froi: fs-fast roi

sections

other documentation

html manual :: introduction

Freesurfer and FS-FAST (Freesurfer Functional Analysis Stream) are software suites developed at the Massachusetts General Hospital imaging center for the analysis of functional MRI (fMRI) data. **insert freesurfer and fs-fast refs** FS-FAST is a set of tools for the analysis of slice-based fMRI data, including pre-processing (motion correction, image normalization, spatial smoothing), design specification (including gamma and finite impulse response (FIR) models), contrast specification, and statistical map creation. In a complimentary fashion, Freesurfer creates 2D surface- and 3D volume-based reconstructions of the brain. Data from the slices can then be mapped onto a previous- or current-session anatomical, or on a 2D surface, such as an inflated or flattened representation. Surfaces can also be registered into spherical and Talairach coordinate systems.

The advantages of FS-FAST and Freesurfer over other fMRI analysis packages are numerous. The tight integration of the two complimentary suites provides for relatively easy transformation of data from one representation to another. The majority of the programs are command-line based, allowing for easy scriptability as well as transparency of the analysis process for the user, something that is sometimes difficult with other packages. With FS-FAST and Freesurfer there is no need for an individual subjects brain to be automatically smoothed and transformed into Talairach space; all analyses can be done in the subjects native slices.

However, FS-FAST is lacking one major feature, that being the analysis of regions of interest (ROIs). ROI analyses are an important part of functional MRI research. **insert old nancy refs** An ROI approach proceeds as follows:

  1. Define a functionally-specific region by a localizer scan. For example, you might be interested in areas that respond more to faces than to objects, perhaps the fusiform face area (FFA) **insert FFA refs**
  2. Perform an experiment that tests your condition of interest; for example, you might be interested in whether or not the FFA responds different to upright faces versus inverted faces.
  3. Using the ROI defined in 1), look to see if there are any differences between conditions in 2).

The advantages of an ROI approach are numerous. For one, because each subjects anatomy is different, anatomical ROIs are usually not sufficient for most analyses. Defining an independent functional ROI allows your analysis to be tailored to the idiosyncrasies of the particular subjects brain. As well, since you are only looking in a small region, you do not have the problems associated with whole brain analyses. ** are there problems? i think there are... **

froi

To this end, I wrote a suite of programs to do functional ROI analyses in conjunction with FS-FAST and Freesurfer called froi: FS-FAST ROI. The meat of the suite is to allow one to functionally define ROIs and then extract information from the ROIs, such as mean time-course (for blocked data) as well as mean and percent signal change for each condition (along with the associated standard deviations and errors). ROIs are saved in a space-efficient format and can be edited quite easily. The results of an ROI computation can be viewed using a Matlab GUI and are saved in a Matlab structure. In addition, there is an option to create tab-delimitted text files of the results for importation into other programs like Excel or SPSS.

The functional definition of an ROI can be done in three ways: by t value, by significance (sig) value, and by false discovery rate (FDR) **insert FDR refs**. Defining an ROI by t or sig value is done in a simple threshold fashion; all values greater (or lower, in the case of negative values) are taken to be in the ROI. Using FDR you provide a specific false discovery rate and a hypothesis of how you think the noise is correlated, and all voxels above the FDR-calculated p value are included in the ROI.

Such ROIs should not be taken verbatim into the subsequent statistical analysis, however. Many times the ROIs will include scattered and non-contiguous voxels throughout your slices, or may contain multiple functional regions that you would like to analyze separately. After masking the image with a t, sig, or FDR threshold, you need to edit the thresholded map in order to define the appropriate ROIs. Once these ROIs are created, you are ready for further statistical analysis.

Data analysis within the ROI uses the results of the selxavg program of FS-FAST. selxavg computes the estimated deviation from baseline for each condition for each voxel in your volume. Depending on the type of analysis (either gamma-fit or FIR), you may have only one point for each condition (gamma) or a time-course (FIR). ROI analyses proceeds by averaging these data over your ROI and computing the appropriate standard deviation and standard error. These results can then be viewed or taken to a third level of analysis, such as a group analysis.

froi provides many other ROI functions. For example, it can combine multiple ROIs (up to three) by intersection or union. As well, it can create an overlap map of ROIs (up to three). Finally, it can create a selectivity map for one condition versus another, thus allowing you to see how the selectivity of voxels varies within and across slices.

Slice-based ROIs provide a simple solution to the problem of converting your data to another coordinate system: there is no problem. However, there are some times when you might want to view an ROI in a different coordinate system; for example, to look at stability across sessions or to perform a group analysis. froi provides scripts to simplify these tasks. It can take an ROI and convert it into a Freesurfer label while also taking a label and converting it to a froi ROI. As well, it can combine labels in much the same way as you can combine ROIs; however, as of the current release, it only works for two labels and only for the intersection method.