Open Data Resources

Highly multiplexed 3D cyclic immunofluorescence (CyCIF) imaging of human tumor tissue, enabling spatial profiling of cell states and immune niches at single-cell resolution in three dimensions. Hosted through the Human Tumor Atlas Network (HTAN) tissue-atlas portal.

High-resolution light-sheet fluorescence microscopy volumes of cells exhibiting complex protrusive morphologies (blebs, filopodia, and lamellipodia), used to benchmark and demonstrate the u-segment3D universal 3D cell segmentation pipeline. Provided as example data accompanying the software release.

Multi-centre benchmark dataset for artefact detection and segmentation in clinical endoscopy, aggregated from six facilities across the UK, France, Italy, Switzerland, and Russia. Covers gastroscopy, cystoscopy, and colonoscopy across white-light, fluorescence, and narrow-band imaging modalities. Supports three challenge tasks: multi-class bounding-box detection, region segmentation, and cross-domain generalisation.

Time-lapse microscopy videos of two co-cultured cell populations undergoing collective migration, used to develop and validate the Motion Sensing Superpixels (MOSES) computational framework for quantifying and discovering cellular motion phenotypes.

Time-lapse microscopy videos documenting the response of EPC2 and CP-A oesophageal cell lines to EGF stimulation, capturing changes in collective migration phenotype used to benchmark the MOSES motion analysis pipeline.

Computationally simulated organoid conditions generated as ground-truth benchmarks for evaluating organoid analysis and segmentation methods. Two datasets of varying scale (~19 GB and ~800 MB) paired with a detailed PDF of simulation parameters, enabling rigorous, reproducible assessment of computational pipelines without requiring manual annotation of real imaging data.

Two augmented shape-texture benchmark collections derived from the MPEG-7 corpus and Brodatz textures, created to evaluate the SAM phenomics framework. The first applies binary dilation and topological in-filling to improve mask contiguity. The second combines MPEG-7 shape classes with normalised Brodatz texture features, yielding 10,000 synthetic image patches for systematic benchmarking of appearance descriptors.

Four high-resolution 3D light-sheet fluorescence microscopy datasets used for the main biological demonstrations in the surface-guided computing / u-unwrap3D paper. Covers diverse morphodynamic phenotypes: K14⁺ breast organoid (3.5 GB), NK–cancer cell interaction (8.9 GB), MV3 melanoma blebbing cell (3.1 GB), and ruffling SU8686 pancreatic cancer cell (2.4 GB). All provided as compressed tar.xz archives.