MRI驱动的脑皮质厚度快速无模型估计方法

需积分: 9 41 浏览量更新于2024-07-19 收藏 836KB PDF 举报

"使用MRI进行脑皮层厚度估计的快速、模型独立的方法" 本文介绍了一种利用磁共振成像（MRI）快速且不依赖模型的方法来估计人脑的脑皮层厚度。作者包括M.L.J. Scott、P.A. Bromiley、N.A. Thacker、C.E. Hutchinson和A. Jackson，他们来自曼彻斯特大学癌症与成像科学学院的影像科学与生物医学工程分部。在研究领域，已经有许多方法被提出用于从MRI体积图像中测量脑皮层的厚度，大多数这些方法依赖于对内皮质表面和外皮质表面的可变形模型拟合。然而，模型拟合过程中为了保持球形拓扑结构以及在狭窄的脑沟中适应外皮质表面，所施加的约束可能会在某些情况下引入偏差，并显著增加处理时间。论文中，作者提出了一种替代方法，这种方法可能避免了上述问题。传统方法在狭小的脑沟区域，由于部分容积效应，脑脊液（CSF）通道可能被遮挡，导致皮层厚度测量的不准确。新方法的目标是减少这种误差并提高计算效率。脑皮层厚度的精确测量对于理解和研究各种神经退行性疾病（如阿尔茨海默病）和精神障碍（如自闭症）具有重要意义。通过更准确地估计皮层厚度，研究人员可以更深入地探究大脑结构与功能之间的关系，以及这些疾病如何影响大脑的微观结构。该方法的创新之处在于它的独立性，即不依赖于特定的模型拟合技术，这可能使其更具适应性和准确性。此外，减少处理器时间的需求对于大规模的脑成像研究来说是一个显著的优点，因为它允许更快地处理大量数据。 "使用MRI进行脑皮层厚度估计的快速、模型独立的方法"为神经科学家提供了一个强大的工具，用于更准确、更高效地评估大脑皮层的结构变化，这将有助于推进神经科学和临床医学的研究。关键词包括：MRI、大脑、皮层厚度、灰质。该论文经过了多次修改，并于2008年10月22日被接受，同年11月6日在线发布。

image noise. This removed the need for the application of an inho-

mogeneity correction algorithm prior to data processing and

thus simpliﬁed the analysis of the results, since no consideration

of the inhomogeneity correction was needed in the analysis of

the performance of the algorithm. However, several authors

(e.g. MacDonald et al., 2000) have presented cortical thickness

measurements performed on inhomogeneity corrected data. This

implies that the algorithm presented here will be applicable to

other MR image types following the application of a suitable cor-

rection algorithm. The scanning protocols used are typical of scans

acquired clinically, in that the in-plane resolution is sub-millime-

tre, whereas the through-plane resolution is of the order of a few

millimetres, in order to increase the signal to noise ratio of the

data. From a purely geometrical perspective, if it is possible to

show that the method presented below produces reliable results

when applied to such data, then there is no reason to expect that

it will not work on more research-oriented protocols that may

have isotropic voxels: the reverse may not be the case, hence the

use here of the clinical scans. Reducing the slice thickness will

inevitably reduce the signal-to-noise ratio, but this will manifest

itself as an increase in the random errors on the results, rather than

the introduction of a systematic error.

2.2. Data pre-processing

The ﬁrst of the data pre-processing stages involved application

of the partial volume segmentation algorithm described by Pokric

et al. (2001). The algorithm ﬁtted a model consisting of Gaussian

distributions for each pure tissue and uniform distributions, con-

volved with Gaussian distributions representing the noise process,

for each partial volume class, to the intensity histogram of the im-

age volume. This resulted in an estimate of the mean and standard

deviation of each pure tissue for use in later stages of the cortical

thickness measurement algorithm. The histogram was constructed

using voxel intensities within a block of the brain containing only

WM, GM and CSF, manually deﬁned in the Talairach coordinates

and then transformed into the coordinate system of each image

volume using the result of an afﬁne registration. The position of

the block is illustrated in Fig. 2. The positioning of the volume in

the frontal lobe was selected in order to obtain a sufﬁcient number

of voxels of the three pure tissue types. In elderly brains, the differ-

entiation between all three tissue types becomes less distinct for

several reasons. The boundary between GM and WM appears much

more diffuse than in younger brains (the gradient between the two

becomes much more shallow) presumably due to demyelination of

the WM axons such that the WM appears more like the non-mye-

linated GM. In addition, the presence of ﬂuid-ﬁlled lesions in WM

may result in the apparent presence of GM in the WM, and the

greater partial voluming of GM and CSF (due to increased CSF) re-

sults in the actual GM appearing to have anomalous (lower)

grey level values. In order to obtain realistic estimates of the pure

tissue values, CSF values were taken from the ventricles, WM from

the corpus callosum, where there did not appear to be much

CSF/WM partial voluming, and GM from the cortex in front of the

anterior cingulate, with relatively little CSF/GM partial voluming.

This resulted in a histogram with well-deﬁned peaks (see Fig. 3),

which could be ﬁtted in order to calibrate the mean tissue values.

Although not strictly necessary for algorithmic purposes, it was

desirable to increase the degree of isotropy of the voxel dimensions

in order to allow a single step size to be used in the search along

the normal to the inner cortical surface, as described below,

regardless of position within the image volume. Therefore, the data

were up interpolated in the through-plane direction by a factor of

two. The partial volume grey level analysis described above al-

lowed estimation of the fractional volumetric contributions of

pairs of tissues to partial volume voxels. This identiﬁes a set of pos-

sible boundary locations within the voxel, differing in which

surfaces of the voxel they intersect with. The most probable

boundary location was then selected from amongst these possibil-

ities according to the best agreement with the adjacent slices.

Fig. 2. Sections illustrating the approximate position of the cuboid, here shown in

greater contrast than the surrounding tissue, from which the three tissue histogram

is constructed: (a) axial view, (b) sagittal view.

-2500 -2000 -1500 -1000 -500 0

Number

Image Intensity Values

Data

Mixture Model

Fig. 3. Histogram of voxel intensity values (solid line) and ﬁt (dashed line) of the

Gaussian mixture model to the histogram. There are clear peaks for the cerebro-

spinal ﬂuid, grey matter and white matter at approximately 1800, 500 and 100

grey levels, respectively.

272 M.L.J. Scott et al. / Medical Image Analysis 13 (2009) 269–285

剩余16页未读，继续阅读

霍夫曼vx_helloworld7352

粉丝: 313
资源: 34

MRI驱动的脑皮质厚度快速无模型估计方法

Real-time brain extraction method from cerebral MRI volume based on graphic processing units

Cerebral palsy to cerebral palsy spectrum disorder.pdf

cerebral-uikit

react-native-cerebral

Cerebral Debugger-crx插件

cerebral-react-native-immutable-store:带有 React Native 和 Immutable Store 的 Cerebral 包

cerebral-streaming-spike:秒杀看看是否可以将Cerebral + Baobab用于流媒体应用程序

Muscle Synergy Analysis in Children with Cerebral Palsy

Designing an assessment of the child with cerebral palsy

脑调试器「Cerebral Debugger」-crx插件

最新资源