Unlocking the Secrets of the Mind: AI's Groundbreaking Impact on Brain Imaging Technology
Published on: March 10, 2024
Researchers from Weill Cornell Medicine, Cornell Tech, and Cornell's Ithaca campus have embarked on an innovative journey using artificial intelligence to explore the mysteries of the human brain's visual processing areas. In a study published on October 23 in Communications Biology, they harnessed AI-selected natural images and AI-generated synthetic images as neuroscientific tools to delve into the brain's visual processing mechanisms. The goal was to employ a data-driven approach to comprehend the organization of vision while mitigating potential biases that can arise when examining responses to a limited set of researcher-selected images.
Volunteers participated in the study, where they viewed images chosen or generated based on an AI model of the human visual system. These images were predicted to strongly activate various visual processing areas. The researchers employed functional magnetic resonance imaging (fMRI) to record the volunteers' brain activity. The results revealed that the images significantly activated the target areas more effectively than control images.
Furthermore, the researchers demonstrated the potential for personalizing their vision model for individual volunteers using image-response data. Images generated to optimally activate specific individuals outperformed those generated based on a generalized model.
Dr. Amy Kuceyeski, a professor at Weill Cornell Medicine, stated, 'We think this is a promising new approach to study the neuroscience of vision.'
The study was a collaborative effort with Dr. Mert Sabuncu, a professor at Cornell Engineering and Cornell Tech. Dr. Zijin Gu, a doctoral student co-mentored by Dr. Sabuncu and Dr. Kuceyeski at the time of the study, served as the study's first author.
Mapping and modeling the human visual system, especially by understanding brain responses to specific images, is a significant challenge in modern neuroscience. Researchers aim to achieve this goal mainly through non-invasive methods, such as fMRI, given the complexities of recording brain activity directly. The researchers trained an artificial neural network (ANN) using a dataset of tens of thousands of natural images and corresponding fMRI responses. The ANN was designed to model the human brain's visual processing system and predict images that would maximally activate targeted vision areas.
The results indicated that both natural and synthetic images, predicted to be maximal activators, activated the targeted brain regions more significantly than a set of average activator images. This validates the ANN-based model's effectiveness and suggests that even synthetic images hold value as probes for testing and improving such models.
In a subsequent experiment, the researchers utilized the image and fMRI-response data from the first session to create personalized ANN-based visual system models for each of the six subjects. They then used these individualized models to select or generate predicted maximal-activator images for each subject. The fMRI responses indicated greater activation of the targeted visual region, FFA1, for synthetic images compared to the responses to group-based model images. This result highlights the potential of AI and fMRI for personalized visual-system modeling and exploring differences in visual system organization across populations.
The researchers are continuing their experiments with an advanced image generator called Stable Diffusion. This approach can also be applied to studying other senses, such as hearing.
Dr. Kuceyeski envisions therapeutic applications of this approach, stating, 'In principle, we could alter the connectivity between two parts of the brain using specifically designed stimuli, for example, to weaken a connection that causes excess anxiety.'
This collaborative research endeavors to unlock new insights into the human brain's visual processing mechanisms, paving the way for innovative discoveries in neuroscience.