Self-Staining Tissues for Scalable Cell Segmentation

We propose a dataset generated with a novel "self-staining" strategy, Universal Confetti, which repurposes established genetically encoded cell coloring technologies to generate highly scalable 3D cell segmentation. Crucially, our method is independent of human annotation, a costly process that has greatly limited training data availability. By pairing Universal Confetti with other imaging modalities, we propose a pan-tissue resource for training and evaluating the next generation of segmentation models. We further envision that this dataset will enable the joint embedding of morphology and molecular measurements, opening the door to more sophisticated single-cell analyses as well as integrative and generative tasks.