Abstract
摘要:本研究主要目的在開發一套具有高信賴度和高精確度的電腦輔助腦部影像自動擷取(Segmentation)系統。由於現代科學的進步,人們可以經由強而有力的磁振造影(Magnetic Resonance Imaging)來探索正常和有病變的大腦之結構和功能。因為大腦的高複雜性,使得我們必須利用電腦程式化的輔助技術來幫助測量和分析大腦資料。而頭
骨去除(Skull-stripping)在腦部磁振影像處理中,扮演了一個相當關鍵也重要的角色。它不僅關係著大腦灰質(Gray matter)分佈的分析成敗,也影響到大腦皮質形態(Cortical morphology)的定量檢測。因此,一個自動而有效的電腦輔助頭骨去除工具,將能大幅減輕人力和物力在磁振影像前處理上的投資。
因為頭骨去除在磁振影像處理的重要性,世界各國的研究學者紛紛提出自己的擷取演算法。然而,大部份的方法依賴許多的經驗值參數且對腦部影像的些微變動相當敏感。因此,其精確度並無法達到後續處理的要求。我們提出一個以電荷流體模型(Charged Fluid Model)和水平集(Level set)數值方法為基礎的大腦頭骨去除系統。在這計
畫中,我們將結合此兩方法的優點來發展一個全新又有效的頭骨去除演算法。此方法將會考慮到自動調適大腦不同組織的影像強度分佈影響和邊緣偵測。我們預計發展成一個具有三度空間的可變形模型(Deformable model),並直接在立體的影像上執行頭骨去除。此外,我們也會研發一個新的雜訊去除技術,以提昇擷取精確度。最後,我們將用我們發表的成果測量(Performance measure)係數來評估這一新腦部擷取系統的準確度和擷取誤差。
Abstract: The human brain is a fascinating, complex system whose mysteries have become increasingly accessible due to the tools of modern science. Medical imaging is one powerful tool used to investigate the structure and function of the brain in both health and disease. The complexity of human brain structure mandates the use of computerized approaches derived from computer vision, image analysis, and applied mathematics fields to manipulate, analyze, and measure brain data. Within the range of potentially beneficial strategies, the greater use of computer and bioinformation technology in neuroscience research possesses particular promise and prospect. In that spirit, computer-aided segmentation of anatomic structures in the brain plays a crucial role in neuroimaging analysis.
Skull-stripping in brain MR images is one of the most important pre-processing stages for the analysis of spatial distribution of gray matter and the quantification of the cortical morphology. It is also useful for a wide range of applications from data compression to preparing for subsequent analyses which provide quantitative brain structural data. A number of skull stripping techniques have been proposed which can generally be classified into three categories: intensity-based, morphology-based, and deformable model-based. One potential disadvantage of many existing methods is that they are often dependent upon many parameters that are usually empirically generated and sensitive to small changes in the image data sets. In general, deformable models have the potential to produce more robust and accurate skull-stripping results than methods using edge detection and threshold classification.
In this study, we propose to develop a new brain extraction system based upon the charged fluid model (CFM) and the level set methods. We will investigate the mathematical relationship between these two methods and incorporate the level set numerical techniques into this hybrid deformable model to achieve better performance regarding computation speed and segmentation accuracy. The resulting algorithm will take advantages from both curve evolution and electric flows in conjunction with optimal numerical techniques. We will also investigate the development of a trilateral filter for efficient noise removal in MR images, which is anchored in the preprocessing step in the overall skull stripping procedure.
The ultimate goal of this project is to develop an efficient and robust skull stripping system for neuroimaging processing applications both clinically and academically. Toward that end, image-based forces of measuring the brain intensity and boundaries are introduced into the new deformable model for reliable skull stripping brain MR images. The entire system will be extended to direct three-dimension (3-D) computation on the image volumes, which will be evaluated using our performance measure coefficients. The success of this study will dramatically reduce the labor and effort of intracranial segmentation to facilitate the subsequent studies in neuroscience and processing applications.
Keyword(s)
可變形模型
影像分割
電荷流
體模型
水平集
磁振造影
Deformable model
image segmentation
charged fluid model
level set
magnetic resonance imaging