FreeSurfer

FreeSurfer is a set of software tools that can be used to process structural (T1-weighted) and functional MR images. The tools can construct models of the white matter, cortical gray matter, as well as the pial surface. Once these surfaces are known, an host of anatomical measures becomes available, such as cortical thickness, surface area, curvature, and surface normal at each point on the cortex. Tools also exist to inflate or flatten for further measurement or visualization. Much of the FreeSurfer pipeline is automated, thereby making it ideal for processing large data sets.

The FreeSurfer source code has been released under an open-source licensing agreement. Making the FreeSurfer source code available to the public has many advantages: users can gain a greater understanding of the underlying algorithms; users can build FreeSurfer on currently unsupported OS platforms; and users can inspect the code for possible bugs, or suggest improvements, thus improving the overall quality. Approximately 2500 FreeSurfer licenses have been distributed.

Now that FreeSurfer is open source, commercial-level support and quality assurance is being provided. Currently there are 407 unit tests, 64 system tests, nightly builds/tests on an array of platforms, and integration of the Doxygen documentation system for FreeSurfer documentation.

More information on FreeSurfer can be found on the FreeSurfer wiki pages here and the download page is located here.

DTI Studio

DTI Studio is a Windows-based software package, developed at Johns Hopkins University for the processing and visualization of diffusion tensor data. The package supports a number of different image data types from Philips PAR and REC files to DICOM.  DTI Studio can be obtained from the S. Mori group website here. The package is supported through a user list here as well as by the developers (dtistudio@mri.jhu.edu).

The Center for Imaging Science (CIS) at Johns Hopkins University (JHU) has been successful in providing Shape Analysis tools to the BIRN community. These tools are able to compare complex-shaped brain structures and detect differences over the entire surface of the structure. Large scale shape analysis processing has been established through the use of Large Deformation Diffeomorphic Metric Mapping (LDDMM) and the use of the TeraGrid. The JHU group has been optimizing the LDDMM semi-automated shape analysis (SASHA) pipeline and it is now ready to move into production mode to accommodate additional datasets. A list of current LDDMM projects and their status is shown below. This group has, contributed significant information to mBIRN and the BIRN-CC on using the SRB to perform large computations. To date there have been 40,800 LDDMM-volume processed jobs using 244,824 cpu/hrs of processing (~27 cpu/yrs) and requiring over 40 TeraBytes of storage. To accommodate data sets of this size, the group used storage on the TeraGrid (GPFS-WAN 220TB), the SRB, and local filesystems. In addition, they have implemented a system using sshfs that essential "glues" the GPFS-WAN and local file system together to make working with data distributed across all of the storage sites a seamless experience. Work is beginning on performing LDDMM shape analysis on the OASIS dataset and the derived results will be included with the OASIS data in the BIRN data repository.

Table 1. Status of LDDMM shape analysis projects to date at JHU. This list includes projects outside of BIRN collaborations as well as collaborations with the Mouse BIRN.

Caption: Average shapes of the Alzheimer's patients (purple) and the population of normals (colored vectors). The color and length of the vector corresponds to the amount of displacement needed to match the corresponding points on each surface. LDDMM was used to generate the vector field.

Spherical Harmonic Shape Analysis: This is a collaboration between the Duke group and the UCLA group to characterize hippocampal shape in various cohorts. It is a prototype study using BIRN infrastructure: the original data is stored in an instance of XNAT, the data is exported to the LONI pipeline for shape analysis and the derived data is imported back into the instance of XNAT. Briefly, the SPHARM project is closely related to the LDDMM work at JHU. LDDMM is a sophisticated approach to volume registration that has been used for hippocampal shape analysis. Duke has several studies (late-life depression, bipolar disorder, childhood abuse and neglect) for which the hippocampus volume is hand-segmented. The group at Duke has been collaborating with JHU through the BIRN for volumetric hippocampal shape analysis. An alternate, less computationally intense approach to shape analysis applies surface registration using the surface point distribution functions rather than volume registration. This approach has been used by a collaboration between Duke and UNC (Dr. Guido Gerig and Dr. Martin Styner). The hippocampal shapes are expressed in terms of spherical harmonic functions (SPHARM) which then allow a uniform set of descriptors of the shape. Dr. Gerig and Dr. Styner are both involved in NAMIC and there is an effort to add such surface registration and SPHARM functions to 3D Slicer. Kurt Zhao at Duke is evaluating the SPHARM analysis results and comparing it to results obtained using volume registration (LDDMM analysis through the portal). The researchers are experimenting with projecting the statistical maps on a representative member of the population to show the head and tail of the hippocampus as well as the alveus, subiculum and CA1 regions. Preliminary results from this work are below.

Caption: A p-value map projected on a hippocampal average shape of the areas that have statistically significant hippocampal shape differences between elderly depressed subjects and age matched non-depressed subjects.

BELL (BIRN Evaluation of Late-life Lesions) Project: The goal of this project is to use the expertise of Johns Hopkins University in white matter tract atlases and MGH in image registration to analyze the damage done by white matter lesions to specific white matter tracts and to examine the lesion correlations with clinical measures of depression. Kurt Zhao of Duke is collaborating with Steve Pieper at BWH to implement a new non-linear approach for white matter tract registration. The registration tool is implemented in a program called "Rview" developed by Dr. D. Ruckert and provided by Dr. Gerig at UNC as well as the Large Deformation Diffeomorphic Metric Mapping (LDDMM) tool in collaboration with Dr. Miller's group at JHU.

Distortion corrections in diffusion MRI using LDDMM: A significant source of error in brain image data is spatial distortions due to inhomogeneities in the main magnetic field (B0), which can arise due to imperfect shimming, magnetic susceptibility effects and chemical shift. These effects can become particularly important for longitudinal studies in which different shim settings can result in substantially different distortions between scan sessions. Although these distortions are most pronounced in functional or diffusion-weighted imaging using echo planar imaging (EPI) sequences, the effect can be significant (on the order of several millimeters) even in conventional structural images.

A method to correct B0- and diffusion-weighting distortions is to apply a non-linear transform to the individual DWI's and register them to a non-diffusion-weighted image (normally a b=0 image). Susumu Mori's group at JHU has implemented a method of image distortion correction using the Large Deformation Diffeomorphic Metric Mapping (LDDMM) algorithm. This method was first implemented using landmarks identified by a user, but now has been updated as an automated method. Results comparing these two methods to a standard non-linear transform implemented in the Statistical Parametric Mapping package are shown below. This work has been submitted and is currently under review.

Demonstration of efficacy of the LDDMM-based non-linear registration method. The automated and landmark-based LDDMM methods are compared with correction using information from the measured B0 field and also to the results from a non-linear registration algorithm implemented in the Statistical Parametric Mapping (SPM) processing package. The red overlay is generated from the undistorted T2-weighted image. The automated LDDMM-based method gives the best registration.

AVID

Morphometry BIRN members at Duke have developed a computer program for using fMRI activated regions to guide DTI fiber tracking called AVID. This software package will be made available to the neuroscience community upon completion of testing.