Machine Learning Differentiates AD and Behavioral Variant Frontotemporal Dementia

Machine learning (ML) techniques demonstrate high accuracy in distinguishing Alzheimer disease (AD) from behavioral variant frontotemporal dementia (bvFTD), according to findings published in Alzheimer’s & Dementia.

Since AD and bvFTD often present with overlapping symptoms, early diagnosis remains clinically challenging.

Researchers analyzed data from 1616 autopsy-confirmed participants, including 1498 with AD and 118 with bvFTD, in the National Alzheimer’s Coordinating Center database. Participants with AD were older at their initial clinical visit (mean age, 72.2 years) than those with bvFTD (mean age, 62.2 years) and were more often women (49.8% vs 33.1%). Both groups were predominantly non-Hispanic White (88.4% vs 93.2%), and most participants met dementia criteria at first visit (79.8% with AD vs 96.6% with bvFTD).

Findings emphasize the importance of neuropsychological and neuropsychiatric assessment during the initial clinic visit.

The researchers evaluated 20 clinical variables, including 12 neuropsychiatric symptoms and 8 neuropsychological test scores, using four ML algorithms: logistic regression, support vector machines, random forest, and artificial neural networks.

All models achieved strong performance, with accuracy ranging from 80% to 90% and area under the curve (AUC) values between 0.89 and 0.95. The logistic regression model performed best overall (accuracy, 88%; AUC, 0.95), followed by the artificial neural networks model (accuracy, 90%; AUC, 0.89). Sensitivity and specificity across models were consistently high (78%-91% and 80%-91%, respectively).

Apathy, disinhibition, and performance on the Digit Symbol Substitution Test were identified as the strongest predictors of diagnostic classification across multiple models. Participants with bvFTD showed higher neuropsychiatric symptom severity (mean total score, 9.7) compared with those with AD (mean total score, 4.3). Anxiety was the most common symptom in AD, while apathy and disinhibition predominated in bvFTD. Patients with bvFTD performed better on tests of processing speed and executive function, while those with AD performed better on language tasks.

Misclassified bvFTD cases showed lower neuropsychiatric symptom severity and poorer memory scores, resembling AD profiles.

Study limitations include sample imbalance between AD and bvFTD groups, limited racial diversity, and the absence of social functioning measures.

“Findings emphasize the importance of neuropsychological and neuropsychiatric assessment during the initial clinic visit,” the study authors concluded.

Disclosures: This research was supported by the National Institute on Aging, the National Institute of Health and other funding partners through the NACC program. One study author declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of disclosures.