AI That Reads PET Scans Can Distinguish Dementia Types

The tool can make it easier to point out causes of dementia that are reversible

Marlene Cimons,

AARP

Comments

En español

Published July 10, 2025

Reviewing a PET scan of the brain, an AI tool can boost speed and accuracy of a dementia diagnosis.

AARP (Getty Images, Dr. David Jones)

Researchers at the Mayo Clinic using artificial intelligence (AI) to analyze specific brain activity can detect at least nine different types of dementia. Their aim is faster and more accurate diagnoses, along with the possibility of detecting dementias that are reversible, they say.

“A proper diagnosis can change everything,” says David Jones, M.D., a neurologist who directs the Mayo Clinic’s Neurology Artificial Intelligence Program. “There are many people with dementia-like symptoms that are reversible. This tool enhances the ability to diagnose … these disorders, which is hard for humans to do alone.” That, in turn, makes it more likely patients will get the correct treatments.

The researchers used a tool called StateViewer to correctly identify nine different kinds of dementia (they later ID’d five more), according to a paper published in the journal Neurology. The findings pointed to variations within Alzheimer’s disease, the most common dementia, affecting more than 7 million Americans, as well as Lewy body disease and frontotemporal dementia, according to Jones.

“StateViewer focuses on brain function closely tied to symptoms, so dementias with different symptoms will appear as different forms to StateViewer,” says Jones. This variability is the biological reason why some people with nontraditional forms of Alzheimer’s experience vision problems and go to the eye doctor first; others have speech or language difficulties, and some have issues with executive function, or decision-making, instead of memory loss, he says.

A faster diagnosis with a boost in accuracy

The Mayo scientists trained and tested the AI tool on more than 3,600 scans called FDG-PET (fluorodeoxyglucose positron emission tomography). These scans display brain metabolism, specifically how the brain uses glucose — or sugar — for energy. The tool then compared these scans to a large database of scans from patients with various types of confirmed dementia.

The process of glucose metabolism differs depending on the brain location and the type of dementia, Jones says.

“The patterns are distinctly different from one another,” he says. “[The tool] is looking at which areas of the brain are taking up sugar. These patterns are more subtle than we think. This lack of precision is what makes it hard to treat and why we need AI. To the human eye, they may look similar, but when AI analyzes them, they are clearly very distinct.”

Most Popular

The tool distinguished dementia causes with 88 percent accuracy, and enabled clinicians to interpret brain scans nearly twice as fast and with up to three times greater accuracy than a radiologist examining raw images, sometimes aided by statistical analysis.

Diagnosing dementia typically also involves cognitive tests, blood draws, imaging, clinical interviews and specialist referrals.

The goal is not to replace clinicians — but to use AI and clinicians together to boost diagnostic precision, Jones says. The 88 percent measure “is just the tool itself,” he says. “The tool combined with the clinician is the best performer. You increase the accuracy by including humans in the loop.”

Alzheimer’s typically affects two regions of the brain involved in memory early in the disease. Lewy body dementia involves areas related to movement and visual and spatial processing. Frontotemporal dementia changes regions responsible for language and behavior, with less effect on memory.

StateViewer displays these patterns through color-coded brain maps that highlight key areas of brain activity.

Andrew E. Budson, M.D., chief of cognitive behavioral neurology at the VA Boston Healthcare System, says blood and other biomarkers, substances that characterize specific dementias, are more valuable diagnostic tools.

“The research described in this article is helpful to diagnose less common dementias, but it is not a game changer,” says Budson, who was not involved in the study.“Today, we want to use a specific biomarker for a specific diagnosis as much as possible,," he wrote in an email.

“I think it is a very valuable study for exploring how machine learning can be used eventually to make the diagnoses via FDG-PET by themselves, thereby potentially reducing the need for other tests and for speeding the arrival at a diagnosis and treatment,” says Paul Schulz, M.D., professor of neurology at the McGovern Medical School, University of Texas Health Science Center in Houston.

“Everything that we do to automate performing tests and interpreting them moves us toward progressively accurate, and more timely, diagnoses,” Schulz, who was not involved in the research, wrote in an email. Moreover, use of the FDG-PET scan could help reduce the need for amyloid PET scans, which are more expensive and apply only to Alzheimer’s disease, he says.

Pointing out a reversible condition

StateViewer already is in widespread use at the Mayo Clinic, where Jones says it is having an impact. He tells the story of Minoo Press, a retired engineer from the Chicago area who came to the Mayo Clinic after two years of progressive cognitive and physical decline. Previous clinicians couldn’t figure out what was wrong with him, partly because his dementia symptoms were complicated by other signs of neurogenerative disease.

They considered Alzheimer’s and other forms of dementia, including normal pressure hydrocephalus, where excess fluid builds up in the brain, producing dementia-like symptoms. But they couldn’t provide him with a definitive diagnosis. He came to Mayo hoping clinicians there could help.

The tool ruled out Alzheimer’s and other dementias and confirmed that Press was suffering from normal pressure hydrocephalus. Surgeons there performed a shunt procedure to relieve the pressure in his brain, easing many of his symptoms and putting him on a path to recovery.

“I have patients who’ve been going for years without a diagnosis, and this tool can do it in hours,” Jones says. “Once it becomes clear and you know what you have, you can focus on what comes next — not more tests, second opinions or doubt. It’s terrible to have a reversible condition that no one is able to diagnose, but it’s a tragedy that technology can fix.”

%{postComment}%