Abstract
Contour integration, the process of joining spatially separated elements into a single unified edge, has consistently been found to be impaired in schizophrenia. Whether contour integration is impaired among patients with psychosis more broadly remains unclear. We examined a transdiagnostic sample of 34 participants with psychosis, 24 first degree biological relatives without psychosis, and 24 healthy controls with no family history as a part of the Psychosis Human Connectome Project. We used 7 tesla functional MRI (fMRI) to measure responses across visual regions-of-interest (ROIs), including primary visual cortex (V1), lateral geniculate nucleus (LGN), and lateral occipital complex (LOC), during a contour integration task. Across all participants, univariate V1 fMRI responses were lower for aligned versus scrambled contours, consistent with previous studies of predictive coding. Participants with psychosis had stronger fMRI responses in LOC for contours presented within a field of background elements, also consistent with prior studies. We performed a representational similarity analysis (Kriegeskorte, 2008) to quantify patterns of responses across voxels within ROIs in individual subjects. Across V1 and LOC, similarity was highest for task conditions with the same configuration of background elements, indicating that the presence or absence of background elements had a strong effect on the fMRI response patterns in these regions. Additionally, in both V1 and LOC, highest similarity was seen between task conditions with a discernible contour. Patients with psychosis showed lower similarity in V1 between same-background conditions, as compared to healthy controls, consistent with disorganized contour processing among patients. Our results are consistent with prior findings of impaired contour integration in schizophrenia, and suggest abnormalities in the multivariate pattern of neural responses in early visual cortex among patients with psychosis.