Dementia is a group of disorders that gradually impair memory, thinking, and daily functioning. Alzheimer's disease (AD), the most common form of dementia, is expected to affect approximately 7.2 million Americans age 65 and older by 2025. Frontotemporal dementia (FTD), although rare, is the second most common cause of early-onset dementia, often affecting people in their 40s to 60s.
Both diseases damage the brain, but in different ways. While AD primarily affects memory and spatial awareness, FTD targets areas responsible for behavior, personality, and language. Their symptoms can overlap, often leading to misdiagnosis. Differentiating between these is not only a scientific challenge but also a clinical necessity, as accurate diagnosis can have a major impact on treatment, care, and quality of life.
MRI and PET scans are effective in diagnosing AD, but are expensive, time-consuming, and require specialized equipment. Electroencephalography (EEG) provides a portable, non-invasive, and affordable alternative by measuring brain activity with sensors across different frequency bands. However, the signals are noisy and have individual differences, making analysis difficult. Even when machine learning is applied to EEG data, the results are inconsistent and it remains difficult to distinguish between AD and FTD.
To address this problem, researchers at Florida Atlantic University's School of Engineering and Computer Science created a deep learning model to detect and evaluate AD and FTD. Analyzing both frequency-based and time-based brain activity patterns associated with each disease improves the accuracy and interpretability of EEG.
The findings, published in the journal Biomedical Signal Processing and Control, found that slow delta brain waves, primarily in frontal and central regions of the brain, are important biomarkers for both Alzheimer's disease and FTD. In Alzheimer's disease, brain activity is disrupted more broadly, affecting other areas of the brain and frequency bands such as beta, indicating more widespread brain damage. These differences help explain why AD is generally easier to detect than FTD.
This model achieved more than The accuracy of distinguishing between patients with dementia (AD or FTD) and cognitively normal participants is 90%. It also predicted disease severity with a relative error of less than 35% for AD and less than 15.5% for FTD.
AD and FTD have been difficult to distinguish because their symptoms and brain activity are similar. The researchers used feature selection to increase the model's specificity, or how well it identifies people without the disease, from 26% to 65%. Their two-step design (first detecting healthy people and then separating AD and FTD) achieved 84% accuracy, ranking among the best EEG-based methods to date.
This model combines convolutional neural networks and attention-based LSTM to detect both dementia type and severity from EEG data. Grad-CAM shows which brain signals influenced the model and helps clinicians understand the model's decisions. This approach provides a new perspective on how brain activity evolves and which regions and frequencies influence diagnosis, something that traditional tools have rarely been able to capture.
“What makes our study novel is how we used deep learning to extract both spatial and temporal information from EEG signals,” said first author Tuan Vo, a doctoral student in FAU's Department of Electrical Engineering and Computer Science. “By doing this, we are able to detect subtle brain wave patterns associated with Alzheimer's disease and frontotemporal dementia that would otherwise go unnoticed. Our model not only identifies the disease, but also estimates its severity, providing a more complete picture of each patient's condition.”
The findings also revealed that AD tends to be more severe, affecting a wide range of brain regions and leading to poorer cognitive scores, while the effects of FTD are more localized to the frontal and temporal lobes. These insights are consistent with previous neuroimaging studies, but add new depth by showing how these patterns appear in EEG data, an inexpensive, non-invasive diagnostic tool.
“Our findings show that Alzheimer's disease disrupts brain activity more broadly, especially in frontal, parietal, and temporal regions, whereas frontotemporal dementia primarily affects frontal and central regions,” said co-author Dr. Hanqi Zhuang, associate dean and professor in the FAU School of Electrical Engineering and Computer Science. “This difference explains why Alzheimer's disease is easier to detect. But our study also shows that with careful selection of features, we can significantly improve the accuracy of distinguishing between FTD and Alzheimer's disease.”
Overall, this study shows that deep learning can streamline dementia diagnosis by integrating detection and severity assessment into one system, reducing lengthy assessments and providing clinicians with real-time tools to track disease progression.
“This research shows how bringing together engineering, AI, and neuroscience can transform how we address major health challenges,” said Dr. Stella Batalama, dean of the School of Engineering and Computer Science. “With millions of people affected by Alzheimer's disease and frontotemporal dementia, breakthroughs like this open the door to earlier detection, more personalized care, and interventions that can truly improve lives.”
reference: Vo T, Ibrahim AK, Zhuang H, Bang C. Extracting and interpreting EEG features for diagnosis and severity prediction of Alzheimer's disease and frontotemporal dementia using deep learning. Biosignal process control. 2026;112:108667. doi: 10.1016/j.bspc.2025.108667
This article has been reprinted from the following material: Note: Materials may be edited for length and content. Please contact the citation source for details. You can access our press release publishing policy here.
