A BU-led research team shows how Generative AI can be used to diagnose different forms of dementia
New Generative AI Model can diagnose 10 distinct types of dementia
With over 10 million new cases of dementia diagnosed each year, it’s critical to facilitate accurate diagnosis to ensure patients have access to early, personalized treatment to maximize outcomes. While doctors typically can diagnose individuals with dementia with high confidence, it’s challenging to discern the type of dementia because different types often present with similar symptoms and brain changes.
Now an interdisciplinary research team, led by Boston University, has developed an artificial intelligence (AI) tool that shows promising results in enabling the differential diagnosis of dementia. Their study, published in Nature Medicine, presents a new generative AI method for diagnosing ten different types of dementia, including vascular dementia, Lewy body dementia, and frontotemporal dementia.
Researchers used a broad range of data to develop the AI model from routinely collected clinical data, such as demographic information, individual and family medical history, medication use, neurological and neuropsychological exam scores, and neuroimaging data. The data was drawn from nine independent, geographically diverse datasets of over 50,000 participants. Using this information, researchers identified ten distinct forms of dementia.
“Our generative AI tool enables differential dementia diagnosis using routinely collected clinical data, showing its potential as a scalable diagnostic tool for Alzheimer’s Disease (AD) and related dementias,” says lead study author Vijaya Kolachalama, associate professor of medicine and computer science, affiliate faculty of Hariri Institute for Computing, and a founding member of the Faculty of Computing & Data Sciences at Boston University. “The ability to generate diagnosis with routine clinical data is becoming increasingly important given the significant challenges in accessing gold-standard testing, not only in remote and economically developing regions but also in urban healthcare centers.”
The multimodal machine learning model that researchers developed achieved an area under the receiver operating characteristic (ROC) curve of 0.96 in differentiating the dementia types. The ROC score can range from 0 to 1. A score of 0.5 indicates random guessing, and a score of 1 indicates perfect performance. The model’s predictions aligned with biomarker evidence discovered in postmortem findings.
The team also compared the performance of neurologists and neuroradiologists working alone versus with the AI tool and found that AI can boost the accuracy of neurologists by over 26% across all 10 dementia types. Using 100 randomly selected cases, 12 neurologists were asked to make a diagnosis and provide a confidence score between 0 and 100. This confidence score was then averaged with the probability score obtained by the AI tool to determine an AI-augmented neurologist score.
“There aren’t enough neurology experts around the world, and the number of patients needing their help is growing quickly. This mismatch is putting a big strain on the healthcare system. We believe AI can help by identifying these disorders early and assisting doctors in managing their patients more effectively, preventing the diseases from getting worse,” says Dr. Kolachalama.
With dementia cases set to double in the next 20 years, the researchers hope that this AI tool can provide accurate differential diagnosis and support the increased demand for targeted therapeutic interventions for dementia.
