Neurospectrum: A Geometric and Topological Deep Learning Framework for Uncovering Spatiotemporal Signatures in Neural Activity

Neurospectrum is a novel deep learning framework designed to analyze complex neural signals by encoding them as latent trajectories that reflect spatial and temporal structures. It utilizes graph-based representations and a manifold-regularized autoencoder to preserve temporal geometry, allowing for the extraction of multiscale geometric, topological, and dynamical features. These features are used for downstream predictions in a modular and interpretable manner. Neurospectrum has been tested on both simulated and experimental datasets, demonstrating its ability to track phase synchronization, reconstruct visual stimuli, and identify biomarkers of obsessive-compulsive disorder, outperforming traditional analysis methods. The study was supported by the National Institutes of Health and the National Science Foundation, among others.