New AI Method Cuts Brain Scan Time by 40%

Brain scans called fMRI (pictures of blood flow showing which brain areas talk to each other) are used to study memory, mood, and diseases like Alzheimer's. But there's a long-standing problem: nobody could say for sure how long a scan needs to run before the results become trustworthy. Scan too briefly and you might be measuring the machine's static, not the person's brain. That matters because every extra minute in a scanner costs money, and because shaky data can lead to wrong conclusions about who is sick and who isn't.

A team of researchers took a different approach. Instead of the usual math, they built a model that simulates brain activity and separately tracks scanner noise, then uses AI-style computation to work backwards and estimate what's really going on. Think of it like tuning a radio: older methods couldn't tell whether the crackle was the station or your equipment. This one can. They tested it on 28 healthy volunteers across 55 scans using a high-powered 7T scanner and a standard 3T scanner, the kind found in most hospitals.

The results are practical. On a 7T scanner, a single person's scan reaches its best possible precision in about 7 minutes, compared to 10 minutes on a 3T — roughly 40% less time. Across a whole study group, the weaker 3T scanner needed about 37 times more scanning time per person to match the 7T's clarity. They also found that sampling about 46 tiny cubes of brain tissue per region, roughly half of what's typical, was enough to capture 90% of the achievable detail.

In plain terms: this gives hospitals and labs a recipe for scanning smarter, not longer. Shorter scans are cheaper, more comfortable for patients, and easier on people who struggle to lie still. The researchers released their code so other labs can calculate their own limits. The honest caveat is that this is a small study of 28 healthy adults, so the exact time savings may shift for children, older patients, or people with movement disorders.