Non-Invasive Reconstruction of Intracranial EEG Across the Deep Temporal Lobe from Scalp EEG based on Conditional Normalizing Flow

A research team led by Dongyi He has introduced NeuroFlowNet, a groundbreaking cross-modal generative framework that reconstructs intracranial electroencephalography (iEEG) signals from non-invasive scalp EEG (sEEG) data. This represents the first successful reconstruction of iEEG signals from the entire deep temporal lobe region using only surface measurements, addressing a critical challenge in neuroscience where traditional methods like signal processing and source localization have struggled to capture the complex waveforms and random characteristics of deep brain activity. The breakthrough enables researchers to study deep brain dynamics without requiring invasive surgical implants, opening new possibilities for understanding neurological conditions and brain function.

NeuroFlowNet's technical innovation lies in its use of Conditional Normalizing Flow (CNF), which directly models complex conditional probability distributions through reversible transformations, explicitly capturing the inherent randomness of brain signals while avoiding the pattern collapse issues common in other generative models like GANs. The architecture integrates multi-scale processing and self-attention mechanisms to capture both fine-grained temporal details and long-range dependencies in brain activity patterns. Validation on publicly available synchronized sEEG-iEEG datasets demonstrates the model's effectiveness in temporal waveform fidelity, spectral feature reproduction, and functional connectivity restoration. This establishes a more reliable and scalable paradigm for non-invasive deep brain analysis, with code already available for the research community to build upon.