3D Liver Spheroids Brightfield 256x256 Images + ATP labels

This dataset contains 20,000 brightfield microscopy images of human liver spheroids used in the development and validation of Neural Viability Regression (NViR), a deep learning-based method for non-invasive estimation of cell viability. Each spheroid was subjected to controlled drug treatment conditions within a microphysiological system (MPS), and viability was measured using the ATP-based CellTiter-Glo® 3D Cell Viability Assay. The dataset includes: • 2,359 labeled images with matched ATP viability measurements (in nM). • ~17,500 unlabeled images used in downstream toxicity prediction tasks. • Images span multiple imaging devices (Echo Rebel, Echo Revolve, Cytation, Thorlabs) from two labs to enable testing generalization. • Each image is accompanied by metadata including experimental day, treatment type, compound name and concentration, microscope model, and (if labeled) viability value. This resource supports the development of non-invasive cell viability models and enables further research in AI-enhanced toxicology, drug safety assessment, and 3D cell culture analysis. Please cite the following paper if you use this dataset: Dubinsky et al., Non-Invasive Quantification of Viability in Spheroids Using Deep Learning, 2025. https://doi.org/10.1101/2025.03.09.64224