Cross-Subject Modeling for Widefield Calcium Imaging via Atlas-Aligned Spatiotemporal Tokenization

The paper introduces WiCAT, a self-supervised pretrained model for multi-subject widefield calcium imaging. Its atlas-grounded tokenization avoids session-specific components and aims to learn globally shared spatiotemporal representations. According to the abstract, WiCAT supports lightweight downstream decoding, transfer across subjects, tasks, and datasets, and zero-shot continuous behavior decoding and left-out brain-region reconstruction on unseen subjects. The abstract does not provide dataset sizes, numerical gains, or detailed experimental settings.