deep Learning〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU202002636