Unleashing the Power of AI Eyes: A Neuroscientific Revolution

Dive into the world of cutting-edge neuroscience as researchers at Weill Cornell Medicine harness AI-selected images to decode the secrets of our visual processing. Discover how AI and fMRI collaborate to uncover the mysteries of our brain's visual landscape.

Scientists at Weill Cornell Medicine, Cornell Tech, and Cornell's Ithaca campus have taken a fascinating step in understanding how our brains process visuals. They've employed a unique method using AI-selected natural images and AI-generated synthetic images to explore the visual processing areas of the brain. The primary aim is to unravel the organization of vision through a data-driven approach while minimizing biases that may arise when using a more limited set of images chosen by researchers.

In their study, published on Oct. 23 in Communications Biology, volunteers were shown images selected or generated by an AI model of the human visual system. These images were expected to strongly activate various visual processing areas. By using functional magnetic resonance imaging (fMRI) to monitor brain activity, the researchers discovered that these images significantly activated the target areas compared to control images.

Moreover, the researchers demonstrated the capability to fine-tune their vision model for individual volunteers using the data from the image responses. Images generated to be maximally activating for a specific individual outperformed those generated based on a general model.

Dr. Amy Kuceyeski, a professor at Weill Cornell Medicine, expressed optimism about this novel approach, stating, "We think this is a promising new approach to study the neuroscience of vision."

The collaborative effort involved Dr. Mert Sabuncu from Cornell Engineering and Cornell Tech, with Dr. Zijin Gu serving as the first author of the study. To build an accurate model of the human visual system, the researchers utilized a dataset of tens of thousands of natural images and corresponding fMRI responses. They trained an artificial neural network (ANN) to model the human brain's visual processing system and used it to predict images that would maximally activate targeted vision areas.

The study, conducted with six volunteers, demonstrated that both natural and synthetic images predicted to be maximal activators significantly activated the targeted brain regions. This supports the validity of the ANN-based model, indicating that even synthetic images can be valuable for testing and improving such models.

In a subsequent experiment, the researchers created individualized visual system models for each subject based on the data from the first session. This personalized approach showed greater activation of the targeted visual region for synthetic images compared to those based on a group model, suggesting the potential for AI and fMRI in individualized visual system modeling.

The researchers are now exploring similar experiments using an advanced image generator called Stable Diffusion. This innovative approach could also be applied to studying other senses, such as hearing. Dr. Kuceyeski envisions exploring the therapeutic potential of this method, suggesting the possibility of altering brain connectivity using specifically designed stimuli to address conditions like excess anxiety.