Extended Data Fig. 1: A flowchart of systematic characterization of heterogeneity in brain atrophy patterning.

(a) A total of cross-sectional MRI from 2170 individuals (1124 patients with schizophrenia) was used to characterize heterogeneity in brain atrophy patterning of schizophrenia. (b) Brain images were processed using voxel-based morphometry. GMV was extracted from ROIs based on the Automated Anatomical Labeling (AAL) atlas and adjusted by regressing out the effects of sex, age, the square of age, TIV and site effects. (c) Adjusted GMV values were normalized relative to control population using z scores. Higher z scores represent larger deviations from the normal (that is, more severe atrophy in patients with schizophrenia). (d) Brain pathophysiological model (that is, SuStaIn [31]) requires both spatial (brain regions) and temporal (z scores representing advancing atrophy severity) features as input (that is, an M × N z score matrix). Here, N represents the number of individuals with schizophrenia (N = 1124 in this study). M represents the number of ROIs (M = 17). (e) SuStaIn was used to identify diverse but distinct patterns of progression using cross-sectional neuroimaging data and to cluster individuals while accounting for disease progression. (f) Individuals with schizophrenia were classified according to the sequence of atrophy in different brain regions. For each subtype, brain-based staging was assessed from progressive spatial patterns with distinct origins. (g) Using a longitudinal sample, we examined whether subtype classification based on baseline brain features predict differential treatment response to antipsychotic medications and TMS.