New artificial intelligence tool helps clinicians distinguish between dementia types

Alzheimer's disease and related dementias are expected to more than double by 2060. As June marks Alzheimer's and Brain Awareness Month, three University of Florida researchers are working to improve clinicians' ability to distinguish between these diseases - a critical step toward earlier diagnosis and better outcomes. 

In a recent study published in Neurology, researchers developed a new tool called Automated Imaging Differentiation for Dementia, or AIDD. The tool combines brain scans with AI to distinguish between two common forms of dementia: Alzheimer's disease dementia and dementia with Lewy bodies. The results showed that AIDD identified the two diseases with near-perfect accuracy, suggesting it could be a promising future tool for clinicians. 

"The use of AI and advanced imaging technology holds considerable promise to uncover brain degeneration patterns for dementia," said David Vaillancourt, Ph.D., a distinguished professor and the Orchid Endowed Chair for the UF Department of Applied Physiology & Kinesiology in the College of Health and Human Performance.

While both conditions are forms of dementia, they can present differently. For example, dementia with Lewy bodies often begins with attention, alertness and movement issues, whereas patients with Alzheimer's demonstrate memory problems. Unlike Alzheimer's, dementia with Lewy bodies requires a different treatment. 

Unfortunately, the two diseases are frequently confused, with up to 50% of patients living with dementia with Lewy bodies being misdiagnosed as having Alzheimer's. Today's diagnosis methods rely on a mix of evaluations, testing and brain scans rather than a single definitive test. In some cases, misdiagnosis can lead to treatments that worsen cognitive and motor functions. 

To build this tool, researchers analyzed 519 brain scans from patients with Alzheimer's, dementia with Lewy bodies and no disease (control group), collected from January 2007 to March 2022 at multiple research data centers. From this group, a subset of 387 scans (129 Alzheimer's, 129 dementia with Lewy bodies, 129 controls) was used to train and test the AI model. Eighty percent of the scans were used to train the machine, while the remaining 20% were used to test it. 

"To ensure the highest standards of reliability, we performed extensive validation experiments using data collected from multiple scanners and imaging centers," said Angelos Barmpoutis, Ph.D., a professor in the UF College of the Arts' Digital Worlds Institute, who worked on the study alongside Vaillancourt and Robin Chen, Ph.D., a postdoctoral student in the J. Crayton Pruitt Family Department of Biomedical Engineering.

The scans used a specialized MRI technique that measures extra fluid in the brain, often signaling brain cell damage and inflammation. These subtle water-movement patterns in the brain were analyzed with AI, allowing for more accurate identification of each disease. Across multiple brain scan comparisons, the tool demonstrated strong performance. To further test the system, researchers applied the tool to a separate group of 13 patients whose diagnoses were confirmed after death through autopsy. The tool correctly identified all 13 cases. 

"Since the therapies for Alzheimer's disease and dementia with Lewy bodies differ, developing precision biomarkers will offer better outcomes for patients," Vaillancourt said.