A review of SPD matrix learning for neuroimaging analysis has been published. The authors propose a framework that combines classical geometric statistics with modern machine learning techniques to analyze brain activity and structure across different modalities. This approach uses Riemannian geometry to provide a non-Euclidean framework for statistical inference and learning on symmetric positive-definite matrices. The review surveys the progression from modality-specific representations to shallow and deep learning paradigms, highlighting how SPD matrix learning preserves structural constraints while extending to AI applications in neuroimaging and brain-computer interfaces. --- Why it matters: This matters because it provides a new framework for analyzing complex neuroimaging data, which can improve our understanding of brain function and structure. It also has potential applications in developing more accurate brain-computer interfaces. Source: https://arxiv.org/abs/2504.18882