Researchers at Mayo Clinic and the University of Newcastle have introduced a new model in Brain that may transform scientific perspectives on memory formation and neurodegenerative disorders. Led by neurologist David T. Jones, M.D.—director of Mayo Clinic’s Biological Intelligence Lab—and Michael Breakspear, M.B.B.S., Ph.D., the team applied latent diffusion, a technique from generative AI, to analyze how the limbic system organizes information.
How the brain processes data
In generative AI models, latent diffusion engines convert complex information, such as an image or text prompt, into a compressed representation of its underlying features. The model then processes that information within a latent, or hidden, space before converting it back into an output. They propose a framework called stochastic latent oscillatory diffusion (SLOD) that breaks this process into four core steps: compression, diffusion, conditioning, and decompression.
The cerebral cortex, which processes sensory and cognitive stimuli, is where SLOD begins. These stimuli are compressed and travel to the hippocampus via the entorhinal cortex, a major pathway to the inner brain. Once the stimuli reach the hippocampus, they undergo diffusion and denoising. Underlying representations of the stimuli are conditioned there until the brain has a need to recall or reconstruct them, such as for memory or even imagined memory, for output back through connected brain structures.
Classifying disease by function
This research also presents new avenues to identifying neurodegenerative dementias. The scientists propose that disruptions to the four diffusion processes might help physician-scientists classify neurodegenerative diseases, rather than relying solely on cellular or protein abnormalities. For example, a brain’s failure to compress stimuli may be indicative of Alzheimer’s disease.
Read Also: Ultrarunner completes race after stomach removal surgery
Dr. Jones likens the hippocampus to a creative engine. “We propose the hippocampus doesn’t store memories the way a filing cabinet stores documents,” he says. “Instead, it learns the underlying features of the world and models it at a deeper level. The format that it’s using is so flexible that the hidden features of the world can be combined and produced into new things beyond what you know or have experienced.”
The framework grew in part from a problem Dr. Jones encountered while treating a memory-loss syndrome impacting the brain’s limbic system called limbic-predominant amnestic neurodegenerative syndrome, or LANS. TDP-43, the protein associated with LANS, often accumulates in the brains of people with the syndrome. While caring for patients with LANS, Dr. Jones observed that sometimes TDP-43 was absent, and when it was present, it often occurred alongside other protein abnormalities.
Dr. Jones recalls that he was puzzled by this phenomenon. “If this is all occurring in the limbic system, where several brain structures converge, protein buildups may be telling us where the problem is rather than being the whole of the problem,” he notes. This observation led him to consider alternative explanations for the syndrome, ultimately contributing to the development of the SLOD framework.
