A new framework published in the journal Brain offers a different way to think about how memories are formed and how that process may go wrong in Alzheimer's disease and other dementias. David T. Jones, M.D., a Mayo Clinic neurologist who directs the Mayo Clinic Biological Intelligence Lab, and Michael Breakspear, M.B.B.S., Ph.D., of the University of Newcastle in Australia, describe the brain's limbic system using a concept from generative artificial intelligence called latent diffusion.
In AI models, latent diffusion converts complex input, such as an image or a text prompt, into a compressed representation of its underlying features, processes it in a hidden space, and then converts it back into an output. The researchers' proposal, called stochastic latent oscillatory diffusion (SLOD), splits this process across inner brain structures into four steps: compression, diffusion, conditioning and decompression. Sensory and cognitive stimuli begin in the cerebral cortex, are compressed, and travel to the hippocampus through the entorhinal cortex. In the hippocampus they are denoised and conditioned until the brain needs to recall or reconstruct them, at which point they pass back out through connected structures.
Dr. Jones contrasts this with a filing-cabinet picture of memory. In his description, the hippocampus learns the underlying features of the world and models them, which allows those features to be recombined into things a person has never directly experienced.
The framework grew partly from Dr. Jones's clinical work with limbic-predominant amnestic neurodegenerative syndrome, or LANS, a memory-loss condition affecting the limbic system. TDP-43, the protein linked to LANS, often builds up in patients' brains, but Dr. Jones observed that it was sometimes absent, and when present it often appeared alongside other protein abnormalities. He came to view protein buildups as possibly signaling where the problem lies in the limbic system rather than being the whole problem.
The authors suggest that disruptions to the four processes could help physician-scientists classify neurodegenerative diseases, instead of relying only on cellular or protein abnormalities. Dr. Jones gives one example: a brain's failure to compress stimuli may be indicative of Alzheimer's disease. He presents the work as an opening step, saying it suggests new ways to approach treatment of dementia.