Mayo Clinic develops AI capable of substantially improving dementia diagnoses
Experts at the Mayo Clinic have developed an artificial intelligence tool that could markedly improve the diagnosis of different types of dementia.
Diagnosing the disease—and determining the type—typically requires a slew of exams that include imaging analyses, behavioral assessments, lab tests, psychological evaluations and more. And even with a file full of evidence, the process can still be somewhat subjective, making it difficult to decide on the best course of treatment.
Thanks to numerous advances in imaging and other technologies, providers are now able to spot signs of impending cognitive decline before the onset of symptoms, enabling them to initiate preventive strategies earlier. The recent emergence of artificial intelligence has taken these diagnostic advances a step further.
At Mayo Clinic, a team of experts developed and trained an AI tool to spot patterns on fluorodeoxyglucose positron emission tomography (FDG-PET) scans that can help physicians distinguish between dementias. It displays the patterns in color-coded maps of different regions of the brain, helping identify altered activity in areas related to specific symptoms. The tool, known as StateViewer, not only improves diagnostic accuracy, but it also reduces interpretation times.
David Jones, MD, a Mayo Clinic neurologist and director of the Mayo Clinic Neurology Artificial Intelligence Program, spearheaded the AI’s development.
“Every patient who walks into my clinic carries a unique story shaped by the brain’s complexity,” Jones said in a statement. “That complexity drew me to neurology and continues to drive my commitment to clearer answers. StateViewer reflects that commitment—a step toward earlier understanding, more precise treatment and, one day, changing the course of these diseases.”
StateViewer was trained on thousands of FDG-PET images, including those from individuals with dementia and patients with no signs of cognitive decline. During testing, it was able to differentiate between nine neurodegenerative phenotypes with a sensitivity of 0.89 ± 0.03 and an area under the receiver operating characteristic curve of 0.93 ± 0.02. Readers using the tool were three times more accurate with its help compared to those who did not. Additionally, interpretation times were nearly twice as fast with the help of StateViewer.
Leland Barnard, PhD, a data scientist who led the AI engineering team, expressed optimism for how it could benefit providers and patients in the future.
“As we were designing StateViewer, we never lost sight of the fact that behind every data point and brain scan was a person facing a difficult diagnosis and urgent questions,” Barnard said. “Seeing how this tool could assist physicians with real-time, precise insights and guidance highlights the potential of machine learning for clinical medicine.”
StateViewer is continuing to be analyzed in varying clinical settings. To learn more, read the study abstract in Neurology.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.