DWI Denoising
This view provides several methods to denoise diffusion-weighted images. Simply select the image to denoise and press the start button. The default parameters should work relatively well. The NLM method is by far the slowest。
该视图提供了几种对扩散加权图像进行去噪的方法。只需选择要去噪的图像并按下开始按钮即可。默认参数的效果应该相对较好。到目前为止,NLM 方法是最慢的。
MITK Diffusion Imaging
Data formats, import and export
MITK Diffusion supports many standard image formats such as NIFTI, NRRD and DICOM as well as common tractography file formats such as vtk/fib, trk, tck, and tractography DICOM. By including multiple tractography file formats native to other tools (e.g. 3D Slicer, MRtrix, DIPY), MITK Diffusion integrates seamlessly in complex workflows involving other tools. Additional file types such as spherical-harmonic coefficient files and voxel-wise fiber orientation or peak images are compliant with MRtrix.
Data can be loaded using the open file dialog via the menu bar or by simply dragging and dropping the file into the data manager or one of data display windows.
MITK Diffusion uses a left-posterior-superior coordinate system convention. Other toolkits, e.g. MRtrix, use an RAS coordinate system. This can cause flips for example of the complete tractogram or of the voxel-wise peaks or ODFs. MITK Diffusion implements mechanisms to deal with this by automatically catching and converting some cases and by providing the tools to manually correct for it, but it is important to keep this issue in mind and to perform the data processing in a corresponding thorough manner.
MITK Diffusion 支持 NIFTI、NRRD 和 DICOM 等多种标准图像格式,以及 vtk/fib、trk、tck 和 tractography DICOM 等常见牵引成像文件格式。MITK Diffusion 包含其他工具(如 3D Slicer、MRtrix、DIPY)原生的多种牵引造影文件格式,可无缝集成到涉及其他工具的复杂工作流程中。其他文件类型,如球形谐波系数文件和体素纤维方向或峰值图像,均与 MRtrix 兼容。
可以通过菜单栏的打开文件对话框加载数据,也可以简单地将文件拖放到数据管理器或数据显示窗口中。
MITK Diffusion 使用左后上方坐标系惯例。而其他工具包(如 MRtrix)则使用 RAS 坐标系。这可能会导致整个束图或体素峰值或 ODF 等的翻转。MITK Diffusion 通过自动捕捉和转换某些情况以及提供手动纠正的工具来解决这个问题,但重要的是要记住这个问题,并以相应的彻底方式进行数据处理。
Diffusion-weighted Images扩散加权图像
This section describes the different file formats of raw diffusion-weighted images that are supported by MITK Diffusion. General points:
MITK Diffusion internally applies the image rotation matrix to the diffusion-gradient vectors while loading the image. The original gradient directions are retained and used when saving the image again.
本节介绍 MITK Diffusion 支持的不同原始扩散加权图像文件格式。一般要点:
MITK Diffusion 在加载图像时会在内部将图像旋转矩阵应用于扩散梯度向量。原始梯度方向会保留下来,并在再次保存图像时使用。
NRRD (.nrrd/.dwi)
NRRD (http://teem.sourceforge.net/nrrd/format.html) is the default file format for saving diffusion-weighted images with MITK Diffusion. The gradient and b-value information is directly stored in the file header. The image information is stored as a 3D vector image. Storing the file as .dwi or .nrrd is equivalent, only the file ending is different. Diffusion-weighted images are discerned from other images by the tag "modality:=DWMRI" in the NRRD image header. The gradient information is stored in the NRRD header in the following way:
DWMRI_b-value:=1000.000000
DWMRI_gradient_0000:=0.000000 0.000000 0.000000
DWMRI_gradient_0001:=0.000000 0.000000 1.000000
DWMRI_gradient_0002:=-0.051773 0.252917 0.966102
...
The b-value of the individual gradient directions is encoded via the squared norm of the respective vector. The tag "DWMRI_b-value" defines which b-value corresponds to a vector with norm 1.
In some cases, dMRI NRRD files contain an additional "measurement frame" matrix, which specifies an additional rotation of the gradient vectors. This matrix is automatically applied when loading the file. The original gradient directions are retained and used when saving the image again.
NRRD (http://teem.sourceforge.net/nrrd/format.html) 是使用 MITK Diffusion 保存扩散加权图像的默认文件格式。梯度和 b 值信息直接存储在文件头中。图像信息则存储为三维矢量图像。将文件存储为 .dwi 或 .nrrd 格式是相同的,只是文件结尾不同。扩散加权图像与其他图像的区别在于 NRRD 图像标头中的 "modality:=DWMRI "标签。梯度信息以如下方式存储在 NRRD 标头中:
DWMRI_b-value:=1000.000000
DWMRI_gradient_0000:=0.000000 0.000000 0.000000
DWMRI_gradient_0001:=0.000000 0.000000 1.000000
DWMRI_gradient_0002:=-0.051773 0.252917 0.966102
...
各个梯度方向的 b 值通过相应矢量的平方常模进行编码。标签 "DWMRI_b-value"(DWMRI_b-value)定义了哪个 b 值对应于常模为 1 的矢量。
在某些情况下,dMRI NRRD 文件包含一个额外的 "测量框架 "矩阵,用于指定梯度向量的额外旋转。该矩阵会在加载文件时自动应用。原始梯度方向将被保留,并在再次保存图像时使用。
NIFTI (.nii/.ni.gz)
Diffusion-weighted images can be saved and loaded as NIFTI files (NIFTI-1 https://nifti.nimh.nih.gov/nifti-1). The gradient vector information is stored in to separate files for b-values (filename.bvals) and gradient vectors (filename.bvecs). When loading a nifti file (.nii or .nii.gz), MITK Diffusion looks for these two additional files (.bval/.bvals and .bvec/.bvecs) and if they are found, MITK will offer to load the image as diffusion-weighted image. In contrast to the NRRD format, all gradient vectors should have a length of 1. The b-values are stored explicietly in the .bval/.bvals file. When loaded into MITK, this information is again encoded into the gradient vectors, as it is done in the NRRD file format.
扩散加权图像可以作为 NIFTI 文件保存和加载(NIFTI-1 https://nifti.nimh.nih.gov/nifti-1)。梯度矢量信息分别存储在 b 值文件(filename.bvals)和梯度矢量文件(filename.bvecs)中。在加载 nifti 文件(.nii 或 .nii.gz)时,MITK Diffusion 会查找这两个附加文件(.bval/.bvals 和 .bvec/.bvecs),如果找到它们,MITK 会将图像加载为扩散加权图像。与 NRRD 格式不同的是,所有梯度矢量的长度都应为 1。b 值明确存储在 .bval/.bvals 文件中。当加载到 MITK 中时,这些信息将再次被编码到梯度向量中,与 NRRD 文件格式一样。
DICOM
MITK is capable of importing diffusion-weighted DICOM images from GE, Siemens and Philips. Writing images is DICOM format is NOT supported! Mosaic images can be directly converted to regular images during import. To load a dMRI DICOM, simply drag and drop any file of the series into the application.
MITK 能够从 GE、西门子和飞利浦导入扩散加权 DICOM 图像。不支持以 DICOM 格式写入图像!马赛克图像可在导入时直接转换为普通图像。要加载 dMRI DICOM,只需将该系列的任何文件拖放到应用程序中即可。
Special Image Types
Diffusion Tensor Images (.dti)
The default format for diffusion tensor images in MITK Diffusion is a 3D NRRD file with a "3D-symmetric-matrix" pixel type. A tensor is encoded as 6 float values. The file ending for diffusion tensor files is .dti. MITK Diffusion is also able to read NIFTI DTI files (6 or 9 component format), which are for example generated by the Camino multi tensor reconstruction. To be recognized as DTI files, the .nii or .nii.gz files have to be renamed to .dti.
MITK Diffusion 中扩散张量图像的默认格式是具有 "三维对称矩阵 "像素类型的三维 NRRD 文件。张量编码为 6 个浮点数值。扩散张量文件的文件结尾是 .dti。MITK Diffusion 也能读取 NIFTI DTI 文件(6 或 9 分量格式),例如由 Camino 多张量重建生成的文件。要识别为 DTI 文件,必须将 .nii 或 .nii.gz 文件重命名为 .dti。
ODF Images (.odf, .qbi (deprecated) )
MITK stores ODFs as 252 float values spherically sampled from the continuous ODF. The sampling directions are generate by a 5-fold subdivisions of an icosahedron. ODF images are stored in NRRD file format with the ending .odf. The image information is stored as a 3D vector image with a vector length of 252.
The specific ODF sampling directions can be found here.
MITK 将 ODF(方向分布函数)存储为从连续 ODF 球面采样的 252 个浮点值。这些采样方向是通过五次细分二十面体生成的。ODF 图像以 NRRD 文件格式存储,文件后缀为 .odf。图像信息存储为向量长度为 252 的 3D 向量图像。
具体的 ODF 采样方向可以在此处找到。
Spherical Harmonics (.nii, .nii.gz, .nrrd)球面谐波
Many applications in dMRI are using spherical harmonics to store spherical functions such as ODFs. MITK Diffusion stores these files as 4D float images this is the same format as MRtrix is using.
在扩散磁共振成像(dMRI)中,许多应用使用球谐函数来存储诸如 ODF 之类的球面函数。MITK Diffusion 将这些文件存储为 4D 浮点图像,这与 MRtrix 所使用的格式相同。
Peak Images (.nii, .nii.gz, .nrrd)峰值图像
MITK Diffusion stores peak images, resulting for example from an ODF maxima extraction, as 4D float images. The peak vector components are stored in the 4th dimension, therefore dimension 4 always contains a multiple of 3 entries. This format is the same as the format used by MRtrix.
MITK Diffusion 将峰值图像(例如从 ODF 最大值提取的结果)存储为 4D 浮点图像。峰值向量的分量存储在第四维,因此第四维始终包含 3 的倍数的条目。这种格式与 MRtrix 使用的格式相同。
Tractography Formats
MITK Diffusion supports multiple formats for tractography files that are commonly used in the dMRI community. As in all other toolkits, the fiber point coordinates are stored as physical/world coordinates without any additional transformation.
MITK Diffusion 支持 dMRI 社区常用的多种束成像文件格式。与所有其他工具包一样,纤维点坐标存储为物理/世界坐标,无需任何附加转换。
VTK (.vtk/.fib)
The default format for tractograms is VTK (vtkPolyData) with the file endings .fib or .vtk. The advantage of the VTK format is that it can store additional information such as fiber weights and fiber colors. By default, both are saved with the actual tract information.
束成像的默认格式是 VTK(vtkPolyData),文件扩展名为 .fib 或 .vtk。VTK 格式的优点是可以存储附加信息,如纤维权重和纤维颜色。默认情况下,这些附加信息与实际束成像信息一起保存。
TrackVis (.trk)
TRK is the tractography file format used by TrackVis and DIPY. See http://www.trackvis.org/docs/?subsect=fileformat for a detailed description of a format.
TRK 是 TrackVis 和 DIPY 使用的束成像文件格式。有关格式的详细描述,请参见 TrackVis 文件格式文档。
MRtrix (.tck)
MITK Diffusion is able to read the tck file format native to MRtrix. Writing this format is currently not supported. By default, MRtrix uses a RAS coordinate convention in contrast to the MITK Diffusion (and also ITK) convention of LPS. To compensate for this, the fiber coordinates are negated in the x and y dimension.
MITK Diffusion 能够读取 MRtrix 原生的 tck 文件格式,但目前不支持写入这种格式。默认情况下,MRtrix 使用 RAS 坐标约定,而 MITK Diffusion(以及 ITK)使用 LPS 约定。为了弥补这一差异,纤维坐标在 x 和 y 维度上需要取负值。
DICOM
Tractography DICOM files compliant with the supplement 181 of the DICOM standard can be read and written by MITK Diffusion. This format is for example supported by the neuronavigation software of Brainlab. DICOM tags of the read tractogram can be manually set or modified using the "Properties" view in MITK Diffusion. To do this, enable the "Developer Mode" option in Window>Preferences>Properties>Developer Mode. Then select the fiber bundle in the data manager and select the "Base Data" property list in the corresponding combobox of the "Properties" view.
MITK Diffusion 可以读取和写入符合 DICOM 标准补充 181 的束成像 DICOM 文件。这种格式例如被 Brainlab 的神经导航软件所支持。读取的束成像的 DICOM 标签可以通过 MITK Diffusion 中的 "Properties" 视图手动设置或修改。要进行这些操作,需要在 "Window>Preferences>Properties>Developer Mode" 中启用 "Developer Mode" 选项。然后在数据管理器中选择纤维束,并在 "Properties" 视图的相应组合框中选择 "Base Data" 属性列表。
MITK Fiber Processing
Fiber Clustering纤维聚类
Cluster fibers using an extended version of the QuickBundles method (see [1,2]). Corrseponding command line tool is MitkFiberClustering.
使用扩展版的QuickBundles方法对纤维进行聚类(参见[1,2])。相应的命令行工具是MitkFiberClustering。
Input Data
Tractogram: Input streamlines to be clustered.
Input Centroids: Optionally input a set of streamlines around which the streamlines of the input tractograms are clustered. No new clusters are created in this case. Each input streamline is assigned to the neares centroid.
纤维束图:待聚类的输入纤维轨迹。
输入质心:可选输入一组纤维轨迹,作为聚类中心。此情况下不会创建新的簇,每条输入纤维轨迹将被分配到最近的质心。
Parameters
Cluster Size: Metric distance threshold for a streamline to be assigned to a cluster (cluster size).
Fiber Points: Resample the input streamlines to the given number of points. For scalar map and anatomical metrics this value should be much larger than for streamline shape-based metrics.
Min. Fibers per Cluster: Clusters with a smaller number of streamlines are discarded.
Max. Clusters: Only the N largest clusters are retained.
Merge Duplicate Clusters: Merge clusters based on the distance of their respective centroids using the given metric threshold. No merging is performed for a metric threshold of 0.
Output Centroids: Output the final cluster centroids.
聚类大小:用于将纤维轨迹分配到簇(聚类大小)的度量距离阈值。
纤维点数:将输入的纤维轨迹重采样为指定数量的点。对于标量图和解剖度量,此值应远大于基于纤维形状的度量。
每个簇最少纤维数:舍弃具有较少纤维轨迹的簇。
最大簇数:仅保留最大的N个簇。
合并重复簇:基于它们各自质心的距离使用给定的度量阈值合并簇。如果度量阈值为0,则不执行合并操作。
输出质心:输出最终的簇质心。
Metrics指标
All metrics can be combined using the average weighted metric value.
Euclidean: Equivalent to the the MDF of [1]. Command line tool metric string EU_MEAN.
Euclidean STDEV: Standard deviation of the point-wise euclidean distance. Fibers that run parallel have a low distance with this metric, regardless of their absolute distance. Command line tool metric string EU_STD.
Euclidean Maximum: Use maximum value of the point-weise euclidean distance. Command line tool metric string EU_MAX.
Streamline Length: Absolute streamline length difference. Command line tool metric string LENGTH.
Anatomical: Metric based on white matter parcellation histograms along the tracts (see [3]). Command line tool metric string MAP.
Scalar Map: Use the average point-wise scalar map value differences (e.g. FA) of two streamlines as distance metric. Command line tool metric string ANAT.
所有度量可以使用加权平均度量值进行组合。
欧几里得:等同于[1]中的MDF。命令行工具的度量字符串为EU_MEAN。
欧几里得标准差:点对点欧几里得距离的标准差。平行的纤维在此度量下具有较低的距离,不论它们的绝对距离。命令行工具的度量字符串为EU_STD。
欧几里得最大值:使用点对点欧几里得距离的最大值。命令行工具的度量字符串为EU_MAX。
纤维轨迹长度:绝对纤维轨迹长度差异。命令行工具的度量字符串为LENGTH。
解剖学:基于沿纤维轨迹的白质分区直方图的度量(参见[3])。命令行工具的度量字符串为MAP。
标量图:使用两条纤维轨迹的平均点对点标量图值差异(例如FA)作为距离度量。命令行工具的度量字符串为ANAT。
[1] Garyfallidis, Eleftherios, Matthew Brett, Marta Morgado Correia, Guy B. Williams, and Ian Nimmo-Smith. “QuickBundles, a Method for Tractography Simplification.” Frontiers in Neuroscience 6 (2012).
[2] Garyfallidis, Eleftherios, Marc-Alexandre Côté, François Rheault, and Maxime Descoteaux. “QuickBundlesX: Sequential Clustering of Millions of Streamlines in Multiple Levels of Detail at Record Execution Time.” ISMRM2016 (Singapore), 2016.
[3] Siless, Viviana, Ken Chang, Bruce Fischl, and Anastasia Yendiki. “AnatomiCuts: Hierarchical Clustering of Tractography Streamlines Based on Anatomical Similarity.” NeuroImage 166 (February 1, 2018): 32–45. https://doi.org/10.1016/j.neuroimage.2017.10.058.
Interactively train a classifier for tract dissection. The workflow implemented in atTRACTive has been published in [1].
要使用交互式方法训练分类器以进行纤维束解剖,可以参考atTRACTive中的工作流程。这个方法的具体实现已在[1]中发表。
Input Data
Selected Fiber Bundle: Currently selected fiber bundle (can be necessary for operations not belonging to ATTRACTIVE).
Training Fiber Bundle: Fibers belonging to that bundle are chosen to be annotated, and the model predicts on all streamlines from that bundle. It is possible to switch the bundle during the iterations, e.g., train first with Tractogram 1 and once the prediction is accurate, switch to Tractogram 2 and predict on its data.
选定纤维束:当前选定的纤维束(可能用于非ATTRACTIVE操作)。
训练纤维束:选择该纤维束中的纤维进行标注,模型将在该纤维束中的所有纤维上进行预测。可以在迭代过程中切换纤维束,例如,先使用Tractogram 1进行训练,一旦预测结果准确后,再切换到Tractogram 2并对其数据进行预测。
Parameters
Resample Bundle: Necessary to calculate dissimilarity features proposed by Olivetti et al. [2]. Needs to be done prior to training.
Number of prototypes: Define the number of prototypes and hence, the number of features to be generated per streamline. More prototypes mean more features and eventually increase the performance of the classifier. However, more features also mean more computational effort. We found 75 to be a good tradeoff since during the segmentation process iterative features are calculated (refer to our publication [2]).
重采样纤维束:这是计算Olivetti等人提出的不相似性特征所必需的步骤,需在训练前完成。
原型数量:定义原型的数量,从而确定每条纤维轨迹生成的特征数量。更多的原型意味着更多的特征,从而最终提高分类器的性能。然而,更多的特征也意味着更多的计算开销。我们发现75是一个不错的折中,因为在分割过程中,迭代特征会被计算(参见我们的出版物[2])。
Metrics
Iteratively use active learning to extract the individual chosen target tract by annotating streamlines that the classifier proposes to the human expert by its own uncertainty.
Initialization:
From Region-based Tractography: If the tractogram contains a good balance of fibers belonging to the target tract and fibers not belonging to it, you can simply click the "Add" button. A subset of randomly selected fibers will be presented for annotation. Swipe over the fibers and press
From Whole Brain Tractography: Since the number of fibers in specific white matter tracts might only be a fraction of the whole tractogram's fibers, randomly sampling might not catch enough or even any target fibers. To improve this, draw two ROIs (Regions of Interest). Create the start ROI by using the mouse middle button to define a sphere's center. The sphere's radius is set by cursor movement, and clicking the middle button creates the start ROI. Repeat for the end ROI, generating a "reducedBundle" Bundle. Random sampling is then performed on this reducedBundle. Annotation proceeds as before by clicking the "Add" button.
Once annotation is complete, click "Train classifier." Features are calculated, and a random forest is trained. The classifier predicts on all remaining streamlines of the selected tractogram, providing a "Prediction_n" (where n is the number of active learning iterations).
Uncertainty-based Annotation: If the classifier's prediction is not accurate, more fibers need to be annotated by the human expert. Simply click the "Add" button, and new fibers are presented for annotation. These fibers are selected depending on the uncertainty and dissimilarity. Repeat until the prediction of the classifier is satisfactory. If a small number of false positive streamlines are presented, you can exclude these fibers manually.
Outlier Exclusion: If the prediction is almost correct, but some single outliers exist, it is possible to exclude those in the same manner as annotating fibers not belonging to the bundle. Afterwards, it is possible to save the bundle and continue with the next subject.
Reset Classifier: To delete all annotated fibers and start dissection from scratch, it is possible to reset the classifier. However, it is possible to use the ability of the classifier to generalize and to use the already annotated fibers, so the already achieved knowledge, to segment the same white matter tract for a different subject.
Enable/Disable Workflow Help Messages: If the workflow is novel for you, help messages are presented before each action. After some time, they become obsolete, so feel free to disable.
迭代使用主动学习,通过注释分类器根据自身不确定性向专家推荐的纤维轨迹,来提取目标纤维束。
初始化:
基于区域的纤维束成像:如果纤维束图包含目标纤维束和非目标纤维束的良好平衡,可以直接点击“添加”按钮。系统会随机选择一部分纤维进行注释。如果某根纤维属于目标纤维束,按
基于全脑纤维束成像:由于特定白质纤维束中的纤维可能仅占整个纤维束图的一小部分,随机抽样可能无法捕捉到足够的目标纤维。为改善这一点,可以绘制两个感兴趣区域(ROI)。使用鼠标中键定义球体中心来创建起始ROI,通过光标移动设置球体半径,并点击中键创建起始ROI。重复该操作以生成结束ROI,形成一个“reducedBundle”束。随后在此reducedBundle上进行随机抽样。注释过程与之前相同,点击“添加”按钮即可。
注释完成后,点击“训练分类器”。系统会计算特征并训练随机森林。分类器会对选定纤维束图的剩余纤维进行预测,生成“Prediction_n”(其中n表示主动学习迭代次数)。
基于不确定性的注释:如果分类器预测不准确,需要专家注释更多纤维。点击“添加”按钮,会有新的纤维供注释。这些纤维根据不确定性和不相似性进行选择。重复这一过程,直到分类器预测令人满意。如果出现少量假阳性纤维,可以手动排除这些纤维。
异常值排除:如果预测基本正确,但存在一些异常值,可以像注释不属于目标束的纤维那样排除这些异常值。之后,可以保存纤维束并继续处理下一个对象。
重置分类器:如果需要删除所有注释纤维并重新开始,可以重置分类器。然而,也可以利用分类器的泛化能力,使用已经注释的纤维来分割不同对象的同一白质纤维束。
启用/禁用工作流程帮助信息:如果您对该流程不熟悉,系统会在每个操作前显示帮助信息。一段时间后,帮助信息可能变得不再必要,可以选择禁用。
[1] Peretzke, R. (2023). atTRACTive: Semi-automatic white matter tract segmentation using active learning. Proceedings of the 26th International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI 2023) (pp. 123-135). Springer. [2] Olivetti, E., Avesani, P.: Supervised segmentation of fiber tracts. In: International Workshop on Similarity-Based Pattern Recognition. pp. 261–274. Springer (2011)
MITK Fiber Fitting
Fiber Fit纤维拟合
Linearly fits a scalar weight for each streamline in order to optimally explain the input image. Corrseponding command line tool is MitkFitFibersToImage.
线性拟合每条纤维轨迹的标量权重,以最佳方式解释输入图像。对应的命令行工具是MitkFitFibersToImage。
Input Data and Parameters
Image: The image data used to fit the fiber weights. Possible input types:
Peak images. The input peak magnitudes are approximated using the voxel-wise fixel magnitudes obtained from the input tractogram.
Raw diffusion-weighted images. The dMRI signal is approximated using a tensor model with fixed diffusion parameters. Tensor orientations are determined by the input tractogram.
Sclar valued images (e.g. FA, axon density maps, ...).
Tractogram: The method fits a weight for each streamline in the input tractogram. In the command line tool, a list of separate bundles can be used as input.
Regularization:
Voxel-wise Variance: Constrain the fiber weights to be similar in one voxe (similar to [1]). Command line tool string "VoxelVariance".
Variance: Constrain the fiber weights to be globally similar (mean squared deaviation of weights from mean weight). Command line tool string "Variance".
Mean-squared magnitude: Enforce small weights using the mean-squared magnitude of the streamlines as regularization factor. Command line tool string "MSM".
Lasso: L1 regularization of the streamline weights. Enforces a sparse weight vector. Command line tool string "Lasso".
Group Lasso: Useful if individual bundles are used as input (command-line tool only). This regularization tries to explain the signal with as few bundles as possible penalizing the bundle-wise root mean squared magnitude of the weighs (see [2]). Command line tool string "GroupLasso".
Group Variance: Constrain the fiber weights to be similar for each bundle individually (command-line tool only). Command line tool string "GrouplVariance".
No regularization. Command line tool string "NONE".
Suppress Outliers: Perform second optimization run with an upper weight bound based on the first weight estimation (99% quantile).
Output Residuals: Add residual images to the data manager.
输入数据和参数
图像:用于拟合纤维权重的图像数据。可能的输入类型包括:
- 峰值图像:输入的峰值大小通过输入纤维束图的体素级别纤维大小进行近似。
- 原始扩散加权图像:dMRI信号通过具有固定扩散参数的张量模型进行近似。张量方向由输入纤维束图确定。
- 标量值图像(例如FA,轴突密度图等)。
纤维束图:该方法为输入纤维束图中的每条纤维轨迹拟合一个权重。在命令行工具中,可以将多个独立的束作为输入。
正则化:
- 体素级别方差:约束一个体素内的纤维权重相似(类似于[1])。命令行工具字符串为“VoxelVariance”。
- 方差:约束纤维权重在全局范围内相似(权重从平均权重的均方偏差)。命令行工具字符串为“Variance”。
- 均方幅度:使用纤维轨迹的均方幅度作为正则化因子,强制权重较小。命令行工具字符串为“MSM”。
- Lasso:L1正则化纤维轨迹权重,强制稀疏权重向量。命令行工具字符串为“Lasso”。
- 组Lasso:当使用独立束作为输入时(仅命令行工具),这种正则化尝试用尽可能少的束来解释信号,惩罚束内权重的均方根幅度(见[2])。命令行工具字符串为“GroupLasso”。
- 组方差:对每个束分别约束纤维权重相似(仅命令行工具)。命令行工具字符串为“GroupVariance”。
- 无正则化:命令行工具字符串为“NONE”。
抑制异常值:基于第一次权重估计的99%分位数,进行第二次优化运行,设置权重上限。
输出残差:将残差图像添加到数据管理器。
MITK Fiberfox
Fiber Generator纤维生成器
This view provides the user interface for defining artificial white matter fibers. Arbitrary fiber configurations like bent, crossing, kissing, twisting, and fanning bundles can be intuitively defined by positioning only a few 3D waypoints to trigger the automated generation of synthetic fibers. From these fibers a diffusion-weighted signal can be simulated using the Fiberfox View. The view also enables the automatic generation of random fiber configurations (Fig. 3).
Available sections:
Manual Fiber Definition
Known Issues
References
此视图提供了一个用户界面,用于定义人工白质纤维。用户可以通过定位少量的3D路径点来直观地定义任意纤维配置,如弯曲、交叉、接触、扭曲和扇形束,从而触发合成纤维的自动生成。通过这些纤维,可以使用Fiberfox视图模拟扩散加权信号。该视图还支持随机纤维配置的自动生成(图3)。
可用部分:
手动纤维定义
已知问题
参考资料
Manual Fiber Definition
Fiber strands are defined simply by placing markers in a 3D image volume. The fibers are then interpolated between these fiducials.
Example:
Chose an image volume to place the markers used to define the fiber pathway. If you don't have such an image available switch to the "Signal Generation" tab, define the size and spacing of the desired image and click "Generate Image". If no fiber bundle is selected, this will generate a dummy image that can be used to place the fiducials.
Start placing fiducials at the desired positions to define the fiber pathway. To do that, click on the button with the circle pictogram, then click at the desired position and plane in the image volume and drag your mouse while keeping the button pressed to generate a circular shape. Adjust the shape using the control points (Fig. 2). The position of control point D introduces a twist of the fibers between two successive fiducials. The actual fiber generation is triggered automatically as soon as you place the second control point.
In some cases the fibers are entangled in a way that can't be resolved by introducing an additional fiber twist. Fiberfox tries to avoid these situations, which arise from different normal orientations of succeeding fiducials, automatically. In rare cases this is not successful. Use the double-arrow button to flip the fiber positions of the selected fiducial in one dimension. Either the problem is resolved now or you can resolve it manually by adjusting the twist-control point.
To create non elliptical fiber profile shapes switch to the Fiber Extraction View. This view provides tools to extract subesets of fibers from fiber bundles and enables to cut out arbitrary polygonal fiber shapes from existing bundles.
手动纤维定义
纤维束的定义非常简单,只需在三维图像体积中放置标记。然后,纤维将在这些标志点之间进行插值。
示例:
- 选择一个图像体积来放置用于定义纤维路径的标记。如果没有这样的图像,可以切换到“信号生成”标签,定义所需图像的大小和间距,然后点击“生成图像”。如果没有选择纤维束,这将生成一个可以用于放置标志点的虚拟图像。
- 开始在所需位置放置标志点,以定义纤维路径。为此,点击带有圆形图标的按钮,然后在图像体积中的所需位置和平面上单击,并在按住按钮的同时拖动鼠标以生成圆形。使用控制点调整形状(图2)。控制点D的位置会在两个连续的标志点之间引入纤维的扭曲。当你放置第二个控制点时,实际的纤维生成会自动触发。
- 在某些情况下,纤维会以一种无法通过引入额外纤维扭曲来解决的方式缠绕在一起。Fiberfox会尝试自动避免这种由连续标志点的不同法线方向引起的情况,但在极少数情况下,这并不成功。使用双箭头按钮在一个维度上翻转所选标志点的纤维位置。问题可能得到解决,或者你可以通过调整扭曲控制点手动解决它。
- 若要创建非椭圆形的纤维轮廓形状,请切换到纤维提取视图。此视图提供了从纤维束中提取纤维子集的工具,并能从现有束中剪切出任意多边形纤维形状。
Fig. 1: Control points defining the actual shape of the fiducial. A specifies the fiducials position in space, B and C the two ellipse radii and D the twisting angle between two successive fiducials.
图1:定义标志物实际形状的控制点。A表示标志物在空间中的位置,B和C表示两个椭圆半径,D表示两个相邻标志物之间的扭转角度。
Fiber Options:
Real Time Fibers: If checked, each parameter adjustment (fiducial position, number of fibers, ...) will be directly applied to the selected fiber bundle. If unchecked, the fibers will only be generated if the corresponding button "Generate Fibers" is clicked.
Fiber Distribution: Specifies if the fiber distribution inside the bundle follows a uniform or normal distribution.
Fibers: Specifies the number of fibers that will be generated for the selected bundle.
Advanced Options: Show/hide advanced options
Fiber Sampling: Adjusts the distenace of the fiber sampling points (in mm). A higher sampling rate is needed if high curvatures are modeled.
Tension, Continuity, Bias: Parameters controlling the shape of the splines interpolation the fiducials. See Wikipedia for details.
Fiducial Options:
Use Constant Fiducial Radius: If checked, all fiducials are treated as circles with the same radius. The first fiducial of the bundle defines the radius of all other fiducials.
Align with grid: Click to shift the selected fiducial center points to the next voxel center.
Operations:
Rotation: Define the rotation of the selected fiber bundle around each axis (in degree).
Translation: Define the translation of the selected fiber bundle along each axis (in mm).
Scaling: Define a scaling factor for the selected fiber bundle in each dimension.
Transform Selection: Apply specified rotation, translation and scaling to the selected Bundle/Fiducial
Copy Bundles: Add copies of the selected fiber bundles to the datamanager.
Join Bundles: Add new bundle to the datamanager that contains all fibers from the selected bundles.
Include Fiducials: If checked, the specified transformation is also applied to the fiducials belonging to the selected fiber bundle and the fiducials are also copied.
光纤选项:
实时光纤: 若选中,每个参数调整(如标志物位置、光纤数量等)将直接应用于所选光纤束。若未选中,则只有点击相应的“生成光纤”按钮后才会生成光纤。
光纤分布:指定光纤束内的光纤分布是否遵循均匀分布或正态分布。
光纤: 指定将为所选光纤束生成的光纤数量。
高级选项: 显示/隐藏高级选项
光纤采样: 调整光纤采样点的距离(以毫米为单位)。若需要建模高曲率,则需要更高的采样率。
张力、连续性、偏差: 控制样条插值标志物形状的参数。详情请见维基百科。
标志物选项:
使用恒定标志物半径: 若选中,所有标志物将被视为具有相同半径的圆。光纤束的第一个标志物定义了所有其他标志物的半径。
与网格对齐: 点击以将所选标志物中心点移动到下一个体素中心。
操作:
旋转: 定义所选光纤束围绕每个轴的旋转(以度为单位)。
平移: 定义所选光纤束沿每个轴的平移(以毫米为单位)。
缩放: 为所选光纤束在每个维度定义一个缩放因子。
变换选择: 对所选光纤束/标志物应用指定的旋转、平移和缩放。
复制光纤束: 将所选光纤束的副本添加到数据管理器。
合并光纤束: 将包含所选光纤束中所有光纤的新光纤束添加到数据管理器。
包含标志物: 若选中,指定的变换也将应用于属于所选光纤束的标志物,并且这些标志物也会被复制。
Fig. 2: Examples of artificial crossing (a,b), fanning (c,d), highly curved (e,f), kissing (g,h) and twisting (i,j) fibers as well as of the corresponding tensor images generated with Fiberfox.
图2:人工交叉(a,b),扇形(c,d),高度弯曲(e,f),接吻(g,h)和扭曲(i,j)纤维的示例,以及使用Fiberfox生成的相应张量图像。
Fig. 3: Example of a random fiber phantom containing all kinds of configurations.
图3:包含各种配置的随机纤维幻影示例。
Known Issues
If a scaling factor is applied to the selcted fiber bundle, the corresponding fiducials are not scaled accordingly.
In some cases the automatic update of the selected fiber bundle is not triggered even if "Real Time Fibers" is checked, e.g. if a fiducial is deleted. If this happens on can always force an update by pressing the "Generate Fibers" button.
If any other issues or feature requests arises during the use of Fiberfox, please don't hesitate to send us an e-mail or directly report the issue in our bugtracker: https://phabricator.mitk.org/maniphest/
如果对选定的纤维束应用缩放因子,相应的标志不会相应缩放。在某些情况下,即使选中了“实时纤维”,也不会自动更新选定的纤维束,例如删除标志时。如果发生这种情况,可以随时按下“生成纤维”按钮强制更新。如果在使用Fiberfox期间出现任何其他问题或功能请求,请随时给我们发送电子邮件或直接在我们的错误跟踪系统中报告问题:https://phabricator.mitk.org/maniphest/
References
[1] Neher, P.F., Laun, F.B., Stieltjes, B., Maier-Hein, K.H., 2014. Fiberfox: facilitating the creation of realistic white matter software phantoms. Magn Reson Med 72, 1460–1470. doi:10.1002/mrm.25045
[2] Neher, P.F., Laun, F.Neher, P.F., Stieltjes, B., Laun, F.B., Meinzer, H.-P., Fritzsche, K.H., 2013. Fiberfox: A novel tool to generate software phantoms of complex fiber geometries, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[3] Neher, P.F., Stieltjes, B., Laun, F.B., Meinzer, H.-P., Fritzsche, K.H., 2013. Fiberfox: A novel tool to generate software phantoms of complex fiber geometries, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[4] Hering, J., Neher, P.F., Meinzer, H.-P., Maier-Hein, K.H., 2014. Construction of ground-truth data for head motion correction in diffusion MRI, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[5] Maier-Hein, Klaus, Neher, Peter, Houde, Jean-Christophe, Caruyer, Emmanuel, Daducci, Alessandro, Dyrby, Tim, … Descoteaux, Maxime. (2015). Tractography Challenge ISMRM 2015 Data [Data set]. Zenodo. http://doi.org/10.5281/zenodo.572345
[6] Maier-Hein, Klaus, Neher, Peter, Houde, Jean-Christophe, Caruyer, Emmanuel, Daducci, Alessandro, Dyrby, Tim, … Descoteaux, Maxime. (2017). Tractography Challenge ISMRM 2015 High-resolution Data [Data set]. Zenodo. http://doi.org/10.5281/zenodo.579933
MITK Fiber Processing
Fiber Processing纤维加工
This view provides tools to extract subbundles from a given fiber bundle, to remove unwanted fibers (e.g. via length, curvature, directional and fiber weight constraints) and to modify the properties of a fiber bundles such as coloring, fiber weights or fiber sampling.
Fiber Extraction: Place ROIs in the 2D render widgets (cricles or polygons) and extract fibers from the bundle that pass through these ROIs by selecting the according ROI and fiber bundle in the datamanger and starting the extraction. The ROIs can be combined via logical operations. All fibers that pass through the thus generated composite ROI are extracted. The extraction can also be performed using 3D ROIs represented as binary mask images. In this extraction method, the logical operations are not implemented at the moment.
此视图提供了从给定纤维束中提取子束、移除不需要的纤维(例如通过长度、曲率、方向和纤维权重约束)以及修改纤维束属性(如着色、纤维权重或纤维采样)的工具。
纤维提取:在2D渲染窗口中放置ROI(圆形或多边形),通过在数据管理器中选择相应的ROI和纤维束并开始提取,从纤维束中提取通过这些ROI的纤维。ROI可以通过逻辑操作进行组合。所有通过组合ROI的纤维都会被提取。该提取还可以使用表示为二值掩码图像的3D ROI来执行。在这种提取方法中,逻辑操作目前尚未实现。
MITK Fiber Processing
Fiber Quantification纤维定量
This view provides tools to derive additional information (such as tract density images and principal fiber direction maps) from tractograms.
该视图提供了一些工具,可以从纤维轨迹图中提取额外的信息(如纤维密度图像和主要纤维方向图)。
Input Data
Tractogram: The input streamlines.
Reference Image: The output images will have the same geometry as this reference image (optional). If a reference image with DICOM tags is used, the resulting tract envelope can be saved as DICOM Segmentation Object.
输入数据
- 纤维轨迹图:输入的纤维轨迹。
- 参考图像:输出图像将与此参考图像具有相同的几何形状(可选)。如果使用带有DICOM标签的参考图像,生成的纤维包络可以保存为DICOM分割对象。
Fiber-derived Images
Tract density image: The voxel values correspond to the sum of all fiber segment lengths in the respective voxel.
Normalized TDI: 0-1 normalized version of the TDI.
Binary envelope: Generate a binary segmentation from the input tractogram.
Fiber bundle image: Generate a 2D rgba image representation of the fiber bundle.
Fiber endings image: Generate a 2D image showing the locations of fiber endpoints.
Fiber endings pointset: Generate a poinset containing the locations of fiber endpoints (not recommended for large tractograms).
纤维衍生图像
- 纤维密度图像:体素值对应于相应体素中所有纤维段长度的总和。
- 归一化TDI:纤维密度图像的0-1归一化版本。
- 二值包络:从输入的纤维轨迹图生成二值分割图像。
- 纤维束图像:生成纤维束的二维RGBA图像表示。
- 纤维端点图像:生成显示纤维端点位置的二维图像。
- 纤维端点点集:生成包含纤维端点位置的点集(不推荐用于大型纤维轨迹图)。
Principal Fiber Directions
Calculate the voxel-wise principal fiber directions (fixels) from a tractogram.
Max. Peaks: Maximum number of output directions per voxel.
Angular Threshold: Cluster directions that are close together using the specified threshold (in degree).
Size Threshold: Discard principal directions with a magnitude smaller than the specified threshold. This value is the vector magnitude raltive to the largest vector in the voxel.
Normalization: Normalize the principal fiber directions by the global maximum, the voxel-wise maximum or each direction individually.
Output #Directions per Voxel: Generate an image that contains the number of principal directions per voxel as values.
主要纤维方向
从纤维轨迹图中计算体素级别的主要纤维方向(fixels)。
- 最大峰值数:每个体素输出方向的最大数量。
- 角度阈值:使用指定的阈值(以度为单位)对接近的方向进行聚类。
- 大小阈值:丢弃幅度小于指定阈值的主要方向。此值是相对于体素中最大向量的向量幅度。
- 归一化:通过全局最大值、体素最大值或单个方向分别归一化主要纤维方向。
- 每体素输出方向数量:生成包含每个体素主要方向数量的图像。
Input fiber bundle输入光纤束
Output principal fiber directions输出主纤维方向
MITK Fiberfox
Fiberfox DW-MRI Simulation
This view provides the user interface for Fiberfox [1,2,3], an interactive simulation tool diffusion-weighted MR images. A diffusion-weighted signal is simulated from arbitrary input fibers using a flexible combination of various diffusion models. It is possible to use manually created artificial fiber bundles (see Fiber Generator View) or fibers obtained in any other way, e.g. using fiber tractography, to simulate the signal (Example: ISMRM Tractography Challenge). The simulation can be modified using specified acquisition settings such as gradient direction, b-value, image size, image resolution, echo time, and much more. Additionally it enables the simulation of magnetic resonance artifacts including thermal noise, Gibbs ringing, N/2 ghosting, aliasing, susceptibility distortions, eddy currents and motion artifacts. The employed parameters can be saved and loaded as xml file with the ending ".ffp" (Fiberfox parameters). It is furthermore possible to add artifacts to an already existing diffusion-weighted image.
该视图为Fiberfox [1,2,3]提供了用户界面,Fiberfox是一种用于模拟扩散加权MR图像的交互式工具。通过灵活组合各种扩散模型,可以从任意输入纤维中模拟扩散加权信号。可以使用手动创建的人工纤维束(参见纤维生成视图)或通过其他方式获得的纤维(例如,使用纤维束成像技术)来模拟信号(例如:ISMRM纤维束成像挑战赛)。模拟可以通过指定的采集设置进行修改,例如梯度方向、b值、图像大小、图像分辨率、回波时间等。此外,该工具还能模拟磁共振伪影,包括热噪声、Gibbs振铃、N/2鬼影、混叠、磁化率失真、涡流和运动伪影。所使用的参数可以保存并加载为以“.ffp”结尾的xml文件(Fiberfox参数)。另外,还可以在已有的扩散加权图像上添加伪影。
Available sections:
Signal Generation
References
Signal Generation
To generate an artificial signal from the input fibers we follow the concepts recently presented by Panagiotaki et al. in a review and taxonomy of different compartment models: a flexible model combining multiple compartments is used to simulate the anisotropic diffusion inside (intra-axonal compartment) and between axons (inter-axonal compartment), isotropic diffusion outside of the axons (extra-axonal compartment 1) and the restricted diffusion in other cell types (extra-axonal compartment 2) weighted according to their respective volume fraction.
A diffusion-weighted image is generated from the fibers by selecting the according fiber bundle in the "Fiber Bundle" combobox and clicking "Generate Image". If some other diffusion-weighted image is selected together with the fiber bundle, Fiberfox directly uses the parameters of the selected image (size, spacing, gradient directions, b-values) for the signal generation process. Additionally a binary image can be selected that defines the tissue area. Voxels outside of this mask will contain no signal, only noise and other effects induced by the acquisiton (ghosts etc.). If a save path is specified, the simualted image will be saved at this location. Eventually generated log files (e.g. recording the head motion) are also saved at this location. If not path is specified, the simualted image will only appear in the data manager and has to be saved manually. Logfiles are then saved in the system specific temp directory.
If no fiber bundle but a diffusion-weighted image is selected, the specified artifacts are added to the selected image. In this mode, signal relaxation is disabled since multiple compartments are not available and the input image alrady contains relaxation effects. Also, introducing head motion is not possible since this qould require a contrast change in the weighted volumes.
信号生成
为了从输入的纤维中生成人工信号,我们采用了Panagiotaki等人最近在不同区段模型的回顾和分类中提出的概念:使用一个灵活的模型结合多个区段来模拟各向异性扩散,包括轴突内的扩散(轴突内区段)和轴突之间的扩散(轴突间区段),以及轴突外的各向同性扩散(轴突外区段1)和其他细胞类型中的受限扩散(轴突外区段2),并根据它们各自的体积分数加权。
从纤维中生成扩散加权图像的方法是,在“纤维束”组合框中选择相应的纤维束,然后点击“生成图像”。如果同时选择了其他的扩散加权图像和纤维束,Fiberfox将直接使用所选图像的参数(大小、间距、梯度方向、b值)进行信号生成过程。此外,还可以选择一个定义组织区域的二进制图像。在该掩模之外的体素将不含信号,仅包含由采集引起的噪声和其他效应(如鬼影等)。如果指定了保存路径,则模拟的图像将保存在此位置。最终生成的日志文件(如记录头部运动)也将保存在此位置。如果没有指定路径,模拟的图像将仅出现在数据管理器中,需要手动保存。日志文件将保存在系统特定的临时目录中。
如果未选择纤维束而是选择了扩散加权图像,则指定的伪影将添加到所选图像中。在这种模式下,由于没有多个区段可用且输入图像已包含松弛效应,信号弛豫被禁用。此外,由于这会在加权体积中引入对比度变化,因此无法引入头部运动。
Basic Image Settings:
Image Dimensions: Specifies actual image size (number of voxels in each dimension).
Image Spacing: Specifies voxel size in mm. Beware that changing the voxel size also changes the signal strength, e.g. increasing the resolution from 2x2x2 mm to 1x1x1 mm decreases the signal obtained for each voxel by a factor 8.
Gradient Directions: Number of gradients directions distributed equally over the half sphere. 10% baseline images are automatically added.
b-Value: Diffusion weighting in s/mm². If an existing diffusion-weighted image is used to set the basic parameters, the b-value is defined by the gradient direction magnitudes of this image, which also enables the use of multiple b-values.
Advanced Image Settings (activate checkbox "Advanced Options"):
Acquisition Type: the default acquisition type is a single shot EPI, which acquires a complete k-space slice with one echo. Alternatively, a standard spin echo sequence can be chosen that uses a cartesian k-space sampling scheme and acquires one k-space line with one echo.
Signal Scale: Additional scaling factor for the signal in each voxel. The default value of 100 results in a maximum signal amplitude of 800 for 2x2x2 mm voxels. Beware that changing this value without changing the noise variance results in a changed SNR. Adjustment of this value might be needed if the overall signal values are much too high or much too low (depends on a variety of factors like voxel size and relaxation times).
Number of Channels: Specify the number of coil elements used for the acquisition. The coil elements are circularly arranged around the objects z-axis. Currently the coil distance to the currently imaged object slice in z-direction is not taken into account, so the coil basically seems to move with the currently imaged slice along the z-axis. The signals obtained from the individual coil elements are combined using a sum of squares approach. Beware that the simulation time scales linearly with the number of coils!
Coil Sensitivity: Using multiple acquisition channels only makes sense if the coil elements have a non-constant sensitivity profile. At the moment linearly as well as exponantially decreasing coil sensitivities are implemented. Using a constant coil sensitivity, the signal received by each coil element is equal regardless of the distance to the coil. In case of a non-constant sensitivity profile the received signal intensities decrease with increasing distance from the coil element. Using a linear profile, about 50% of the signal originating from the slice center is received. In case of an exponential coil sensitivity, only about 32% of the signal originating from the slice center is received.
Echo Train Length ETL: Only relevant for Fast Spin Echo sequence (number of k-space lines acquired with one RF pulse). The echo spacing is set to TE.
Echo Time TE: Time between the 90° excitation pulse and the first spin echo. Increasing this time results in a stronger T2-relaxation effect (Wikipedia).
Repetition Time TR: Time between two 90° RF pulses. Important for T1 contrast (use short TE and TR for strong T1 weighting).
Inversion Time TI: Time between 180° inversion pulse and 90° RF pulse. If 0, no inversion pulse is simulated.
Dwell Time: Time to read one line in k-space. Increasing this time results in a stronger T2* effect which causes an attenuation of the higher frequencies in phase direction (here along y-axis) which again results in a blurring effect of sharp edges perpendicular to the phase direction.
Tinhom Relaxation (T2'): Time constant specifying the signal decay due to magnetic field inhomogeneities (also called T2'). Together with the tissue specific relaxation time constant T2 this defines the T2* decay constant: T2*=(T2 T2')/(T2+T2')
Fiber Radius (in µm): Used to calculate the volume fractions of the used compartments (fiber, water, etc.). If set to 0 (default) the fiber radius is set automatically so that the voxel containing the most fibers is filled completely. A realistic axon radius ranges from about 5 to 20 microns. Using the automatic estimation the resulting value might very well be much larger or smaller than this range.
Reverse Phase Encoding Direction: Switch anterior-posterior and posterior-anterior phase encoding.
Simulate Signal Relaxation: If checked, the relaxation induced signal decay is simulated, other wise the parameters TE, Line Readout Time, Tinhom, and T2 are ignored.
Disable Partial Volume Effects: If checked, the actual volume fractions of the single compartments are ignored. A voxel will either be filled by the intra axonal compartment completely or will contain no fiber at all.
Output Additional Images: Output a double image for each compartment. The voxel values correspond to the volume fraction of the respective compartment.
基本图像设置:
图像尺寸:指定图像的实际大小(每个维度的体素数量)。
图像间距:指定体素大小(以毫米为单位)。注意,改变体素大小也会改变信号强度,例如,将分辨率从2x2x2毫米增加到1x1x1毫米时,每个体素的信号会减少8倍。
梯度方向:在半球上均匀分布的梯度方向数量。系统会自动添加10%的基线图像。
b值:扩散加权系数,以s/mm²为单位。如果使用现有的扩散加权图像来设置基本参数,则b值由该图像的梯度方向幅度定义,这也使得可以使用多个b值。
高级图像设置(勾选“高级选项”复选框后可见):
采集类型:默认采集类型为单次EPI(单次激发平面成像),它可以用一次回波采集完整的k空间切片。或者,可以选择标准自旋回波序列,它使用笛卡尔k空间采样方案,用一次回波采集一条k空间线。
信号比例:每个体素信号的附加缩放因子。默认值为100,结果是在2x2x2毫米体素中信号幅度最大为800。注意,在不改变噪声方差的情况下改变此值会改变信噪比。如果整体信号值过高或过低(取决于体素大小和弛豫时间等各种因素),可能需要调整此值。
通道数量:指定用于采集的线圈元素数量。线圈元素环绕物体z轴圆形排列。目前线圈与当前成像切片在z方向的距离没有考虑在内,因此线圈基本上似乎随着当前成像切片沿z轴移动。各个线圈元素获得的信号通过平方和方法组合。注意,模拟时间与线圈数量线性相关!
线圈灵敏度:只有在线圈元素具有非恒定灵敏度轮廓时,使用多个采集通道才有意义。目前实现了线性和指数递减的线圈灵敏度。使用恒定灵敏度时,每个线圈元素接收到的信号在距离相同的情况下是相等的。在非恒定灵敏度轮廓情况下,接收到的信号强度随着距离线圈元素的增加而减少。使用线性轮廓时,约50%的信号来自切片中心;使用指数轮廓时,只有约32%的信号来自切片中心。
回波列长度(ETL):仅对快速自旋回波序列相关(一次射频脉冲采集的k空间线数量)。回波间距设置为TE。
回波时间(TE):从90°激发脉冲到第一次自旋回波的时间。增加此时间会导致更强的T2弛豫效应。
重复时间(TR):两次90°射频脉冲之间的时间。对T1对比度重要(使用短TE和TR以获得强T1加权)。
反转时间(TI):180°反转脉冲和90°射频脉冲之间的时间。如果为0,则不模拟反转脉冲。
驻留时间:读取k空间一行的时间。增加此时间会导致更强的T2效应,导致相位方向(这里沿y轴)的高频衰减,从而导致垂直于相位方向的锐边模糊效应。
非均匀弛豫(T2'):指定由于磁场不均匀性导致的信号衰减的时间常数。与组织特异的弛豫时间常数T2一起定义T2衰减常数:T2* = (T2 T2') / (T2 + T2')
纤维半径(以微米为单位):用于计算使用的不同成分(纤维、水等)的体积分数。如果设置为0(默认),纤维半径将自动设置,以便包含最多纤维的体素完全填充。现实的轴突半径范围大约为5到20微米。使用自动估算,结果值可能会大大超出或低于此范围。
反向相位编码方向:切换前后相位编码。
模拟信号弛豫:如果勾选,将模拟弛豫引起的信号衰减,否则将忽略TE、线读取时间、非均匀弛豫和T2参数。
禁用部分体积效应:如果勾选,将忽略各单个成分的实际体积分数。体素要么完全由轴内成分填充,要么完全不包含纤维。
输出附加图像:为每个成分输出双图像。体素值对应于各成分的体积分数。
Compartment Settings:
The group-boxes "Intra-axonal Compartment", "Inter-axonal Compartment" and "Extra-axonal Compartments" allow the specification which model to use and the corresponding model parameters. Currently the following models are implemented:
Stick: The “stick” model describes diffusion in an idealized cylinder with zero radius. Parameter: Diffusivity d
Zeppelin: Cylindrically symmetric diffusion tensor. Parameters: Parallel diffusivity d|| and perpendicular diffusivity d⊥
Tensor: Full diffusion tensor. Parameters: Parallel diffusivity d|| and perpendicular diffusivity constants d⊥1 and d⊥2
Ball: Isotropic compartment. Parameter: Diffusivity d
Astrosticks: Consists of multiple stick models pointing in different directions. The single stick orientations can either be distributed equally over the sphere or are sampled randomly. The model represents signal coming from a type of glial cell called astrocytes, or populations of axons with arbitrary orientation. Parameters: randomization of the stick orientations and diffusivity of the sticks d.
Dot: Isotropically restricted compartment. No parameter.
Prototype Signal: EXPERIMENTAL FEATURE!!! The signal is not generated from a parametric model but a prototype signal is sampled from the selected diffusion-weighted image. Parameters: The number of prototype signals that are used for the signal generation (at each fiber position one is picked randomly) and the constraining diffusion parameters for a voxel signal to be included in the list. For a fiber signal one would for example probably select a high FA and for a CSF voxel a low FA.
For a detailed description of the individual models, please refer to Panagiotaki et al. "Compartment models of the diffusion MR signal in brain white matter: A taxonomy and comparison".
Additionally to the model parameters, each compartment has its own T1 and T2 signal relaxation constants (in ms). This constants are not relevant if the prototype signal model is used, since in this case signal relaxation is disabled. Furthermore, it is possible to specify a volume fraction map for each compartment:
The volume fraction maps for compartment 1 and 2 (fiber compartments) are optional. If they are not specified, the corresponding volume fractions are directly determined from the fiber bundle. Additionally, it is assumed that in this case all volume fraction maps of the non-fiber compartments contain values relative to the remaining non-fiber volume, not absolute fractions of the complete voxel volume. This ensures that the automatically determined fiber volumes and the map-defined non-fiber volumes sum up to 1 in each voxel.
If one non-fiber compartment is used but no corresponding volume fraction map is specified, the corresponding volume is automatically set to the remaining volume (voxel volume - fiber volume).
If four compartments are used, at least one of the extra axonal compartment volume fraction maps has to be specified. The second one can be automatically determined from the respective other (1-f). If this is the case, the non-fiber volume information is again regarded as relative to the available non-fiber volume.
分区设置:
“轴内分区”、“轴间分区”和“轴外分区”组框允许指定使用哪种模型及其对应的模型参数。目前实现的模型如下:
- Stick: "Stick" 模型描述了在理想化的零半径圆柱体中的扩散。参数:扩散系数 d
- Zeppelin:具有圆柱对称的扩散张量。参数:平行扩散系数 d|| 和垂直扩散系数 d⊥
- Tensor:完整的扩散张量。参数:平行扩散系数 d|| 和两个垂直扩散系数 d⊥1 和 d⊥2
- Ball:各向同性分区。参数:扩散系数 d
- Astrosticks:由指向不同方向的多个 stick 模型组成。单个 stick 的方向可以均匀分布在球面上,也可以随机抽样。该模型表示来自一种名为星形胶质细胞的神经胶质细胞或任意方向的轴突群的信号。参数:stick 方向的随机性和 stick 的扩散系数 d
- Dot:各向同性受限的分区。无参数。
- Prototype Signal:实验功能!!!信号不是从参数模型生成的,而是从所选的扩散加权图像中抽样的原型信号生成的。参数:用于信号生成的原型信号数量(在每个纤维位置随机选择一个)以及包含在列表中的体素信号的约束扩散参数。例如,对于纤维信号,可能会选择高 FA,对于 CSF 体素,则选择低 FA。
关于各个模型的详细描述,请参阅 Panagiotaki 等人的论文《大脑白质中扩散 MR 信号的分区模型:分类和比较》。
除模型参数外,每个分区都有其自身的 T1 和 T2 信号弛豫常数(以毫秒为单位)。如果使用原型信号模型,这些常数则不相关,因为在这种情况下信号弛豫被禁用。此外,还可以为每个分区指定体积分数图:
1 和 2 分区(纤维分区)的体积分数图是可选的。如果未指定,则相应的体积分数直接从纤维束中确定。此外,在这种情况下,假设所有非纤维分区的体积分数图包含相对于剩余非纤维体积的值,而不是整个体素体积的绝对分数。这确保了自动确定的纤维体积和地图定义的非纤维体积在每个体素中总和为1。
如果使用一个非纤维分区但未指定相应的体积分数图,则相应的体积自动设置为剩余体积(体素体积 - 纤维体积)。
如果使用四个分区,至少需要指定一个轴外分区体积分数图。第二个可以从相应的另一个自动确定(1-f)。在这种情况下,非纤维体积信息再次被视为相对于可用的非纤维体积的相对值。
Noise and Artifacts:
Noise: Add Rician or Chi-Square distributed noise with the specified variance to the signal.
Spikes: Add signal spikes to the k-space signal resulting in stripe artifacts across the corresponding image slice.
Aliasing: Aliasing artifacts occur if the FOV in phase direction is smaller than the imaged object. The parameter defines the percentage by which the FOV is shrunk.
N/2 Ghosts: Specify the offset between successive lines in k-space. This offset causes ghost images in distance N/2 in phase direction due to the alternating EPI readout directions.
Distortions: Simulate distortions due to magnetic field inhomogeneities. This is achieved by adding an additional phase during the readout process. The input is a frequency map specifying the inhomogeneities. The "Fieldmap Generator" view provides an interface to generate simple artificial frequency maps. To egnerate realistic distortions for an in vivo like dataset we recommend using a frequency map acquired during a real MR scan or one estimated with tools such as FSL TOPUP.
Motion Artifacts: To simulate motion artifacts, the fiber configuration is moved between the signal simulation of the individual gradient volumes. The motion can be performed randomly, where the parameters are used to define the +/- maximum of the corresponding motion, or linearly, where the parameters define the maximum rotation/translation around/along the corresponding axis at the and of the simulated acquisition.
Eddy Currents: Eddy current induced magnetic field gradient (in mT/m) at the beginning of the k-space readout. A spatially linear eddy current profile in the direction of the respective diffusion-weighting gradient is used. The eddy current induced gradient decays with a time constant τ=70ms.
Gibbs Ringing: Ringing artifacts occurring on edges in the image due to the frequency low-pass filtering caused by the limited size of the k-space.
噪声和伪影:
- 噪声:在信号中添加具有指定方差的莱斯(Rician)或卡方(Chi-Square)分布的噪声。
- 尖峰:在k空间信号中添加尖峰,导致相应图像切片上出现条纹伪影。
- 混叠:如果相位方向的视野(FOV)小于被成像物体,则会产生混叠伪影。参数定义了视野收缩的百分比。
- N/2鬼影:指定k空间中连续行之间的偏移量。由于交替的EPI读出方向,这种偏移会在相位方向上造成N/2距离的鬼影。
- 失真:模拟由于磁场不均匀性引起的失真。这通过在读出过程中添加额外相位来实现。输入是指定不均匀性的频率图。Fieldmap Generator视图提供了生成简单人工频率图的接口。为了生成真实的体内数据失真,我们建议使用在实际MR扫描期间获得的频率图或使用如FSL TOPUP工具估算的频率图。
- 运动伪影:为了模拟运动伪影,在各个梯度体积的信号模拟之间移动纤维配置。运动可以随机进行,参数用于定义相应运动的最大正负值;或者线性进行,参数定义在模拟采集结束时围绕/沿相应轴的最大旋转/平移。
- 涡流:在k空间读出开始时引入涡流引起的磁场梯度(单位:mT/m)。使用在相应扩散加权梯度方向上具有空间线性的涡流剖面。涡流引起的梯度以时间常数τ=70ms衰减。
- 吉布斯振铃:由于k空间大小有限的频率低通滤波器导致图像边缘出现振铃伪影。
Fig. 1: Realistic simulation of a whole brain dataset with multiple artifacts.图1:包含多种伪影的全脑数据集的真实模拟。
References
[1] Neher, P.F., Laun, F.B., Stieltjes, B., Maier-Hein, K.H., 2014. Fiberfox: facilitating the creation of realistic white matter software phantoms. Magn Reson Med 72, 1460–1470. doi:10.1002/mrm.25045
[2] Neher, P.F., Laun, F.Neher, P.F., Stieltjes, B., Laun, F.B., Meinzer, H.-P., Fritzsche, K.H., 2013. Fiberfox: A novel tool to generate software phantoms of complex fiber geometries, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[3] Neher, P.F., Stieltjes, B., Laun, F.B., Meinzer, H.-P., Fritzsche, K.H., 2013. Fiberfox: A novel tool to generate software phantoms of complex fiber geometries, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[4] Hering, J., Neher, P.F., Meinzer, H.-P., Maier-Hein, K.H., 2014. Construction of ground-truth data for head motion correction in diffusion MRI, in: Proceedings of International Society of Magnetic Resonance in Medicine.
[5] Maier-Hein, Klaus, Neher, Peter, Houde, Jean-Christophe, Caruyer, Emmanuel, Daducci, Alessandro, Dyrby, Tim, … Descoteaux, Maxime. (2015). Tractography Challenge ISMRM 2015 Data [Data set]. Zenodo. http://doi.org/10.5281/zenodo.572345
[6] Maier-Hein, Klaus, Neher, Peter, Houde, Jean-Christophe, Caruyer, Emmanuel, Daducci, Alessandro, Dyrby, Tim, … Descoteaux, Maxime. (2017). Tractography Challenge ISMRM 2015 High-resolution Data [Data set]. Zenodo. http://doi.org/10.5281/zenodo.579933
MITK Fiberfox
Fieldmap Generator
This view allows the creation of artificial frequency maps used by Fiberfox to introduce distortions into diffusion weighted images. The generated images can contain a linear frequency gradient and/or multiple 3D gaussian shaped field inhomogeneities.
Example:
Select a reference image with the combo box. The generated fieldmap will feature the same geometry as the selected image.
Move the crosshair to the any position in the image and click "Place Field Source".
A position marker will appear in the render windows and in the datamanager, which indicates the position of a 3D gaussian field distortion that will be introduced upon clicking "Generate Fieldmap".
The strength and variance of the placed sources can be modified by selecting the corresponding data node in the data manager and adjusting the parameters in the lower part of the view (below "Edit Selected Source").
To introduce an (additional) linear frequency gradient, specify the gradient below "Add Gradient".
To finally generate the fieldmap, press "Generate Fieldmap".
此视图允许创建由Fiberfox用来在扩散加权图像中引入失真的人工频率图。生成的图像可以包含线性频率梯度和/或多个3D高斯形状的场不均匀性。
例如:
- 通过组合框选择一个参考图像。生成的场图将具有与所选图像相同的几何形状。
- 将十字准线移动到图像中的任意位置,然后点击“放置场源”。
- 渲染窗口和数据管理器中将出现一个位置标记,指示在点击“生成场图”时将引入一个3D高斯场失真的位置。
- 可以通过选择数据管理器中的相应数据节点并在视图下方(“编辑选定源”下)调整参数,来修改放置源的强度和方差。
- 要引入(额外的)线性频率梯度,请在“添加梯度”下指定梯度。
- 最后,按下“生成场图”以生成场图。
The Fieldmap Generator View. The render window shows a diffusion weighted image of the brain superimposed by a frequency map with two 3D gaussian field inhomogeneities (red).场图生成器视图。渲染窗口显示了大脑的弥散加权图像,并叠加了一个包含两个三维高斯场不均匀性的频率图(红色部分)。
MITK Tractography
Global Gibbs Tractography全局吉布斯径迹成像
This view provides the user interface for the global Gibbs tractography algorithm, a global fiber tracking algorithm, originally proposed by Reisert et.al. [1].
The corresponding comman line application is named "MitkGlobalTractography".
该视图提供了全局吉布斯径迹算法的用户界面,这是一种全局纤维追踪算法,最初由Reisert等人提出[1]。
对应的命令行应用程序名为 "MitkGlobalTractography"。
Input Data
Mandatory Input:
One ODF or tensor image selected in the datamanager
Optional Input:
Mask Image: White matter probability mask. Corresponds to the probability to generate fiber segments in the respective voxel.
必需输入:
在数据管理器中选择的一个ODF或张量图像。
可选输入:
蒙版图像:白质概率蒙版,对应于在各自体素中生成纤维段的概率。
Parameters
Number of iterations: More iterations causes the algorithm to be more stable but also to take longer to finish the tracking. Recommended: minimum 10^8 iterations for full brain tractography.
Particle length/width/weight controlling the contribution of each particle to the model M
Start and end temperature controlling how fast the process reaches a stable state. (usually no change needed)
Weighting between the internal (affinity of the model to long and straigt fibers) and external energy (affinity of the model towards the data). (usually no change needed).
Minimum fiber length constraint (in mm). Shorter fibers are discarded after the tracking.
The automatic selection of parameters for the particle length/width and weight are determined directly from the input image using information about the image spacing and GFA.
- 迭代次数:增加迭代次数可以使算法更稳定,但同时会延长追踪时间。建议:进行全脑径迹成像至少需要 108108 次迭代。
- 粒子长度/宽度/权重:控制每个粒子对模型M的贡献。
- 起始和结束温度:控制过程达到稳定状态的速度。(通常情况下不需要更改)
- 内部权重和外部能量之间的加权:内部权重指模型对长且直纤维的亲和力,外部能量指模型对数据的亲和力。(通常情况下不需要更改)
- 最小纤维长度约束(单位:毫米):追踪后会丢弃长度较短的纤维。
粒子长度/宽度和权重的自动选择直接从输入图像中获取,利用图像间距和GFA(Generalized Fractional Anisotropy)的信息。
Surveilance of the tracking process追踪过程监控
Once started, the tracking can be monitored via the textual output that informs about the tracking progress and several stats of the current state of the algorithm. If enabled, the intermediate tracking results are displayed in the renderwindows each second. This live visualization should usually be disabled for performance reasons. It can be turned on and off during the tracking process via the according checkbox. The button next to this checkbox allows the visualization of only the next iteration step.
开始后,可以通过文本输出监视追踪进度和算法当前状态的多个统计信息。如果启用了实时可视化,在渲染窗口每秒显示中间追踪结果。通常情况下,出于性能考虑,应该禁用此实时可视化功能。在追踪过程中可以通过相应复选框随时开启或关闭此功能。复选框旁边的按钮允许仅可视化下一个迭代步骤的结果。
References
[1] Reisert, M., Mader, I., Anastasopoulos, C., Weigel, M., Schnell, S., Kiselev, V.: Global fiber reconstruction becomes practical. Neuroimage 54 (2011) 955-962
MITK dMRI RegistrationMITK dMRI 配准
Head-Motion Correction头部运动校正
This view allows head-motion and eddy-current correction by affinely registering all volumes to the first unweighted volume of the complete diffusion-weighted image. The individual gradient directions are roated accordingly.
该视图通过将所有体积仿射配准到完整扩散加权图像的第一个无权重体积来实现头部运动和涡流校正。各个梯度方向也会相应旋转。
MITK IVIM
Intra-voxel incoherent motion estimation (IVIM)微观体素内不相干运动估算(IVIM)
This view enables the estimation of "Intra-voxel incoherent motion estimation" (IVIM) and diffusion kurtosis parameters from diffusion weighted images. The required input is a diffusion weighted image that was acquired with multiple different b-values (unidirectional).
The view enables an interactive exploration of the selected image (click around in the image and watch the estimated parameters in the figure of the view) as well as generation of parameter maps.
此视图使得可以从多b值(单方向)获取的扩散加权图像中估算“微观体素内不相干运动估算”(IVIM)和扩散峰度参数。所需输入是一幅用多个不同b值采集的扩散加权图像。
该视图允许用户互动式地探索选定的图像(点击图像中的任意位置,可以在视图的图形中查看估算的参数)以及生成参数图。
IVIM fit methods IVIM拟合方法
The IVIM model is a bi-exponential model defined by the parameters f, D and D: (1-f)e{-bD}+fe{-b(D+D^)}. All fit methods available in the view ultimatle use this model but they differ how the three parameters are estimated:
Jointly fit D, f and D: All three parameters are estimated simultaneously.
Fit D & f with fixed D value: Only f and D are estimated, D* is fixed (user defined).
Fit D & f (high b), then fit D: First fit f and D (monoexponentially (1-f)e^{-bD}) and use these then fixed parameters in a second fit of D with the complete bi-exponential model.
Linearly fit D & f (high b), then fit D: First fit f and D linearly and use these then fixed parameters in a second fit of D with the complete bi-exponential model.
Negative values for D and D* are penalized during the fit.
IVIM模型是一个双指数模型,由参数f、D和D定义:$(1-f)e^{-bD} + fe{-b(D+D)}$。视图中提供的所有拟合方法最终都是使用该模型,但它们在估算这三个参数的方法上有所不同:
- 联合拟合D、f和D*:同时估算所有三个参数。
- 在固定D*值的情况下拟合D和f:仅估算f和D,D*值固定(由用户定义)。
- 高b值下拟合D和f,然后拟合D:首先以单指数$(1-f)e^{-bD}$拟合f和D,然后在第二次拟合中使用完整的双指数模型固定这些参数来估算D。
- 高b值下线性拟合D和f,然后拟合D:首先线性拟合f和D,然后在第二次拟合中使用完整的双指数模型固定这些参数来估算D。
在拟合过程中,D和D*的负值会受到惩罚。
Suggested Readings
Toward an optimal distribution of b values for intravoxel incoherent motion imaging. Lemke A, Stieltjes B, Schad LR, Laun FB. Magn Reson Imaging. 2011 Jul;29(6):766-76. Epub 2011 May 5. PMID: 21549538
Differentiation of pancreas carcinoma from healthy pancreatic tissue using multiple b-values: comparison of apparent diffusion coefficient and intravoxel incoherent motion derived parameters. Lemke A, Laun FB, Klauss M, Re TJ, Simon D, Delorme S, Schad LR, Stieltjes B. Invest Radiol. 2009 Dec;44(12):769-75. PMID: 19838121
MITK Reconstruction
ODF Details详情
This view provides detailed information about the orentation distribution function at the current crosshair position (if a Tensor/ODF image is selected). A visualization of the ODF as well as statistical information are displayed.
该视图提供了当前十字准线位置上方向分布函数的详细信息(如果选择了张量/ODF图像)。不仅展示了ODF的可视化效果,还提供了统计信息。
MITK Reconstruction
ODF Peak Extraction ODF峰值提取
This view provides the user interface to extract the principal diffusion direction of tensors and the peaks of spherical harmonic ODFs using gradient descent optimization.
The output peaks can for example be used for streamline tractography.
此视图为用户提供界面,利用梯度下降优化提取张量的主要扩散方向和球谐ODF的峰值。
这些输出的峰值可以用于流线束成像
MITK Preprocessing
Preprocessing of Diffusion Images扩散图像的预处理
Preprocessing预处理
The Preprocessing View bundles a selection of features and tools for dealing with diffusion weighted MR images. It can be used to modify header and geometry information, manipulate gradients and gradient images as well as calculating ADC maps.
预处理视图集合了一系列用于处理扩散加权MR图像的功能和工具。它可以用于修改头部和几何信息,操纵梯度和梯度图像以及计算ADC图。
Details
Gradients tab梯度选项卡
The Gradients tab allows you to examine the number of b values and how many gradients were acquired for each. Here you can also visualize the gradients as points on a sphere, mirror opposite gradients so all are oriented within the same half sphere, round b values and merge similar gradients.
在 "梯度 "选项卡中,您可以查看 b 值的数量以及每个值获得的梯度数量。在这里,您还可以将梯度可视化为球面上的点、镜像相反的梯度以便所有梯度都在同一个半球内、舍入 b 值并合并相似的梯度。
Image values tab图像值选项卡
The Image values tab enables you to manipulate the voxel values by e.g. resampling images, averaging repetitions of the same gradient, normalizing the image values or merging several DWIs into one.
在 "图像值 "选项卡中,您可以通过对图像重新取样、对同一梯度的重复图像取平均值、对图像值进行归一化处理或将多个 DWI 合并为一个等方式来处理体素值。
Header tab 标题选项卡
In the Header tab you can selectively change geometry information of the image and delete or extract specific gradient volumes.
在 "标题 "选项卡中,您可以有选择性地更改图像的几何信息,删除或提取特定的渐变体积。
Other tab其他选项卡
Miscellaneous tools in the Other tab allow for ADC map calculation and the extraction of the b0 image with the possibility of averaging all b0 images.
其他选项卡中的其他工具可用于 ADC 图计算和提取 b0 图像,并可对所有 b0 图像进行平均处理。
MITK Reconstruction
Q-Ball Reconstruction重建
The q-ball reonstruction view implements a variety of reconstruction methods. The different reconstruction methods are described in the following:
Numerical: The original, numerical q-ball reconstruction presented by Tuch et al. [5]
Standard (SH): Descoteaux's reconstruction based on spherical harmonic basis functions [6]
Solid Angle (SH): Aganj's reconstruction with solid angle consideration [7]
ADC-profile only: The ADC-profile reconstructed with spherical harmonic basis functions
Raw signal only: The raw signal reconstructed with spherical harmonic basis functions
Q-ball 重建视图实现了多种重建方法。以下是这些不同的重建方法:
数值法:由Tuch等人提出的原始数值q-ball重建方法【5】。
标准法(SH):基于球谐基函数的Descoteaux重建方法【6】。
立体角法(SH):考虑立体角的Aganj重建方法【7】。
仅ADC轮廓:使用球谐基函数重建的ADC轮廓。
仅原始信号:使用球谐基函数重建的原始信号。
The q-ball resonstruction view
B0 threshold works the same as in tensor reconstruction. The maximum l-level configures the size of the spherical harmonics basis. Larger l-values (e.g. l=8) allow higher levels of detail, lower levels are more stable against noise (e.g. l=4). Lambda is a regularisation parameter. Set it to 0 for no regularisation. lambda = 0.006 has proven to be a stable choice under various settings.
B0阈值在张量重构中的作用与其相同。最大l级别确定了球谐基函数的大小。较大的l值(例如l=8)允许更高级别的细节,较低级别则对噪声更稳定(例如l=4)。Lambda是正则化参数。将其设置为0表示不进行正则化。在各种设置下,lambda = 0.006 已被证明是一个稳定的选择。
Advanced q-ball reconstruction settings高级 q 球重建设置
This is how a q-ball image should initially look after reconstruction. Standard q-balls feature a relatively low GFA and thus appear rather dark. Adjust the level-window to solve this.
这是重建后Q球图像的初始外观。标准的Q球通常具有相对较低的GFA值,因此看起来比较暗淡。调整级别窗口可以解决这个问题。
q-ball image after reconstruction重建后的 q 球图像
MITK dMRI Registration
Registration配准
This view enables the simple rigid or affine registration of two images. It is also possible to transform a tractogram with a registration object obtained from a previous registration of two images.
这个视图允许简单地对两幅图像进行刚性或仿射配准。同时,也可以使用从前一次两幅图像配准中获得的配准对象来转换一条束跟踪图。
MITK Reconstruction
Scalar Indices标量指标
The scalar indices view allows the calculation of different scalar measures for raw diffusion-weighted images (MD, ADC), tensors (Fractional Anisotropy, Relative Anisotropy, Axial Diffusivity, Radial Diffusivity) or ODFs (Generalized Fractional Anisotropy).
标量指标视图允许对原始扩散加权图像(MD、ADC)、张量(分数各向异性、相对各向异性、轴向扩散率、径向扩散率)或ODF(广义分数各向异性)进行不同标量测量的计算。
MITK Tractography
Streamline Tractography纤维束追踪
This view enables streamline tractography on various input data. The corresponding command line application is named "MitkStreamlineTractography".
这个视图允许在各种输入数据上进行流线追踪。相应的命令行应用程序名为 "MitkStreamlineTractography"。
Available sections:
Input Data
Seeding
ROI Constraints
Tractography Parameters
Neighbourhood Sampling (for details see [1])
Data Handling
Output and Postprocessing
References
输入数据
播种
ROI 约束
追踪参数
邻域采样(详情见[1])
数据处理输出和后处理
参考文献
Input Data
Select the data you want to track on in the datamanager. Supported file types are:
One or multiple DTI images selected in the datamanager.
One ODF image, e.g. obtained using MITK Q-ball reconstruction or MRtrix CSD (tractography similar to [6]). Input can bes discretely sampled ODFs or an SH representation of the ODFs.
One peak image (4D float image).
在数据管理器中选择要追踪的数据。支持的文件类型包括:
在数据管理器中选择的一个或多个DTI图像。
一个ODF图像,例如使用MITK Q-ball重建或MRtrix CSD获得的(类似于[6]的追踪)。输入可以是离散采样的ODFs或ODFs的SH表示。
一个峰值图像(4D浮点图像)。
Seeding
Specify how, where and how many tractography seed points are placed. This can be either done statically using a seed image or in an interactive fashion. Interactive tractography enables the dynamic placement of spherical seed regions simply by clicking into the image (similar to [5]).
Image based seeding:
Seed Image: ROI image used to define the seed voxels. If no seed mask is specified, the whole image volume is seeded.
Seeds per voxel: If set to 1, the seed is defined as the voxel center. If > 1 the seeds are distributet randomly inside the voxel.
Interactive seeding:
Update on Parameter Change: When "Update on Parameter Change" is checked, each parameter change causes an instant retracking with the new parameters. This enables an intuitive exploration of the effects that the other tractography parameters have on the resulting tractogram.
Radius: Radius of the manually placed spherical seed region.
Num.Seeds: Number of seeds placed randomly inside the spherical seed region.
Parameters for both seeding modes:
Trials Per Seed: Try each seed N times until a valid streamline is obtained (only for probabilistic tractography).
Max. Num. Fibers: Tractography is stopped after the desired number of fibers is reached, even before all seed points are processed.
指定如何、在哪里以及放置少追踪播种点。可以静态地使用播种图像完成,也可以交互式地完成。交互式追踪允许通过点击图像来动态放置球形播种区域(类似于[5])。
基于图像的播种:
播种图像:用于定义播种体素的ROI图像。如果未指定播种掩模,则对整个图像体积进行播种。
每体素播种点数:如果设置为1,则播种被定义为体素中心。如果大于1,则在体素内随机分布播种点。
交互式播种:
参数更改时更新:当选中“参数更改时更新”时,每次参数更改都会导致使用新参数进行即时重追踪。这使得可以直观地探索其他追踪参数对生成的追踪束的影响。
半径:手动放置球形播种区域的半径。
播种点数:在球形播种区域内随机放置的播种点数。
两种播种模式的参数:
每播种点试验次数:仅对概率追踪有效,尝试每个播种点N次,直到获得有效的流线。
最大纤维数:在达到期望的纤维数后停止追踪,即使没有处理完所有播种点。
ROI Constraints ROI 约束
Specify various ROI and mask images to constrain the tractography process.
Mask Image: ROI image used to constrain the generated streamlines, typically a brain mask. Streamlines that leave the regions defined in this image will stop immediately.
Stop ROI Image: ROI image used to define stopping regions. Streamlines that enter the regions defined in this image will stop immediately.
Exclusion ROI Image: Fibers that enter a region defined in this image will be discarded.
Endpoint Constraints: Determines which fibers are accepted based on their endpoint location. Options are:
No constraints on endpoint locations (command line option NONE)
Both EPs are required to be located in the target image (command line option EPS_IN_TARGET)
Both EPs are required to be located in the target image and the image values at the respective position needs to be distinct (command line option EPS_IN_TARGET_LABELDIFF)
One EP is required to be located in the seed image and one in the target image (command line option EPS_IN_SEED_AND_TARGET)
At least one EP is required to be located in the target image (command line option MIN_ONE_EP_IN_TARGET)
Exactly one EP is required to be located in the target image (command line option ONE_EP_IN_TARGET)
No EP is allowed to be located in the target image (command line option NO_EP_IN_TARGET)
Target Image: ROI image needed for endpoint constraints.
指定各种ROI和掩模图像以约束追踪过程。
掩模图像:用于约束生成的流线的ROI图像,通常为脑掩模。离开此图像定义的区域的流线将立即停止。
停止ROI图像:用于定义停止区域的ROI图像。进入此图像定义的区域的流线将立即停止。
排除ROI图像:进入此图像定义的区域的纤维将被丢弃。
端点约束:根据端点位置确定接受哪些纤维。选项包括:
端点位置无约束(命令行选项NONE)
要求两个端点都位于目标图像中(命令行选项EPS_IN_TARGET)
要求两个端点都位于目标图像中,并且在相应位置的图像值必须不同(命令行选项EPS_IN_TARGET_LABELDIFF)
一个端点必须位于播种图像中,另一个必须位于目标图像中(命令行选项EPS_IN_SEED_AND_TARGET)
至少一个端点必须位于目标图像中(命令行选项MIN_ONE_EP_IN_TARGET)
精确地一个端点必须位于目标图像中(命令行选项ONE_EP_IN_TARGET)
不允许任何端点位于目标图像中(命令行选项NO_EP_IN_TARGET)
目标图像:端点约束所需的ROI图像。
Tractography Parameters纤维追踪参数
Mode: Toggle between deterministic and probabilistic tractography (also affects tracking prior proposals). Probabilistic ODF tractography samples the next direction from the discrete probability distribution provided by the discretized ODF. In case of probabilistic tensor tractography, an ODF is calculated from the tensor. Probabilistic peak tracking does not derive probabilities from the data but simply adds a normally distributed jitter to the proposed direction.
Sharpen ODFs: If you are using dODF or tensor images as input for probabilistic tractography, it is advisable to sharpen the ODFs (raise to the power of X). This is not necessary (and not recommended) for CSD fODFs, since they are naturally much sharper.
Cutoff: If the streamline reaches a position with an FA value or peak magnitude lower than the speciefied threshold, tracking is terminated. Typical values are 0.2 for FA/GFA and 0.1 for CSD peaks.
FA/GFA image used to determine streamline termination. If no image is specified, the FA/GFA image is automatically calculated from the input image. If multiple tensor images are used as input, it is recommended to provide such an image since the FA maps calculated from the individual input tensor images can not provide a suitable termination criterion.
ODF Cutoff: Additional threshold on the ODF magnitude. This is useful in case of CSD fODF tractography. For fODFs a good default value is 0.1, for normalized dODFs, e.g. Q-ball ODFs, this threshold should be very low (0.00025) or 0.
Step Size: The algorithm proceeds along the streamline with a fixed stepsize. Default is 0.5minSpacing.
Min. Tract Length: Shorter fibers are discarded.
Max. Tract Length: Longer fibers are discarded.
Angular Threshold: Maximum angle between two successive steps (in degree). Default is 90° * step_size. For probabilistic tractography, candidate directions exceeding this threshold have probability 0, i.e. the respective ODF value is set to zero. The probabilities of the valid directions are normalized to sum to 1.
Loop Check: Stop streamline if the threshold on the angular stdev over the last 4 voxel lengths is exceeded. -1 = no loop check.
f and g values to balance between FACT [2] and TEND [3,4] tracking (only for tensor based tractography). For further information please refer to [2,3]
Peak Jitter: Used for probabilistic peak tractography. Since no probability are provided by the data, a gausian jitter is added to the peaks. The value influences the standard deviataion of the gaussian (dir[i] += normal(0, peak_jitter|dir[i]|)).
模式:在确定性和概率性纤维追踪之间切换(也影响追踪之前的建议)。概率性ODF纤维追踪从离散的ODF概率分布中采样下一个方向。对于概率性张量纤维追踪,从张量计算ODF。概率性峰值追踪不会从数据中派生概率,而是简单地在建议的方向上添加一个正态分布的抖动。
锐化ODFs:如果你使用dODF或张量图像作为概率性纤维追踪的输入,建议锐化ODFs(提高到X次幂)。对于CSD fODFs来说,这不是必须的(也不推荐),因为它们本身就非常锐利。
截断值:如果纤维束到达的位置的FA值或峰值幅度低于指定的阈值,则追踪终止。典型值为FA/GFA的0.2和CSD峰值的0.1。
用于确定纤维束终止的FA/GFA图像。如果未指定图像,将自动从输入图像计算FA/GFA图像。如果使用多个张量图像作为输入,建议提供这样的图像,因为从单个输入张量图像计算的FA图不能提供合适的终止标准。
ODF截断值:对ODF幅度的附加阈值。这对于CSD fODF纤维追踪非常有用。对于fODFs,良好的默认值是0.1,对于归一化的dODFs,例如Q-ball ODFs,这个阈值应该非常低(0.00025)或0。
步长:算法以固定步长沿纤维束前进。默认值为0.5 * minSpacing。
最小纤维长度:较短的纤维将被丢弃。
最大纤维长度:较长的纤维将被丢弃。
角度阈值:两个连续步骤之间的最大角度(以度为单位)。默认值为90° * 步长。对于概率性纤维追踪,超过此阈值的候选方向的概率为0,即相应的ODF值被设为零。有效方向的概率归一化为总和为1。
循环检查:如果最后4个体素长度的角度标准差超过阈值,则停止纤维束追踪。-1表示不进行循环检查。
f和g值:在FACT和TEND追踪之间平衡(仅适用于基于张量的纤维追踪)。更多信息请参考相关文献。
峰值抖动:用于概率性峰值追踪。由于数据不提供概率,添加高斯抖动到峰值。这个值影响高斯的标准偏差(dir[i] += normal(0, peak_jitter * |dir[i]|))。
纤维追踪先验
Tractography Prior
It is possible to use a peak image as prior for tractography on arbitrary other input images. The local progression direction is determined as the weighted average between the direction obtained from the prior and the input data.
Weight: Weighting factor between prior and input data directions. A weight of zero means that no prior iformation is used. With a weight of one, tractography is performed directly on the prior directions itself.
Restrict to Prior: The prior image is used as tractography mask. Voxels without prior peaks are excluded.
New Directions from Prior: By default, the prior is used even if there is no valid direction found in the data. If unchecked, the prior cannot create directions where there are none in the data.
Flip directions: Internally flips prior directions. This might be necessary depending on the input data.
可以使用峰值图像作为任意其他输入图像的纤维追踪先验。局部进展方向由先验和输入数据之间的加权平均确定。
权重:先验和输入数据方向之间的加权因子。权重为零表示不使用先验信息。权重为一表示直接在先验方向上进行纤维追踪。
限制为先验:先验图像用作纤维追踪掩码。没有先验峰值的体素被排除。
从先验生成新方向:默认情况下,即使数据中没有找到有效方向,先验仍被使用。如果未勾选,先验不能在数据中没有方向的情况下生成方向。
翻转方向:内部翻转先验方向。这可能根据输入数据的不同而必要。
Data Handling数据处理
Flip directions: Internally flips progression directions. This might be necessary depending on the input data.
Interpolate Tractography Data: Trilinearly interpolate the input image used for tractography.
Interpolate ROI Images: Trilinearly interpolate the ROI images used to constrain the tractography.
翻转方向:内部翻转进展方向。这可能根据输入数据的不同而必要。
插值纤维追踪数据:三线性插值用于纤维追踪的输入图像。
插值ROI图像:三线性插值用于限制纤维追踪的ROI图像
Neighbourhood Sampling (for details see [1])邻域采样
Neighborhood Samples: Number of neighborhood samples that are used to determine the next fiber progression direction.
Sampling Distance: Distance of the sampling positions from the current streamline position (in voxels).
Use Only Frontal Samples: Only neighborhood samples in front of the current streamline position are considered.
Use Stop-Votes: If checked, the majority of sampling points has to place a stop-vote for the streamline to terminate. If not checked, all sampling positions have to vote for a streamline termination.
邻域样本:用于确定下一个纤维进展方向的邻域样本数量。
采样距离:从当前纤维束位置到采样位置的距离(以体素为单位)。
仅使用前方样本:仅考虑当前纤维束位置前方的邻域样本。
使用终止投票:如果勾选,采样点的大多数必须投票终止纤维束。如果未勾选,所有采样点必须投票终止纤维束。
Output and Postprocessing输出与后处理
Compress Fibers: Whole brain tractograms obtained with a small step size can contain billions of points. The tractograms can be compressed by removing points that do not really contribute to the fiber shape, such as many points on a straight line. An error threshold (in mm) can be defined to specify which points should be removed and which not.
Output Probability Map: No streamline are generated. Instead, the tractography outputs a visitation-count map that indicates the probability of a fiber to reach a voxel from the selected seed region. For this measure to be sensible, the number of seeds per voxel needs to be rather large.
压缩纤维:使用小步长获得的全脑纤维束可以包含数十亿个点。通过移除不真正贡献纤维形状的点(如直线上的许多点)可以压缩纤维束。可以定义一个误差阈值(以毫米为单位)来指定哪些点应该被移除,哪些不应该。
输出概率图:不生成纤维束。纤维追踪输出一个访问计数图,表示从选定种子区域到达体素的纤维的概率。为了使这一测量有意义,每个体素的种子数量需要相当大。
References
[1] Neher, Peter F., Marc-Alexandre Côté, Jean-Christophe Houde, Maxime Descoteaux, and Klaus H. Maier-Hein. “Fiber Tractography Using Machine Learning.” NeuroImage. Accessed July 19, 2017. doi:10.1016/j.neuroimage.2017.07.028.
[2] Mori, Susumu, Walter E. Kaufmann, Godfrey D. Pearlson, Barbara J. Crain, Bram Stieltjes, Meiyappan Solaiyappan, and Peter C. M. Van Zijl. “In Vivo Visualization of Human Neural Pathways by Magnetic Resonance Imaging.” Annals of Neurology 47 (2000): 412–414.
[3] Weinstein, David, Gordon Kindlmann, and Eric Lundberg. “Tensorlines: Advection-Diffusion Based Propagation through Diffusion Tensor Fields.” In Proceedings of the Conference on Visualization’99: Celebrating Ten Years, 249–253, n.d.
[4] Lazar, Mariana, David M. Weinstein, Jay S. Tsuruda, Khader M. Hasan, Konstantinos Arfanakis, M. Elizabeth Meyerand, Benham Badie, et al. “White Matter Tractography Using Diffusion Tensor Deflection.” Human Brain Mapping 18, no. 4 (2003): 306–321.
[5] Chamberland, M., K. Whittingstall, D. Fortin, D. Mathieu, and M. Descoteaux. “Real-Time Multi-Peak Tractography for Instantaneous Connectivity Display.” Front Neuroinform 8 (2014): 59. doi:10.3389/fninf.2014.00059.
[6] Tournier, J-Donald, Fernando Calamante, and Alan Connelly. “MRtrix: Diffusion Tractography in Crossing Fiber Regions.” International Journal of Imaging Systems and Technology 22, no. 1 (March 2012): 53–66. doi:10.1002/ima.22005.
MITK Reconstruction
Tensor Reconstruction
The tensor reconstruction view allows ITK based tensor reconstruction [3]. The advanced settings for ITK reconstruction let you configure a manual threshold on the non-diffusion weighted image. All voxels below this threshold will not be reconstructed and left blank. It is also possible to check for negative eigenvalues. The according voxels are also left blank.
张量重建视图允许基于ITK的张量重建[3]。ITK重建的高级设置让您可以在非扩散加权图像上配置手动阈值。所有低于此阈值的体素将不进行重建并留为空白。还可以检查负特征值,相应的体素也将留为空白。
ITK tensor reconstruction
A few seconds (depending on the image size) after the reconstruction button is hit, a colored image should appear in the main window.
点击重建按钮几秒钟后(取决于图像大小),主窗口中就会出现彩色图像。
Tensor image after reconstruction重建后的张量图像
To assess the quality of the tensor fit it has been proposed to calculate the model residual [9]. This calculates the residual between the measured signal and the signal predicted by the model. Large residuals indicate an inadequacy of the model or the presence of artefacts in the signal intensity (noise, head motion, etc.). To use this option: Select a DWI dataset, estimate a tensor, select both the DWI node and the tensor node in the datamanager and press Residual Image Calculation. MITK Diffusion can show the residual for every voxel averaged over all volumes or (in the plot widget) summarized per volume or for every slice in every volume. Clicking in the widget where the residual is shown per slice will automatically let the cross-hair jump to that position in the DWI dataset. If Percentage of outliers is checked, the per volume plot will show the percentage of outliers per volume. Otherwise it will show the mean together with the first and third quantile of residuals. See [9] for more information.
为了评估张量拟合的质量,有人提出计算模型残差[9]。这涉及到测量信号与模型预测信号之间的残差。较大的残差表明模型不充分或信号强度中存在伪影(如噪声、头部运动等)。使用该选项的方法如下:选择一个DWI数据集,估计一个张量,在数据管理器中同时选择DWI节点和张量节点,然后按下残差图像计算按钮。MITK Diffusion可以显示每个体素的残差,平均于所有体积之上,或在绘图窗口中按体积或每个体积中的每一层总结显示。点击显示每层残差的窗口,将自动使十字光标跳到DWI数据集中的该位置。如果勾选了“异常值百分比”,每体积图将显示每体积的异常值百分比。否则,它将显示残差的均值及其第一和第三四分位数。更多信息请参见[9]。
The residual widget剩余小部件
The view also allows the generation of artificial diffusion weighted or ODF images from the selected tensor image. The ODFs of the Q-Ball image are directly initialized from the tensor values and afterwards normalized. The diffusion weighted image is estimated using the l2-norm image of the tensor image as B0. The gradient images are afterwards generated using the standard tensor equation.
该视图还允许从选定的张量图像生成人工扩散加权图像或ODF图像。Q球图像的ODF直接从张量值初始化,然后进行归一化处理。扩散加权图像使用张量图像的l2范数图像作为B0进行估计。之后,使用标准的张量方程生成梯度图像。
MITK Diffusion Imaging
Visualization Control Panel可视化控制面板
ODF VisualizationODF 可视化
In this small view, the visualization of ODFs and diffusion images can be configured. Depending on the selected image in the data storage, different options are shown here.
For tensor or ODF images, the visibility of glyphs in the different render windows (T)ransversal, (S)agittal, and (C)oronal can be configured here. The maximal number of glyphs to display can also be configured here for. This is usefull to keep the system response time during rendering feasible. The other options configure normalization and scaling of the glyphs.
This is how a visualization with activated glyphs should look like:
在此小视图中,可以配置 ODF 和扩散图像的可视化。根据数据存储中选择的图像,这里会显示不同的选项。
对于张量或 ODF 图像,可以在此处配置不同渲染窗口(横断面 (T)、矢状面 (S) 和冠状面 (C))中字形的可见性。也可以配置显示的最大字形数量,以确保渲染时系统响应时间在可接受范围内。其他选项用于配置字形的归一化和缩放。
以下是激活字形后的可视化效果:
ODF image with glyph visibility toggled ON 打开字形可见性的 ODF 图像
Tensor image with glyph visibility toggled ON打开字形可见性的张量图像
Tractogram Visualization纤维束成像可视化
If a tractogram is selected in the data manager, this view enables to visualize the fibers as tubes or thick lines as well as to play with the 2D and 3D clipping of the fiber visualization. The 3D clipping works per tractogram individually, which enables nice visualization of e.g. a CST tract extending upwards out of the bottom half of a whole brain tractogram.
如果在数据管理器中选择了纤维束成像 (tractogram),该视图可以将纤维显示为管状或粗线,并可以对纤维的2D和3D剪辑进行调整。3D剪辑功能针对每个纤维束成像单独工作,这样可以实现如 CST 纤维束从整个大脑纤维束成像的下半部向上延伸的漂亮可视化效果。
CST tube visualization with sagittal 3D clipping of the whole-brain tractogram.实现全脑纤维束成像的矢状面 3D 剪辑,并特别显示 CST(皮质脊髓束)的管状可视化
标签:diffusion,Diffusion,fiber,image,mitk,图像,纤维,MITK From: https://www.cnblogs.com/cupwym/p/18329413