health

Researchers tested several AI tools — called "foundation models" — designed to read brain MRI scans and estimate things like a person's biological brain age or biological sex. Foundation models are large AI systems trained on vast amounts of data that can then be adapted to many tasks. The team compared how well these models performed on two specific jobs: guessing how old a brain looks based on its structure ("brain age regression"), and identifying whether a brain scan belongs to a male or female ("sex classification"). They ran these tests across multiple brain-imaging datasets and compared the AI models against each other to see which approaches work best.

The study found that no single foundation model dominated across all tasks and datasets — performance varied considerably depending on the model and the specific challenge. Some models were better at estimating brain age, others at sex classification, and results shifted depending on which dataset was used. Importantly, the researchers highlight that simple, well-tuned traditional approaches can still be competitive with cutting-edge AI foundation models, meaning bigger and newer isn't always better.

For people living with Parkinson's, this matters because brain age — how "old" your brain looks structurally compared to your actual age — is increasingly studied as a potential marker of neurodegeneration and disease progression. If AI tools can reliably and accurately measure brain age from routine MRI scans, they could one day help doctors detect Parkinson's earlier, track how fast someone's disease is progressing, or predict who might develop problems with thinking and memory. However, this study is a technical benchmarking exercise — a comparison of tools, not a clinical trial — so it does not directly change care today. It lays important groundwork for building trustworthy AI brain-imaging tools, which could begin influencing research study design within the next few years.