DIVER-1: Scaling Intracranial EEG Foundation Models for Transferable Representations
Researchers have developed DIVER-1, a large-scale foundation model for intracranial EEG (iEEG) data, capable of handling variable electrode layouts and recording conditions. This model was pre-trained on over 5,310 hours of iEEG data and demonstrated strong performance on cognitive decoding and seizure detection benchmarks. Separately, a new framework called NeuroFlowNet uses conditional normalizing flow to reconstruct deep temporal lobe iEEG signals from non-invasive scalp EEG, addressing limitations in understanding deep brain dynamics. Additionally, a brain-to-image system has been created that decodes visual stimuli from EEG signals, enabling both retrieval of stimulus images from EEG and reconstruction of images consistent with perceived visuals. AI
IMPACT Advances in AI models for EEG analysis offer new tools for neuroscience research and clinical diagnostics, potentially improving understanding of brain dynamics and aiding in seizure detection.