The activity of ROI's is extracted from the activation maps (maybe more than one map). This reduces data set considerably and can be later analyzed with conventional methods of multivariate analysis e.g. ANOVA, ANCOVA 
This is still the best method for reliable assessment of significance of fMRI findings.

• Regions Of Interest can be defined :
• by hand drawing on anatomical images using ADOBE Photoshop, this method is most precise but very labor intensive
• by choosing regions in the uniform Talairach space, in this case regions are defined as composed of small 8x8 mm rectangular "bricks" using ROI3Ddef (this method is carrying an error due to misregistration in the Talairach space, the error in location can approach 10 mm).

DoRegions program used to calculate ROI measures can create various intensity measures (calculated by the get_region3D function) :

NOTE :two statistical maps maybe used in the calculation of ROI mesures: one "tcut_map" is used only to determine if voxel is activated for given threshold and cluster filter, the other "main_stat" is describing the intensity of activation, one may use the same statistics for both purposes, or e.g. use t-statistics to decide which voxels are activated, but raw signal change to measure the intensity of activation.

• n - number of "active" voxels
• s - sum of intensities of active voxels (intensity is defined by variable main_stat while active status by tcut_stat, those maybe two different maps, e.g. t-statistics and signal change)
• r = s/n
• R - dispersion of intensity for activated pixels (mean distance form the center of mass.
• q = n/ROIsize*image_size (n normalized by region size)
• Q = n/total_n/ROIsize*image_size (n normalized total number of activations in the whole brain)
• Q1 = n/total_n*image_sizeÝ (n normalized total number of activations in the whole brain)
• Q2 = s/total_n/ROIsize*image_size (s normalized by region size and total number of activations in the whole brain)
• Q3 = s/total_n*image_sizeÝ (s normalized by region size and total number of activations in the whole brain)
• Q4 = s/ROIsize*image_size (s normalized by region size)
• Q5 = n/total_n/ROIcount*image_sizeÝ (n normalized by region size and total number of activations in the whole brain, here ROIsize is replaced by number of voxels in the ROI that actually carry signal, are not equal to zero)
• Q6 = s/total_s/ROIcount*image_sizeÝ (s normalized by region size and total number of activations in the whole brain, here ROIsize is replaced by number of voxels in the ROI that actually carry signal, are not equal to zero)
• x,y,z - coordinates of the center of mass of activations within ROI

Constable RT, Skudlarski P, Mencl E, et al. Quantifying and comparing region-of-interest activation patterns in functional brain MR imaging: Methodology considerations   MAGN RESON IMAGING 16: (3) 289-300 APR 1998
 
• for maximum power ROI should match the size of activated region,
• none of the above measures can be used to compare activation between ROIs of various sizes,
• s - measure used with very low threshold (even as low as 0) is most powerful,
• ROI analysis is reliable for comparing of the same ROI between various tasks and subject groups, it is much more prone to false activations in comoparisons between different ROIs, even of similar size.