Face Decoding and Reconstruction from 7T Laminar fMRI Data using A Diffusion Generative Model

Motivation: Face recognition and perception are vital for social interactions, yet its neural mechanisms, including the interplay between feedforward and feedback processes, remain poorly understood. Goal(s): Our goal is to use ultrahigh field fMRI data to investigate the contributions of different cortical layers to face perception in the visual cortex of a subject engaged in a naturalistic viewing paradigm. Approach: We employed a diffusion generative model to reconstruct video scenes, particularly those featuring human faces, from laminar brain activity. Results: The number of accurately reconstructed face images suggests that the superficial and deep layers across the visual areas significantly contribute to face recognition. Impact: Brain disorders, like stroke and prosopagnosia, can impair brain regions for facial processing, making face perception difficult. By understanding the neural circuitry involved in face perception, researchers may identify pathways that could be targeted to alleviate symptoms in these conditions