A ROUTINE EEG could soon change dementia diagnosis, as researchers show that artificial intelligence applied to EEG scans can reliably identify frontotemporal dementia while distinguishing it from Alzheimer’s disease and estimating disease severity with notable accuracy.
Why Frontotemporal Dementia is Often Missed
Frontotemporal dementia is the second most common cause of dementia after Alzheimer’s disease, yet it is frequently misdiagnosed due to overlapping cognitive symptoms. Unlike Alzheimer’s, which typically presents with memory loss, frontotemporal dementia often begins with changes in behaviour, personality, or language linked to degeneration of the frontal and temporal lobes. Current diagnostic pathways rely heavily on clinical assessment and expensive imaging, which can delay accurate identification. This has driven interest in electroencephalography, a portable and low-cost technique that measures brain electrical activity but has historically lacked diagnostic precision when used alone.
How EEG and AI can Identify Frontotemporal Dementia
In this study, researchers developed a deep learning framework to analyse EEG signals from individuals with Alzheimer’s disease, frontotemporal dementia, and cognitively normal controls. The model extracted both spectral and temporal features across all frequency bands, allowing subtle patterns to be detected beyond human visual inspection. A hybrid architecture combining a Convolutional Neural Network with an attention-based Long Short-Term Memory network enabled simultaneous disease classification and severity prediction. Key biomarkers included increased delta band activity in frontal and central regions, which appeared in both conditions. However, Alzheimer’s disease showed broader disruption across multiple brain regions, while frontotemporal dementia displayed more localised frontal and temporal changes.
Using a two-stage approach, the system first identified cognitively normal individuals, then differentiated frontotemporal dementia from Alzheimer’s disease. This method achieved an overall accuracy of 84% in classifying Alzheimer’s disease, cognitively normal status, and frontotemporal dementia. Feature selection also improved specificity for separating Alzheimer’s disease from frontotemporal dementia, increasing it from 26% to 65%. Severity prediction showed relative errors of less than 35% for Alzheimer’s disease and approximately 15.5% for frontotemporal dementia.
Implications For Clinical Practice
These findings suggest that AI enhanced EEG could become a practical clinical tool for earlier and more accurate identification of frontotemporal dementia. Faster differentiation may support timely counselling, more appropriate care planning, and personalised management, particularly in settings where access to advanced imaging is limited. Further validation could help integrate this approach into routine memory services.
Reference
Vo T et al. Extraction and interpretation of EEG features for diagnosis and severity prediction of Alzheimer’s disease and frontotemporal dementia using deep learning. Biomedical Signal Processing and Control. 2026;112:108667.
