Generative AI Is Helping Stanford Researchers Better Understand Brain Diseases

If you could visually see how your habits today might affect your brain 10 or 20 years from now, would you change your behavior?

While that’s a hypothetical question today, advances in generative AI and neuroscience may one day turn that option into a reality. Researchers at Stanford are leading the way using generative AI to create synthetic brain MRIs (magnetic resonance imaging) to increase the scale and diversity of training data sets to accelerate our understanding of brain disorders. Someday, this technology might even be able to simulate what your future brain looks like.

Kilian M. Pohl, professor of psychiatry and behavioral sciences and, by courtesy, of electrical engineering at Stanford, says that “future breakthrough discoveries in neuroscience will rely on AI technology. The problem currently is that this technology tends to produce unreliable results, as most brain MRI studies are simply not large enough.” That is why Pohl is taking advantage of large studies to create deep-learning models for generating realistic-looking brain MRIs that then can be used by smaller studies. 

Enabled by funding from the Stanford Institute for Human-Centered AI (HAI) Google Cloud Credits grant program and the National Institutes of Health, Pohl worked jointly with former research scientist Wei Peng and other researchers from the Computational Neuroscience Laboratory on creating a model called BrainSynth that synthesizes realistic, high-resolution MRIs to help replicate disease effects. The generated MRIs can augment data sets with countless more samples to better conduct brain research. That means a data set that might have had only 100 samples before could now have 5,000 for training AI methods on. 

Such enriched datasets could be used to understand common conditions (like depression, substance abuse disorders, or neurocognitive impairment) in the general population as well as specific subgroups, such as people with HIV. Longitudinal studies could also become more cost-effective, allowing researchers to simulate the brain in between less frequent participant scans.

Pohl, who co-directs the AI for Mental Health Initiative and is a faculty affiliate of Stanford HAI and the Wu Tsai Neurosciences Institute, is most excited about applying BrainSynth toward learning about diseases that subtly affect the brain. “Many diseases or conditions that I study are ones that are not well understood, and the impact on the brain has subtle effects that you can’t often see with the naked eye,” Pohl said. “I want to use this generative AI technology to capture those subtle effects.”

Since current generative AI technology is far from perfect, sometimes hallucinating, Pohl cautions that the synthetic MRIs are only used for training for now, not testing or inference. Synthetic MRIs must be reviewed to ensure they are anatomically correct and possible in a human. Pohl says his research team compares real MRIs with the synthetic images to see how well they overlap to ensure these systems are working and to improve training.

Down the road, Pohl is optimistic that the technology could also be used for education and prevention: What will my brain look like if I keep doing X? It could also be used for surgery planning to project the long-term consequences of a treatment and how the brain might look differently in the future.

Right now, Pohl and the research team are focused on improving the realism of the synthetic MRIs, including factoring in multimodal images (different types of MRIs like functional, structural, or diffusion) and accounting for population-specific characteristics. Doing so is part of Pohl’s vision to “build trustworthy AI technology that will transform psychiatry from subjective observations to objective assessments, resulting in more effective and accessible research and care.”