 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
Updated: copyright, 2004 |
- Expectation-Maximization Segmentation (EMS)
- Statistical Parametric Mapping (SPM)
- Analyze
- ITK-SNAP
- IRIS
- MatLab
- MrX
- GRID
Expectation-Maximization Segmentation (EMS)
EMS is a powerful suite of MatLab functions developed by Dr. K. Van Leemput that provides fully automated segmentation of brain tissues in MR Images using a geometric model and unifying statistical classification based approach. NIRL has acquired this tool courtesy of Dr. Van Leemput through Dr. Guido Gerig. EMS is a general program that we, at NIRL, have customized with subroutines and applications specific to our study projects.
Statistical Parametric Mapping (SPM)
The SPM programs are a suite of MatLab functions developed by The Wellcome Trust for interpretation of neuroimaging data. SPM uses a voxel based approach to spatially normalize and smooth neuroimages into a standard space.
The Analyze Program was developed by Biomedical Imaging Resource at the Mayo Foundation and used for display, manipulation and measurement of multidimensional biomedical images. This program can be used on UNIX workstations as well as high powered PCs.
ITK-SNAP is a software application used to segment structures in 3D medical images. It provides semi-automatic segmentation using active contour methods, as well as manual delineation and image navigation. In addition to these core functions, SNAP provides a number of supporting utilities.
Iris is a 3D interactive segmentation tool developed by Dr. Guido Gerig’s group. 3D structures are segmented by drawing regions slice by slice in arbitrary slice directions. The segmented regions are simultaneously displayed in all the three orthogonal cuts. A 4th window displays the segmented object as a 3D graphical rendering. Full 3D navigation is possible by linked cursors in the three orthogonal cuts and the 3D window. Points and lines can be painted on the object surface by a newly developed 3D painting option. These marked points are simultaneously labeled in the data volume. This interactive segmentation tool is platform independent and runs on UNIX and PC computers.
The MrX software was developed by GE Research and Development Center in Schenectady, NY, and has been modified by Brigham and Women's Hospital in Boston, MA, as well as by NIRL.
MrX uses dual contrast data to provide a feature space labeling of pixels for subsequent segmentation using the cluster analysis (KNN algorithms). The K nearest neighbor (KNN) algorithm is a very simple, yet relatively powerful, technique. To classify a new example, the K closest training examples are identified, and the most common class amongst these K neighbors is chosen as the classification for the new item (i.e. a majority vote is taken). More formally, the maximum likelihood estimators of the probability that an item lying in a small volume of variable space (with radius equal to the distance between the point to be classified and the Kth nearest neighbor) belongs to each class is calculated. This is done by considering the number of training examples of each class in this volume. The novel point is then classified as belonging to the class with the largest estimated probability.
GRID is an image analysis program developed by programmers at NIRL. This program uses the point counting method for volumetric measurements of various neuro-anatomical regions of the brain. The program uses stereologic and Cavalieri principles to estimate volumes which is particularly suited for MRI since sample slices are obtained at regular intervals. This method is simple and efficient.
Using the GRID interface, the analyst can load a 2D or 3D image series of the brain which may contain T1, T2 and/or Proton Density images. These original series of images can be realigned and the grid size can be set based on various study specific protocols. A new realigned series is created, and an ROI can be processed.