A method for analyzing brain images includes receiving a plurality of three-dimensional image datasets comprising a plurality of voxels for a corresponding plurality of subjects; dividing the plurality of three-dimensional image datasets into at least a first and a second group; dividing the first group into A1 to An subgroups; dividing the second group into B1 to Bn subgroups; determining statistical individual variability between datasets selected from the A1 to An subgroups and the B1 to Bn subgroups; determining statistical individual variability between datasets in the first and second groups responsive to the statistical individual variability between datasets selected from the plurality of three-dimensional image datasets in the A1 to An subgroups and the B1 to Bn subgroups; and for each of the first and second groups, generating a reliability map comprising a map of probabilities that a voxel in an image dataset of the respective first and second groups satisfies a predetermined statistical threshold.