Multi-view manifold learning of human brain-state trajectories

The complexity of the human brain gives illusion that activity is intrinsically high-dimensional. Nonlinear dimensionality-reduction methods such as uniform manifold approximation and t-distributed stochastic neighbor embedding have been used for high-throughput biomedical data. However, they not extensively data those from functional magnetic resonance imaging (fMRI), primarily due to their inability maintain dynamic structure. Here we introduce a nonlinear learning method time-series data—including fMRI—called temporal potential heat-diffusion affinity-based transition (T-PHATE). In addition recovering low-dimensional intrinsic geometry data, T-PHATE exploits data’s autocorrelative structure faithfully denoise unveil trajectories. We empirically validate on three fMRI datasets, showing it greatly improves visualization, classification, segmentation relative several other state-of-the-art benchmarks. These improvements suggest many applications high-dimensional datasets temporally diffuse processes. A called developed applied (functional imaging) where denoises signals unveils latent brain-state trajectories which correspond with cognitive processing.