We propose an automated method to segment cortical necrosis from brain FLAIR-MR Images. as cortical necrosis if they are in the cortex FJX1 and have the intensity profile of CSF. We evaluate our method by using a set of 72 healthy subjects to model cortical variation.We use this model to successfully detect and segment cortical necrosis in a set of 37 patients with CVD. A comparison of the results with segmentations from two impartial human experts shows that the overlap between our approach and either of the human experts is in the range of the overlap between the two human experts themselves. where indexes the normal subjects in our database. The preprocessing done before the registration involves skullstrip-ping with BET [6], bias correction with N3 [7] and 12-dof linear registration to template using FLIRT [8]. For a given subject scan in which we wish to detect and segment cortical necrosis, we repeat the preprocessing and the non-linear registration in a similar manner. We denote the Jacobian determinant obtained from the subject scan as 𝒥t. It is important to note that both and 𝒥t are in the template space. 2.2. Extraction of abnormal cortical regions using the Jacobian deformation maps We now use and 𝒥t to pinpoint cortical abnormalities. If the high values of Jacobian determinants generated above were caused exclusively by the presence of cortical necrosis, our problem would be solved at this stage. In practice this does not happen. The extreme variation of sulcal anatomy results in the presence of subject AMG-073 HCl specific false positive spikes in the value of the Jacobian determinant at several cortical locations. AMG-073 HCl We deal with these false positives using the framework described in the following section. The motivating idea is that if we have a large enough set of normal subjects, such false positives will occur in one or a few of our normal subjects. This information can then be used to eliminate such false AMG-073 HCl positives whenever they occur in subject scans. In broad terms we want to detect cortical abnormalities from 𝒥t while modeling cortical variation using are wavelet coefficients sorted in descending order, are the corresponding wavelet basis images (functions), and is the dimensionality of 𝒥 (the number of voxels in the image). In practice a large proportion of the ordered set are very close to zero. This allows us to choose the largest of them and write ? 5:6 106 and we use = 10000. We generate compressed wavelet representation for the Jacobian map(𝒥t) corresponding to the subject scan: are selected based on the subject data only. The exact same basis are then used to generate a compressed representation of each normal Jacobian map is the mean of the rows of A. For the rest of this paper we denote the contains cortical as well as subcortical abnormalities. Since in this paper we focus on cortical abnormalities, we mask out the subcortical regions of using a precomputed binary mask based on the template. is now mapped back to the subject space to get using a precomputed deformation field. is shown in Figure 2. Fig. 2 From left to right, the original FLAIR-MR scan, the Jacobian abnormality map the value of at location . is an abnormality threshold. We consider that values higher than the threshold may indicate necrosis. If 0 then the AMG-073 HCl method flags the subject as containing a cortical necrosis. In large studies the thresholds and 0 can be determined from a small subset of the data for which ground truth is known. 2.4. Segmentation of cortical necrosis The Jacobian abnormality maps give us a coarse delineation of necrosis (Figure 2). In general, they pick up large regions around the cortical necrosis of interest. For the segmentation task, we combine the Jacobian abnormality map with an.