Noninvasive selective activation and suppression of the human brain with natural images
Name
zahorodnii_zaho_meng_bcs_thesis_2026.pdf
Size
13.86 MB
Format
Adobe PDF
Checksum (MD5)
531b22f0dd38c0896721917dc1e3d2a1
Author(s)
Zahorodnii, Andrii
Advisor(s)
Fiete, Ila
Date Issued
February 2026
Publisher
Massachusetts Institute of Technology
Abstract
The ability to selectively modulate neural activity in the human brain has profound implications for both basic neuroscience and clinical applications. However, approaches for achieving neural control in humans remain limited. Here we demonstrate that deep neural network models trained on intracranial recordings can predict which natural images will drive or suppress activity at specific cortical sites in the human brain. We recorded stereotactic electroencephalography (SEEG) from epilepsy patients viewing thousands of images in a rapid serial visual presentation experiment, establishing the first human intracranial dataset designed for model-based stimulus optimization. Using pretrained image models combined with linear regression, we identified visual stimuli that successfully modulated neural activity in predicted directions at targeted recording sites. We found that visual reliability was highest in occipital and temporal regions, though neural target stability varied substantially across recording days. Our systematic analysis of scaling laws revealed that model performance improves approximately linearly with the logarithm of training set size, with stimulus diversity contributing more than repeated presentations for a fixed experimental budget. Models improved iteratively when retrained with feedback from previous sessions, demonstrating successful closed-loop refinement of model predictions. Our framework extends prior work in non-human primates to human intracranial electrophysiology, establishing a proof-of-concept for stimulus-based neural control in humans. These results open new avenues for non-invasive therapeutic interventions for neurological disorders and for causal testing of computational theories of human sensory processing.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
