ISSN – Individual Structure Spatial Normalization
Individual structure spatial normalization (ISSN) is a software tool for analyzing and adjusting spatial (position and orientation) and structural (size and shape) properties of brain structures. This software analyzes regions of interest (ROIs) of brain structures, which can range from whole brain to small structures such as the hippocampus. It works with ROIs formulated and saved using Mango. The ROIs can be generated in a variety of ways ranging from manual entry using Mango's ROI entry tools to Mango's automated shrink-wrapping of pre-segmented structures.
ISSN uses the geometric center of a structure's ROI as it position. Orientation is determined as the orientation of the structures three principal axes using principal axes analysis. Size can be assessed in several ways. The simplest is the volume of the ROI. Another would be surface area for a surface formed about the ROI using Mango surface extraction and measurement capabilities. Surfaces of structures created by Mango can be exported for extensive analysis by other software. Finally, size can be assessed using the magnitude of the three eigenvectors used in representing the three principal axes. The square root of the magnitude of these eigenvectors is a measure of the average distance from the center to the surface along each of the three areas. Shape can be roughly assessed as the anisotropy of these three sizes.
ISSN will perform analyses and generate a spreadsheet tabulating measurements as well as group level statistics. For group analyses all ROIs need to be in the same directory as the brain images from which they were made. Images should be stored in NIFTI format. ROIs and images need to have corresponding names so that they can be properly matched. An example might be as follows:
| Image filename | ROI filename |
| I1_whole_brain.nii | I1_L_caudate.roi.gz |
| I2_whole_brain.nii | I2_L_caudate.roi.gz |
| I3_whole_brain.nii | I3_L_caudate.roi.gz |
| I4_whole_brain.nii | I4_L_caudate.roi.gz |
| ... | ... |
| I40_whole_brain.nii | I40_caudate.roi.gz |
To run ISSN, the operator must select the folder containing the original images and ROIs determined from them. A label is required to identify which ROIs you want to use. The ROI label is a unique portion of the ROI's filename, such as "L_caud" in the example above. Once the directory and label are entered, ISSN indicates the number of matching files found (40 in this example).
Many processing options are available. The simplest is to uncheck the Translation, Rotation, and Scale options and only check Result Spreadsheet for output. This will output a spreadsheet with full details of the group analysis of the ROIs.
Results Spreadsheet. This spreadsheet is stored in a file in the image directory using a name that indicates the ROIs label, such as "results-L_caud.xls" from the above example. There are four tabs in the spreadsheet with the following information for each subject:
- Results Tab. x-y-z coordinates of geometric center, magnitude of the three eigenvalues (V1<V2<V3), x-y-z components of each of the three eigenvectors, and additional info in the future.
- Sense Corrected Tab. Some eigenvectors have the wrong sense so we correct for this and show the sense corrected eigenvectors at this tab.
- Covariance Tab. Covariance matrix data along with an assessment of skewness.
- Excluded Tab. In some cases a structure might not be consistent with others in the group due to a major difference in shape. This can dissociate V1-V3 order so we exclude such structures from the group.
Results, Sense Corrected, and Covariance tabs provide group summary statistics for each measure.
ISSN's Options for Spatial Normalization. ISSN can apply an affine transform to correct each structure within a group for differences in position, orientation, and size. ISSN formulates group specific standards for these so that average position, orientation and size are not affected. ISSN can do this with two types of scaling which can be selected if the Scale option is checked. Shape preserving scaling uses isotropic scaling (x, y,and z scaling are identical) and these are determined to adjust volume to the group mean volume. Shape standardizing determines x, y and z scale factors to adjust dimensions along principal axes to group mean average dimensions. Shape standardizing scaling will also preserve group mean volume.
Output options for spatial normalization support making transformed and average brain images (useful when the ROI is for the whole brain). For individual structures the transformed ROIs and Incidence Map options are quite useful.
Users are invited to try different processing and output options.