Meet the dedicated researchers driving innovation in medical AI
PhD Student
A leading academic lab dedicated to advancing neuroscience through cutting-edge research in EEG, mental health, and neuroimaging
By advancing non-invasive brain monitoring using electroencephalography (EEG) and intracranial EEG (iEEG) to better localize epileptic foci and understand neural dynamics.
Our lab proposes a novel large language model framework focused on epilepsy, seizure activity, or the diagnosis of brain disorders. More broadly, LLMs support the discovery of biomarkers, facilitate patient stratification, and enhance the interpretation of complex neurological data. This growing field bridges artificial intelligence and brain science, offering powerful tools to advance the study and treatment of epilepsy and related neurological conditions.
Combining EEG, MEG, and structural imaging data, we develop multimodal fusion models to map brain source activity with high spatiotemporal accuracy. Our attention-based neural networks extract latent representations from multiple modalities, contributing to next-generation neuroimaging for clinical and research applications.
Our latest contributions to the field of AI Neuroimaging
This study evaluates the use of ChatGPT for localizing epileptogenic zones (EZs) from seizure semiology. ChatGPT outperformed epileptologists in commonly affected brain regions and achieved strong weighted sensitivity, supporting its clinical value in epilepsy surgery decision-making.
This paper proposes XDL-ESI, an interpretable deep learning framework for EEG source imaging. By integrating spatial attention with temporal signal representation and validating against simultaneous iEEG recordings, XDL-ESI achieves state-of-the-art source localization accuracy while offering model explainability. The framework bridges noninvasive and invasive neurophysiological data, aiding precise epileptogenic zone identification.
This paper presents a novel deep fusion framework that integrates EEG and MEG data for precise brain source localization. The proposed attention neural network architecture leverages cross-modal correlations to enhance spatial resolution and improve source imaging accuracy. Evaluated on real and simulated data, the model demonstrates robust performance and potential for clinical neuroimaging applications.
