What is it about?
Alzheimer’s disease changes how brain cells communicate, but understanding these changes from measurements of brain activity remains a major challenge. In this study, we developed an approach that combines artificial intelligence, brain-inspired computing, and computer simulations to investigate these changes using electroencephalography (EEG), a non-invasive technique that records the brain’s electrical activity through sensors placed on the scalp. We trained a spiking neural network, a type of artificial intelligence that mimics how neurons communicate through electrical spikes, to distinguish EEG recordings from people with Alzheimer’s disease from those of healthy individuals. The model achieved promising classification performance. We then used computer simulations of networks containing thousands of artificial neurons to investigate how changes in the balance between excitatory and inhibitory activity could explain the patterns observed in the EEG. We also incorporated information about connections between brain regions to make the simulations more representative of real brain networks. Finally, we deployed the model on specialised neuromorphic hardware designed to mimic aspects of neural computation. This implementation retained competitive classification performance while offering the potential for substantially lower energy consumption than conventional GPU-based computing. By connecting data-driven predictions with biologically grounded simulations, this work offers a step towards AI systems that are not only able to recognise patterns associated with Alzheimer’s disease, but can also help researchers investigate the underlying brain mechanisms.
Featured Image
Photo by Growtika on Unsplash
Why is it important?
This work addresses two important challenges in AI for neuroscience: understanding why a model recognises disease-related patterns and reducing the computational resources required to run it. By linking EEG-based classification to simulations of neuronal excitation, inhibition, and brain connectivity, the framework provides a way to investigate potential physiological explanations for observed brain signals. Its implementation on neuromorphic hardware also demonstrates the potential for more energy-efficient brain-signal analysis. In the longer term, this approach could contribute to accessible, interpretable EEG tools for neurological research and support the development of future screening technologies. Further validation in independent and clinically diverse populations will be necessary before considering clinical use.
Perspectives
The broader vision is to bring artificial intelligence closer to the way the brain itself processes information. Rather than treating EEG as a collection of signals to classify, this approach connects measurable brain activity with computational models of the neural circuits that may generate it. The same framework could be extended to other neurological conditions, alternative brain-recording techniques, and more detailed models of brain connectivity. Combining mechanistic understanding with low-power neuromorphic computing may ultimately help researchers develop AI tools that are both scientifically informative and practical to deploy in real-world settings.
Alessandro Crimi
Universitat Zurich
Read the Original
This page is a summary of: Learning Alzheimer’s disease signatures by bridging EEG with spiking neural networks and biophysical simulations, Neurocomputing, August 2026, Elsevier,
DOI: 10.1016/j.neucom.2026.134890.
You can read the full text:
Contributors
The following have contributed to this page
