Summary: New research demonstrates that deep learning applied to EEG recordings can accurately distinguish Alzheimer’s disease (AD) from frontotemporal dementia (FTD). By evaluating both the timing and frequency components of brain activity, the model revealed distinct signatures: Alzheimer’s produces broader disruptions across multiple regions and frequency bands, while frontotemporal dementia causes more localized changes in frontal and temporal areas.
The system also estimates disease severity, offering clinicians a faster, more affordable complement to MRI and PET scans. These results indicate that widely available EEG technology, combined with advanced artificial intelligence, could streamline diagnosis and support more personalized care for people with cognitive decline.
Key Facts
- EEG biomarkers: Elevated slow (delta) activity in frontal and central regions was a consistent indicator of disease.
- Distinct patterns: Alzheimer’s showed widespread cortical disruption across multiple bands; frontotemporal dementia presented more focal frontal and temporal abnormalities.
- High accuracy: A two-stage deep learning framework achieved 84% accuracy when separating AD, FTD, and cognitively normal participants.
Source: FAU
About the diseases: Dementia describes a set of conditions that progressively impair memory, thinking and daily functioning. Alzheimer’s disease is the most prevalent form, while frontotemporal dementia is a leading cause of early-onset dementia, often affecting people between their 40s and 60s. Although both conditions damage the brain, they affect different networks—AD commonly impairs memory and spatial skills, whereas FTD mainly alters behavior, personality and language.
Misdiagnosis is common because symptoms overlap between AD and FTD. Accurate differentiation matters: it guides treatment choices, care planning and prognosis. While MRI and PET imaging are effective, they are expensive and not always accessible. EEG provides a portable, noninvasive and cost-effective alternative by recording electrical brain activity across multiple frequency bands. Still, EEG signals are noisy and person-to-person variability complicates interpretation, limiting diagnostic use—especially when distinguishing AD from FTD.
Researchers in the College of Engineering and Computer Science at Florida Atlantic University developed a deep learning approach that improves EEG-based detection and interpretation. The model simultaneously extracts spectral (frequency) and temporal (time-based) features from EEG signals, increasing both diagnostic accuracy and clinical interpretability.
Published in Biomedical Signal Processing and Control, the study identified increased delta-band activity in frontal and central electrodes as a key biomarker for both AD and FTD. Alzheimer’s patients, however, demonstrated broader disruptions that extended to parietal regions and affected higher frequency bands such as beta—consistent with more extensive neurodegeneration. These differences help explain why AD tends to be easier to detect from EEG than FTD.
Performance highlights: the model exceeded 90% accuracy when separating people with dementia (AD or FTD) from cognitively normal controls. It also estimated disease severity with mean relative errors under 35% for AD and about 15.5% for FTD. Because AD and FTD share overlapping EEG characteristics, the authors applied feature selection to improve specificity (the ability to correctly identify people without the disease), increasing it from 26% to 65% for differentiating FTD from AD.
The researchers implemented a two-stage classification pipeline: first identify cognitively normal individuals, then separate AD from FTD among the remaining cases. This approach yielded an overall classification accuracy of 84%, placing it among the stronger EEG-based methods to date.
Methodologically, the system combines convolutional neural networks (CNNs) to capture spatial patterns and attention-based long short-term memory networks (aLSTMs) to model temporal dynamics. Visual explanation tools such as Grad-CAM were used to highlight which electrodes, time windows and frequency components most influenced the model’s decisions, improving transparency and helping clinicians interpret results.
Lead author Tuan Vo, a doctoral student in FAU’s Department of Electrical Engineering and Computer Science, notes that the model’s novelty lies in extracting both spatial and temporal EEG information to reveal subtle disease-related brainwave patterns that standard analyses miss. Co-author Hanqi Zhuang, Ph.D., emphasizes that the broader cortical disruption seen in Alzheimer’s—especially in frontal, parietal and temporal regions—makes it more detectable, but careful feature selection can markedly improve FTD identification.
By combining detection and severity estimation in one pipeline, the approach has the potential to reduce lengthy evaluations and provide clinicians timely tools for monitoring progression. As Stella Batalama, Ph.D., dean of FAU’s College of Engineering and Computer Science, stated, integrating engineering, AI and neuroscience can lead to earlier diagnosis, better personalized care and more effective interventions for millions affected by dementia.
Study co-authors include Ali K. Ibrahim, Ph.D., and doctoral student Chiron Bang, both in FAU’s Department of Electrical Engineering and Computer Science.
Key Questions Answered:
A: Overlapping symptoms and similar EEG features often cause misdiagnosis unless specialized imaging or advanced analysis is used.
A: By extracting both spatial and temporal EEG features and applying attention mechanisms, the model finds subtle brainwave differences that conventional techniques can miss.
A: Yes. The model provides severity estimates for both AD and FTD, supporting longitudinal tracking and clinical decision-making.
Editorial Notes:
- Edited by a Neuroscience News editor.
- Full journal article reviewed by staff.
- Additional context added for clarity and reader value.
About this AI and neurotech research news
Author: Gisele Galoustian
Source: FAU
Contact: Gisele Galoustian – FAU
Image: Image credit to Neuroscience News
Original Research: Open access. “Extraction and interpretation of EEG features for diagnosis and severity prediction of Alzheimer’s Disease and Frontotemporal dementia using deep learning” by Tuan Vo et al., published in Biomedical Signal Processing and Control.
Abstract
Extraction and interpretation of EEG features for diagnosis and severity prediction of Alzheimer’s Disease and Frontotemporal dementia using deep learning
Alzheimer’s disease (AD) causes progressive cognitive decline and memory loss. Frontotemporal dementia (FTD) primarily impairs the frontal and temporal lobes, producing changes in personality, behavior and language. Because clinical signs overlap, FTD is frequently misdiagnosed as AD. Although EEG is portable, noninvasive and affordable, its diagnostic utility is challenged by similarities between these dementias and variability in EEG signals.
This study introduces an EEG-based feature extraction pipeline that combines spectral and temporal information using convolutional neural networks and attention-based LSTM units. The approach identifies increased delta-band activity in frontal and central regions as a key biomarker, achieves over 90% accuracy distinguishing dementia cases from cognitively normal controls, and predicts severity with relative errors under 35% for AD and around 15.5% for FTD.
Discriminating AD from FTD remains challenging; however, a feature selection step improved specificity from 26% to 65%. Building on this, the team implemented a two-stage classifier—first isolating cognitively normal individuals, then separating AD and FTD—reaching 84% overall classification accuracy. These results demonstrate that combining EEG with deep learning can produce interpretable, clinically relevant tools for diagnosis and monitoring of dementia.