計算解剖學之腦區域擷取與對位演算法

Brain Extraction and Registration Algorithms for Computational Anatomy

劉嘉修(Jia-Xiu Liu)

指導教授 ： 陳永昇

國立交通大學/資訊學院/資訊科學與工程研究所/博士(2009年)

https://doi.org/10.6842/NCTU.2009.01164

摘要

計算解剖學（computational anatomy）廣泛運用非侵入式造影與影像處理方法於腦部結構分析上。為讓結構分析聚焦於腦組織並使腦結構空間標準化，腦區域擷取（brain extraction）與腦部結構對位（brain registration）在計算解剖學上扮演重要之角色。此兩項技術之準確性可提升結構分析之可信度，且其高執行效率可增加結構分析之處理容量進而提升統計效度。本論文針對快速且準確之腦部結構分析提出腦區域擷取與腦部對位演算法，並將所開發之影像處理技術運用於躁鬱症病患之腦纖維病理分析上。

在腦區域擷取方法中，本論文提出新的隱式形變模型（implicit deformable model）以期能有效的切割出腦部區塊。我們採用一組具區域影響性之Wendland輻射基底函數（radial basis function）以描述平滑的腦部曲線，並降低隱式形變模型之計算複雜度。在內力（internal force）與外力（external force）的交互作用下，輻射基底函數中心點將依次由初始位置逐漸推進至腦部邊界。由於不同視角下的腦部邊界曲率往往差異頗大，因此我們分別對矢狀面（sagittal view）與冠狀面（coronal view）的影像切片進行腦部區域擷取，並將不同切面的擷取結果整合以達到互補之效果。利用Internet Brain Segmentation Repository兩組測試影像所進行的效能評估結果，顯示本論文所提出的腦部擷取技術較Brain Surface Extractor、Brain Extraction Tool、Hybrid Watershed Algorithm與Model-based Level Set等方法有更佳的準確性與強健性（robustness）。

在腦結構影像對位方面，我們藉由影像導數（image derivatives）所得到的腦結構資訊，以仿射轉換（affine transformation）與非線性形變模型建立測試影像與目標影像間的空間對應關係。利用腦結構方向性之不同與腦重心位置之差異可估計一組旋轉角度與位移量，以作為仿射對位的初始值。承續仿射對位之空間對應結果，所提出的非線性方法階層式的將Wendland輻射基底函數佈置於具顯著腦結構的區域上以描述影像的形變，仿射轉換與非線性形變均以階層式之影像解析度進行影像相似度之最佳化。一般而言，非線性最佳化（nonlinear optimization）結果之優劣深受初始值之影響。而本論文所提出之方法能有效的估計仿射對位與非線性對位的初始空間對應關係。運用多組高解析度與低解析度影像之效能評估結果，顯示所提出的影像對位方法較許多已廣泛運用的演算法精準且快速。另外，本論文亦以模擬影像實驗量化影像對位準確度對腦結構分析正確性之影響。

雖然罹患第一型與第二型躁鬱症（bipolar disorders）之病患呈現相異的表徵與認知能力（cognitive functions），然而此二亞型（subtypes）是否亦具有不同之神經基質（neural substrates）卻一直是未知的。為探究第一與第二型躁鬱症病患間之腦部纖維結構性差異，我們運用所開發之影像處理技術以及由擴散張量磁振造影（diffusion tensor imaging）所導出之非等向性指標（fractional anisotropy）於正常人、第一型躁鬱症病患、第二型躁鬱症病患與所有躁鬱症病患間之腦部纖維結構之分析上。此研究採用雙樣本t檢定（two-sample t-test）之體素分析（voxel-wised analysis）方式進行，所發現的顯著性差異區域之平均非等向性指標亦用於探究其與臨床表徵及認知測驗分數之相關性。研究結果顯示第一與第二型躁鬱症病患的腦部纖維均在視丘（thalamus）、前扣帶（anterior cingulate）與下額頁（inferior frontal）等位置有顯著性之結構異常。另外，第二型躁鬱症病患的腦部纖維損傷現象亦呈現於顳頁（temporal）及下前額頁（inferior prefrontal）等區域。第一型躁鬱症的右下額頁與第二型躁鬱症的左中顳頁等區域之平均非等向性指標與認知執行功能具顯著性相關，第二型之躁鬱症病患的左中顳頁與下前額頁等區域之平均非等向性指標與楊氏躁症量表（YMRS）分數及輕躁症發作期（hypomanic episodes）呈顯著性相關。此研究結果指出第一型躁鬱症病患的腦纖維異常處多與認知功能有關，而第二型躁鬱症病患具顯著性差異之位置則涵括了認知與情緒處理等功能。

本論文提出兼具高執行效率與高準確度之腦部擷取與腦結構對位等演算法，所提出之影像處理方法的高準確性可使結構分析結果更值得信賴；而其低計算複雜度可使需要大量運用腦部擷取與腦結構對位的結構分析更有效率。本論文亦將所開發的影像處理方法運用於分析躁鬱症病患的腦部纖維結構之損傷，此結構分析結果顯示第一型與第二型躁鬱症病患具有不同之神經基質。

關鍵字

計算解剖學 ； 型態計量學 ； 腦區域擷取 ； 仿射/非線性影像對位 ； 躁鬱症

並列摘要

Computational anatomy, or morphometry, concentrates upon the quantitative analysis of brain structure, such as gyrification study and the examination of anatomical size and shape. Neuroimaging as well as image processing techniques are extensively utilized in this emerging field. Two key computerized methods of morphometry are brain extraction and registration, which can be applied to remove the non-brain tissues followed by normalizing brain structures into a standard stereotaxtic space. Accurate extraction and registration algorithms facilitate the validity of morphometric analysis. Computational anatomy generally requires large participants to provide the statistical power, and thus efficient image processing approaches support the feasibility of a large-scale study. Toward an accurate and efficient morphometric analysis, this thesis proposes a brain extraction method and brain registration algorithms. The developed image processing techniques were implemented in a voxel-based analysis protocol which was conducted to explore the fiber pathology of bipolar disorders.

The proposed brain extraction method utilizes a new implicit deformable model to well represent brain contours and to segment brain region from magnetic resonance (MR) images. This model is described by a set of Wendland's radial basis functions (RBFs) and has the advantages of compact support property and low computational complexity. Driven by the internal force for imposing the smoothness constraint and the external force for considering the intensity contrast across boundaries, the deformable model of a brain contour can efficiently evolve from its initial state toward its target by iteratively updating the RBF locations. In the proposed method, brain contours are separately determined on 2-D coronal and sagittal slices. The results from these two views are generally complementary and are thus integrated to obtain a complete 3-D brain volume. The proposed method was compared to four existing methods, Brain Surface Extractor, Brain Extraction Tool, Hybrid Watershed Algorithm, and Model-based Level Set, by using two sets of MR images along with manual segmentation results obtained from the Internet Brain Segmentation Repository. Our experimental results demonstrated that the proposed approach outperformed others when jointly considering extraction accuracy and robustness.

The proposed brain registration algorithms (BIRT) can rapidly and accurately register brain images by utilizing the brain structure information estimated from image derivatives. Source and target image spaces are related by affine transformation and non-rigid deformation. The deformation field is modeled by a set of Wendland's RBFs hierarchically deployed near the salient brain structures. In general, nonlinear optimization is heavily engaged in the parameter estimation for affine/non-rigid transformation and good initial estimates are thus essential to registration performance. In this work, the affine registration is initialized by a rigid transformation, which can robustly estimate the orientation and position differences of brain images. The parameters of the affine/non-rigid transformation are then hierarchically estimated in a coarse-to-fine manner by maximizing an image similarity measure, the correlation ratio, between the involved images. T1-weighted brain magnetic resonance images were utilized for performance evaluation.
Our experimental results using four 3-D image sets demonstrated that BIRT can efficiently align images with high accuracy compared to several extensively adopted algorithms. Moreover, a voxel-based morphometric study quantitatively indicated that accurate registration can improve both the sensitivity and specificity of the statistical inference results.

Patients with bipolar I and II disorders exhibit heterogeneous clinical presentations and cognitive functions, however, it remains unclear whether these two subtypes have distinct neural substrates. Fractional anisotropy (FA) maps calculated from diffusion tensor images and processed by the developed techniques were compared among healthy, bipolar I, and bipolar II groups using two-sample t-test analysis in a voxel-wise manner. Correlations between the mean FA value of each survived area and the clinical characteristics as well as the scores of neuropsychological tests were further analyzed. Patients of both subtypes manifested fiber impairments in the thalamus, anterior cingulate, and inferior frontal areas, whereas the bipolar II patients showed more fiber alterations in the temporal and inferior prefrontal regions. The FA values of the subgenual anterior cingulate cortices for both subtypes correlated with the performance of working memory. The FA values of the right inferior frontal area of bipolar I and the left middle temporal area of bipolar II both correlated with execution function. For bipolar II patients, the left middle temporal and inferior prefrontal FA values correlated with the scores of YMRS and hypomanic episodes, respectively. Our findings suggest distinct neuropathological substrates between bipolar I and II subtypes. The fiber alterations observed in the bipolar I patients were majorly associated with cognitive dysfunction, whereas those in the bipolar II patients were related to both cognitive and emotional processing.

This dissertation proposes brain extraction and registration algorithms which can rapidly extract brain volumes and align brain images with high accuracy. The high accuracy of our methods can facilitate computational anatomy to report accurate results. Due to the high execution efficiency, the developed image processing techniques are feasible to morphometric analysis which applies brain extraction and registration processes intensively. The proposed algorithms are also utilized to investigate the fiber impairments of bipolar disorders. Our analysis results demonstrated the distinct neuropathological substrates between bipolar I and II disorders.

並列關鍵字

computational anatomy ； morphometry ； brain extraction ； affine/non-rigid registration ； bipolar disorders

參考文獻

[73] S. Smith, P. R. Bannister, C. Beckmann, M. Brady, S. Clare, D. Flitney, P. Hansen, M. Jenkinson, D. Leibovici, B. Ripley, M. Woolrich, and Y. Y. Zhang. FSL: new tools for functional and structural brain image analysis. NeuroImage, 13(6):S249–S249, 2001.

[39] J. X. Liu, Y. S. Chen, and L. F. Chen. Nonlinear registration based on the approximation of radial basis function coefficients. Journal of Medical and Biological Engineering, 28(3):119–126, 2008.

[1] C. D. Good, I. Johnsrude, J. Ashburner, R. N. A. Henson, K. J. Friston, and R. S. J. Frackowiak. Cerebral asymmetry and the effects of sex and handedness on brain structure: a voxel-based morphometric analysis of 465 normal adult human brains. NeuroImage, 14(3):685–700, 2001.

[2] C. D. Good, I. S. Johnsrude, J. Ashburner, R. N. A. Henson, K. J. Friston, and R. S. J. Frackowiak. A voxel-based morphometric study of ageing in 465 normal adult human brains. NeuroImage, 14(1):21–36, 2001.

[3] C. Brenneis, S. M. B‥osch, M. Schocke, G. K.Wenning, andW. Poewe. Atrophy pattern in SCA2 determined by voxel-based morphometry. NeuroReport, 14(14):1799–1802, 2003.