Researchers at Örebro University have developed two new AI models that can analyze the brain’s electrical activity and accurately distinguish between healthy individuals and patients with dementia, including Alzheimer’s disease.
Subscribe to our newsletter for the latest sci-tech news updates.
“Early diagnosis is crucial in order to be able to take proactive measures that slow down the progression of the disease and improve the patient’s quality of life,” says Muhammad Hanif, researcher in informatics at Örebro University.
In the study, “An explainable and efficient deep learning framework for EEG-based diagnosis of Alzheimer’s disease and frontotemporal dementia,” researchers combined two advanced AI methods—temporal convolutional networks and LSTM networks. The program analyzes EEG signals and can determine almost flawlessly whether a person is sick or healthy. This study appears in Frontiers in Medicine.
Can distinguish healthy from sick with 80% certainty
When comparing three groups—Alzheimer’s, frontotemporal dementia and healthy—the method achieved over 80% accuracy. The researchers also use an explanatory AI technique that shows which parts of the EEG signal affect the diagnosis. This helps doctors interpret how the system reaches its conclusions.
In the second study, “Privacy–preserving dementia classification from EEG via hybrid–fusion EEGNetv4 and federated learning,” the researchers developed a small and resource-efficient AI model—under one megabyte in size—that also safeguards patient privacy. This study was published in Frontiers in Computational Neuroscience.
With the help of federated learning, multiple health care providers can collaborate to train the AI system without sharing patient data. Despite the privacy protection, the model achieves over 97% accuracy.
“Traditional machine learning models often lack transparency and are challenged by privacy concerns. Our study aims to address both issues,” says Hanif, associate senior lecturer of informatics at Örebro University.
AI detects patterns in the brain’s electrical signals
The researchers have succeeded in combining different methods of interpreting the brain’s electrical signals. By dividing EEG signals into various frequency bands—alpha, beta and gamma waves—the AI can identify patterns linked to dementia.
The algorithms can detect long-term changes in the signals and recognize subtle differences between diagnoses. In addition, the explainable AI technology ensures the system is no longer a “black box”—it clearly shows the basis for its decisions.
In their studies, the researchers demonstrate how AI can become a rapid, low-cost and privacy-safe tool for early diagnosis of dementia. EEG is already a simple and inexpensive method that can be used in primary care. Combined with AI models that can run on portable devices, this opens up the potential for wider use in health care—from specialist clinics to future home testing.
The AI test could be used at home in the future
“Early diagnosis is essential for implementing proactive measures that slow disease progression and improve quality of life. If solutions like this are fully implemented, it could ease the burden for everyone involved—patients, care staff, relatives and health care professionals,” says Hanif.
The studies were conducted in collaboration between researchers at Örebro University and several international institutions, including universities in the UK, Australia, Pakistan and Saudi Arabia.
“We plan to continue the research by expanding to larger and more diverse datasets, exploring more EEG features, and including other types of dementia such as vascular dementia and Lewy body dementia. At the same time, we will use explainable AI and ensure strict protection of patient data,” explains Hanif.
More information: Waqar Khan et al, An explainable and efficient deep learning framework for EEG-based diagnosis of Alzheimer’s disease and frontotemporal dementia, Frontiers in Medicine (2025). DOI: 10.3389/fmed.2025.1590201
Muhammad Umair et al, Privacy–preserving dementia classification from EEG via hybrid–fusion EEGNetv4 and federated learning, Frontiers in Computational Neuroscience (2025). DOI: 10.3389/fncom.2025.1617883
Provided by Örebro Universitet
This story was originally published on Medical Xpress.
