Brief

Researchers have developed a new deep learning model called Brain2Text that can decode fMRI signals into textual descriptions of viewed natural images. This model, trained without visual input, achieves state-of-the-art semantic decoding performance by generating meaningful captions that capture the core semantic content of complex scenes. Neuroanatomical analysis using this framework highlights the involvement of higher-level visual cortices and specific semantic dimensions like animacy and motion in visual processing. The work offers a novel methodology for understanding the neural basis of semantic processing and could inspire the development of brain-inspired language models. AI