OrgLine

Dataset Description: A Curated and Refined Benchmark for Organoid Analysis Dataset Overview This repository provides a standardized benchmark dataset for organoid detection and instance segmentation, curated from multiple high-quality public repositories. To ensure the reliability of deep learning models trained on this data, we have unified the data formats (e.g., COCO/YOLO) and performed systematic quality control. This involved refining loose annotations, supplementing missing labels, and converting annotation formats where necessary. Part 1: Object Detection Data & Refinements For the object detection benchmark, we aggregated images from several sources with the following specific enhancements: Intestine Organoids: Data was sourced from Tellu (Domènech-Moreno et al.), OrgaQuant (Kassis et al.), and OrgaSegment (Lefferts et al.). For the subset sourced from Tellu, we identified that some original bounding boxes contained excessive background margins. We performed manual annotation refinement on these samples, "tightening" the bounding boxes to strictly enclose the organoid instances, thereby improving spatial precision for training. Lung Spheroids: Data was sourced from DeepLUMEN (Abdul et al.). The original annotations primarily focused on classifying "lumen" vs. "no lumen" structures and omitted many other organoids in the field of view. We significantly extended this dataset by manually labeling these previously ignored, non-defocused organoid instances to provide a comprehensive detection ground truth. Brain Organoids: Data was sourced from Schröter et al. and standardized into the unified detection format. Part 2: Instance Segmentation Data & Refinements For the instance segmentation benchmark, we processed masks from the following sources: PDAC (Pancreatic Ductal Adenocarcinoma): Data was sourced from OrganoID (Matthews et al.) and OrganoidNet (Ferreira et al.). A key modification was made to the OrganoID subset, which originally provided semantic segmentation masks. We manually processed and converted these into instance-level segmentation masks, effectively separating individual connected instances to support instance segmentation tasks. Other Organs: Intestine masks were curated from OrgaSegment; Brain masks from Schröter et al.; and Colon masks from OrgaExtractor. All were standardized to a compatible format. References and Original Sources Users are requested to cite both this repository and the original publications listed below: Organoid Type Original Dataset Reference URL / DOI Intestine OrgaQuant Kassis et al. (2019) https://osf.io/etz8r Intestine OrgaSegment Lefferts et al. (2024) https://doi.org/10.5281/zenodo.10278229 Intestine Tellu Domènech-Moreno et al. (2023) https://doi.org/10.5281/zenodo.6768583 Brain Brain Organoid Schröter et al. (2024) https://doi.org/10.5281/zenodo.10301912 Lung DeepLUMEN Abdul et al. (2021) https://osf.io/g2a7r/ Colon OrgaExtractor Park et al. (2023) https://github.com/tpark16/orgaextractor PDAC OrganoID Matthews et al. (2022) https://osf.io/xmes4/ PDAC OrganoidNet Ferreira et al. (2025) https://zenodo.org/records/10643410 Usage & License The original images and annotations remain the property of their respective authors and are distributed under their original licenses. The refined annotations provided in this repository are shared to facilitate reproducibility and community research. Acknowledgements We gratefully acknowledge the original authors and maintainers of the public datasets used in this work. All original images and annotations remain the property of their respective authors and are used in accordance with their licenses.