Using cosine similarity (cos), we found that homotopic patches appeared more similar than randomly selected patches (\(cos=0.45\pm 0.28\) for homotopic versus \(cos=0.00\pm 0.29\) for heterotopic random patches, \(p<0.001\), \(d=1.57\)). FCD patches were more similar to each other (\(cos=0.29\pm0.29\)) than to random HV cortical patches (FCD vs heterotopic patches \(cos=0.006\pm0.29\), \(p<0.001\), \(d=0.97\)), but were quite similar to their underlying homotopic regions in HVs (FCD vs homotopic patches \(cos=0.24\pm0.31\), \(p=1.78\), \(d=0.17\)), presumably with more subtle lesions having a fairly typical appearance for their given location.
As expected, therefore, following local normalization patches in the outlier ROIs appeared more "typical" for their location, now with a similar Mahalanobis distance to randomly selected cortical patches (random cortex MD = \(1.86\pm0.52\) versus precentral MD = \(1.92\pm0.52\), \(d=0.10\), and insula MD = \(1.59\pm0.36\), \(d=0.63\)). In contrast, following local normalization, FCDs remained as significant outliers (MD = \(3.55\pm0.81\)) compared to both normal cortex and the previously outlying ROIs (versus normal cortex \(d=2.81\), precentral \(d=2.78\), insula \(d=4.09\)). Cosine similarity between FCD patches also became significantly higher (\(0.40\pm 0.26\)) than similarity to non-lesional patches at any location (to homotopic patches \(cos=0.00\pm 0.30\), \(p<0.001\), \(d=1.40\), and to heterotopic patches \(cos=0.00\pm 0.28\), \(p<0.001\), \(d=1.44\), with no difference between the two, \(p=0.54\), ) (Figure S3). Local normalization, therefore, not only aids with global outlier detection by decreasing the "outlierness" of normal cortical regions, but also with local outlier detection, by further distancing FCDs from the expected appearance in their homotopic regions.