Ian Abenes¹, Yue Wang², Runnan Cao², Anila D'Mello^1,3,4, Shuo Wang², Jiahui Guo¹; ¹The University of Texas at Dallas, ²Washington University in St. Louis, ³The University of Texas Southwestern Medical Center, ⁴Peter O'Donnell Jr Brain Institute

Autism is a neurodevelopmental disorder associated with impairments in social interaction and communication; however, it is also accompanied by deficits in various perceptual processes, concomitant with deficits in category-selective cortex (Humphreys et al., 2008, Weisberg et al., 2012). Traditional functional localizers suffer from inefficiency, limited ecological validity, and are not patient- or kid-friendly enough. To explore advanced alternative localization paradigms, we used the hyperalignment algorithm, which functionally aligns individuals in a high-dimensional information space, to investigate whether their individualized category-selective topography can be estimated with an independent scan and from an independent group of participants. Two runs of dynamic functional localizer data from 24 autistic individuals and 24 controls were included, which incorporated stimuli of faces, objects, and scrambled objects (Wang et al., 2024). We hyperaligned participants using their connectivity (connectivity hyperalignment, CHA) or response profiles (response hyperalignment, RHA) with one run of the localizer scan and applied the derived transformation matrices to the other run. Analysis was completed using a cross-validation procedure. We estimated functional topographies using the hyperalignment predicted localizer runs (RHA/CHA) and the anatomically aligned localizer runs (AA), and correlated them with topographies based on their own localizer runs to assess prediction performance. For face- and object-selective topographies, both RHA and CHA outperformed AA. In a searchlight analysis with 15 mm radiums, we found a general improvement of RHA/CHA over AA across the whole cortex, with the most noticeable improvement in face- and object-selective cortex for their respective category. Importantly, we demonstrated that cross-group predictions (i.e., using control participants’ data to predict autism patients’ topographies, and vice versa) produced estimates similar to within-group predictions. We demonstrate that high-fidelity estimation of individualized functional topographies can be achieved using hyperalignment in autism, offering a new approach for individualized functional localization in this clinical population.