Recently added (10)#
Bio-image Data Science Lectures 2025 @ Uni Leipzig / ScaDS.AI#
Robert Haase
Published 2025-05-29
Licensed CC-BY-4.0
These are the PPTx training resources for Students at Uni Leipzig who want to dive into bio-image data science with Python. The material will develop here and in the corresponding github repository between April and July 2025.
CellBinDB: A Large-Scale Multimodal Annotated Dataset#
Can Shi, Jinghong Fan, Zhonghan Deng, Huanlin Liu, Qiang Kang, Yumei Li, Jing Guo, Jingwen Wang, Jinjiang Gong, Sha Liao, Ao Chen, Ying Zhang, Mei Li
Published 2024-11-20
Licensed CC-ZERO
CellBinDB is a large-scale, multimodal annotated dataset for cell segmentation. It contains 1,044 annotated microscope images and 109,083 cell annotations, covering four staining types: DAPI, ssDNA, H&E, and mIF. CellBinDB contains samples from two species, human and mouse, covering more than 30 histologically different tissue types, including disease-related tissues. The images in CellBinDB come from two sources: 844 mouse images from internal experiments and 200 human images from the open access platform 10x Genomics. We annotated all images in CellBinDB and provide two types of image annotations: semantic and instance masks. A xlsx file is attached to record the detailed information of each image. In addition, we provide the images and annotations of nine other widely used publicly available cell segmentation datasets downloaded from their original sources, retaining their original formats for ease of use. The file ‘mixed_licenses.txt’ contains the original accessions of the public datasets used in our project and their associated licenses. Please refer to these links for more information about each dataset and its licensing terms, and use it according to the specifications.
Tags: Ai-Ready
Content type: Data
Collaborative working and Version Control with Git[Hub]#
Robert Haase
Published 2025-05-10
Licensed CC-BY-4.0
Working together on the internet presents us with new challenges: Who uploaded a file and when? Who contributed to the project when and why? How can content be merged when multiple team members make changes at the same time? The version control tool Git offers a comprehensive solution to these questions. The online platform GitHub.com provides a Git-driven platform that enables effective collaboration. Attendees of this hands-on tutorial will learn:
Introduction to version control with Git[Hub]
Working with Git: Pull requests
Resolving merge conflicts
Artificial intelligence that can respond to GitHub issues
Explainable AI for Computer Vision#
Robert Haase
Published 2025-03-09
Licensed CC-BY-4.0
In this slide deck we learn about the basics of Explainable Artificial Intelligence with a soft focus on Computer Vision. We take a deeper dive in one method: Gradient Class Activation Maps. Releated exercise materials are available online: https://haesleinhuepf.github.io/xai/
HT1080WT cells embedded in 3D collagen type I matrices - manual annotations for cell instance segmentation and tracking#
Estibaliz Gómez-de-Mariscal, Hasini Jayatilaka, Denis Wirtz, Arrate Muñoz-Barrutia
Published 2021-12-13
Licensed CC-BY-4.0
Human fibrosarcoma HT1080WT (ATCC) cells at low cell densities embedded in 3D collagen type I matrices [1]. The time-lapse videos were recorded every 2 minutes for 16.7 hours and covered a field of view of 1002 pixels × 1004 pixels with a pixel size of 0.802 μm/pixel The videos were pre-processed to correct frame-to-frame drift artifacts, resulting in a final size of 983 pixels × 985 pixels pixels.
Hasini Jayatilaka, Anjil Giri, Michelle Karl, Ivie Aifuwa, Nicholaus J Trenton, Jude M Phillip, Shyam Khatau, and Denis Wirtz. EB1 and cytoplasmic dynein mediate protrusion dynamics for efficient 3-dimensional cell migration. FASEB J., 32(3):1207–1221, 2018. ISSN 0892-6638. doi: 10.1096/fj.201700444RR.
Further information about how to use this data is given in esgomezm/microscopy-dl-suite-tf
This dataset is provided together with the following preprint and if you use it, we would like to kindly ask you to cite it properly:
Estibaliz Gómez-de-Mariscal, Hasini Jayatilaka, Özgün Çiçek, Thomas Brox, Denis Wirtz, Arrate Muñoz-Barrutia, Search for temporal cell segmentation robustness in phase-contrast microscopy videos, arXiv 2021 (arXiv:2112.08817)
Tags: Ai-Ready
Content type: Data
Learning and Training Bio-image Analysis in the Age of AI#
Robert Haase
Published 2025-04-07
Licensed CC-BY-4.0
The advent of large language models (LLMs) such as ChatGPT changes the way we analyse images. We ask LLMs to generate code, apply it to images and spend less time on learning implementation details. This also has impact on how we learn image analysis. While coding skills are still required, we can use LLMs to explain code, make proposals how to analyse the images and yet still decide how the analysis is done.
LyNSeC: Lymphoma Nuclear Segmentation and Classification#
Naji Hussein, Büttner Reinhard, Simon Adrian, Eich Marie-Lisa, Lohneis Philipp, Bozek Katarzyna
Published 2023-06-21
Licensed CC-BY-4.0
Over the last years, there has been large progress in automated segmentation and classification methods in histological whole slide images (WSIs) stained with hematoxylin and eosin (H&E). Current state-of-the-art techniques are based on diverse datasets of H&E-stained WSIs of different types of predominantly solid cancer. However, there is a lack of publicly available annotated datasets of lymphoma, which is why we generated a labeled diffuse large B-cell lymphoma dataset and denoted it LyNSeC (lymphoma nuclear segmentation and classification). LyNSeC comprises three subsets: LyNSeC 1 consists of 379 IHC images of size 512 x 512 pixels at 40x magnification. In the images, we annotated the contours of each cell nuclei and the cell class: marker-positive or marker-negative.
In total, LyNSeC 1 contains 87,316 annotated cell nuclei of four different cases, with 48,171 of them assigned the class negative and 39,145 positive. We included three markers in this dataset showing visually different staining patterns: cluster of differentiation 3 (CD3), Ki67 as a marker of proliferation, and erythroblast transformation-specific (EST)-related gene (ERG).
LyNSeC 2 and 3 contain H&E-stained images of 70 different patients. LyNSeC 2 consists of 280 images and LyNSeC 3 of 40 images of size 512 x 512 pixels at 40x magnification. 65,479 and 8,452 nuclei were annotated in LyNSeC 2 and 3, respectively. In LyNSeC 3, the nuclei were also assigned a class label (tumor and non-tumor). 3,747 nuclei were identified as tumors and 4,705 as non-tumors.
In the annotation procedure, the contours of the H&E images (LyNSeC 2 and LyNSeC 3) were annotated by two pathologists and by two students (trained by the pathologists). Annotation of the cell classes in LyNSeC 3 was done by the pathologists only. LyNSeC 1 was annotated by the two students who were additionally trained to annotate the contours and to distinguish marker-positive and marker-negative cells. The pathologists inspected and (if necessary) adjusted the LyNSeC 3 annotations.
The files are uploaded in ‘.npy’ format. The files of LyNSeC 1 (x_l1.npy) and LyNSeC 3 (x_l3.npy) contain five channels, respectively: the first three are the RGB channels of the images, channel 4 contains the instance maps, and channel 5 the class type maps (for LyNSeC 1 a pixel value of 1 corresponds to the class negative and 2 to the class positive, whereas in LyNSeC 3 1 corresponds to the class non-tumor and 2 to the class tumor). The files of LyNSeC 2 (x_l2.npy) have 4 channels (without the class type map).
Additionally, we also make our HoVer-Net-based pre-trained nuclei segmentation and classification models available (he.tar for H&E images and ihc.tar for IHC images).
Tags: Ai-Ready
Content type: Data
Melanoma Histopathology Dataset with Tissue and Nuclei Annotations#
Mark Schuiveling
Published 2025-03-19
Licensed CC-ZERO
Description: This dataset is designed for development of deep learning models for segmentation of nuclei and tissue in melanoma H&E stained histopathology. Existing nuclei segmentation models that are trained on non-melanoma specific datasets have low performance due to the ability of melanocytes to mimic other cell types, whereas existing melanoma specific models utilize older, sub-optimal techniques. Moreover, these models do not provide tissue annotations necessary for determining the localization of tumor-infiltrating lymphocytes, which may hold value for predictive and prognostic tasks. To address this, we created a melanoma specific dataset with nuclei and tissue annotations. Methodology: Sample Collection: Regions of interest (ROIs) were sampled from H&E stained slides of 103 primary melanoma specimens and 102 metastatic melanoma specimens, scanned using a Hamamatsu scanner at 40× magnification (0.23 μm per pixel). All slides were obtained from regular diagnostic procedures.From each specimen, a 40× magnified ROI of 1024×1024 pixels was selected for annotation. Additionally, a context ROI of 5120×5120 pixels was sampled to provide information about the broader context for the annotation process. Selection was performed by a trained medical expert (M.S.) and subsequently verified by a dermatopathologist (W.B.). Manual ROI selection ensured the inclusion of diverse tissue and nuclei types. Annotation Process:
Nuclei segmentationNuclei segmentations were generated using Hover-Net pretrained on the PanNuke dataset. Manual annotation adjustments were performed by author M.S. using QuPath, with the following nuclei categories: tumor, stroma, vascular endothelium, histiocyte, melanophage, lymphocyte, plasma cell, neutrophil, apoptotic cell, and epithelium. All annotations were reviewed and corrected, where needed, by a dermatopathologist (W.B.). Tissue segmentationTissue segmentations were created manually using QuPath by M.S., with the following categories: tumor, stroma, epidermis, necrosis, blood vessel, and background. Annotations were reviewed and corrected, where needed, by a dermatopathologist (W.B.).
Quality Control: To assess the reliability of the annotations, intra- and interobserver agreement (by pathologist G.B.) were determined on 12 randomly selected ROIs.
Nuclei segmentationThe intraobserver overall precision was 84.89%, with a recall of 86.45%, and an F1 score of 85.66%. Interobserver overall precision was 80.34%, with a recall of 80.62%, and an F1 score of 80.20%. These results are based on the sum of all true positive, false positive, and false negative counts for the 12 ROIs. Tissue segmentationThe DICE score was determined on the same 12 randomly selected ROIs. The average intraobserver DICE score was 0.90, and the interobserver DICE score was also 0.90.
Version 3:Removed sample “training_set_metastatic_roi_103” due to inconsistencies in annotation file. Version 4:Sample training_set_metastatic_roi_088 missed one color annotation for a nuclei_apoptosis in the geojson file rendering it qupath uncompatible. This is fixed in the new version. Version 5:Addition of correct sample of training_set_metastatic_roi_103” after deadline of panoptic segmentation of nuclei and tissue in advanced melanoma challenge test phase.
Tags: Ai-Ready
Content type: Data
NeurIPS 2022 Cell Segmentation Competition Dataset#
Jun Ma, Ronald Xie, Shamini Ayyadhury, Cheng Ge, Anubha Gupta, Ritu Gupta, Song Gu, Yao Zhang, Gihun Lee, Joonkee Kim, Wei Lou, Haofeng Li, Eric Upschulte, Timo Dickscheid, de Almeida, José Guilherme, Yixin Wang, Lin Han, Xin Yang, Marco Labagnara, Vojislav Gligorovski, Maxime Scheder, Rahi, Sahand Jamal, Carly Kempster, Alice Pollitt, Leon Espinosa, Tam Mignot, Middeke, Jan Moritz, Jan-Niklas Eckardt, Wangkai Li, Zhaoyang Li, Xiaochen Cai, Bizhe Bai, Greenwald, Noah F., Van Valen, David, Erin Weisbart, Cimini, Beth A, Trevor Cheung, Oscar Brück, Bader, Gary D., Bo Wang
Published 2024-02-27
Licensed CC-BY-NC-ND-4.0
The official data set for the NeurIPS 2022 competition: cell segmentation in multi-modality microscopy images. https://neurips22-cellseg.grand-challenge.org/ Please cite the following paper if this dataset is used in your research. @article{NeurIPS-CellSeg, title = {The Multi-modality Cell Segmentation Challenge: Towards Universal Solutions}, author = {Jun Ma and Ronald Xie and Shamini Ayyadhury and Cheng Ge and Anubha Gupta and Ritu Gupta and Song Gu and Yao Zhang and Gihun Lee and Joonkee Kim and Wei Lou and Haofeng Li and Eric Upschulte and Timo Dickscheid and José Guilherme de Almeida and Yixin Wang and Lin Han and Xin Yang and Marco Labagnara and Vojislav Gligorovski and Maxime Scheder and Sahand Jamal Rahi and Carly Kempster and Alice Pollitt and Leon Espinosa and Tâm Mignot and Jan Moritz Middeke and Jan-Niklas Eckardt and Wangkai Li and Zhaoyang Li and Xiaochen Cai and Bizhe Bai and Noah F. Greenwald and David Van Valen and Erin Weisbart and Beth A. Cimini and Trevor Cheung and Oscar Brück and Gary D. Bader and Bo Wang}, journal = {Nature Methods}, volume={21}, pages={1103–1113}, year = {2024}, doi = {https://doi.org/10.1038/s41592-024-02233-6} } This is an instance segmentation task where each cell has an individual label under the same category (cells). The training set contains both labeled images and unlabeled images. You can only use the labeled images to develop your model but we encourage participants to try to explore the unlabeled images through weakly supervised learning, semi-supervised learning, and self-supervised learning. The images are provided with original formats, including tiff, tif, png, jpg, bmp… The original formats contain the most amount of information for competitors and you have free choice over different normalization methods. For the ground truth, we standardize them as tiff formats. We aim to maintain this challenge as a sustainable benchmark platform. If you find the top algorithms (https://neurips22-cellseg.grand-challenge.org/awards/) don’t perform well on your images, welcome to send us the dataset (neurips.cellseg@gmail.com)! We will include them in the new testing set and credit your contributions on the challenge website! Dataset License: CC-BY-NC-ND
Tags: Ai-Ready
Content type: Data
OCELOT: Overlapped Cell on Tissue Dataset for Histopathology#
Jeongun Ryu, Aaron Valero Puche, JaeWoong Shin, Seonwook Park, Biagio Brattoli, Mohammad Mostafavi, Jinhee Lee, Sérgio Pereira, Wonkyung Jung, Soo Ick Cho, Chan-Young Ock, Kyunghyun Paeng, Donggeun Yoo
Published 2023-03-23
The OCELOT dataset is a histopathology dataset designed to facilitate the development of methods that utilize cell and tissue relationships. The dataset comprises both small and large field-of-view (FoV) patches extracted from digitally scanned whole slide images (WSIs), with overlapping regions. The small and large FoV patches are accompanied by annotations of cells and tissues, respectively. The WSIs are sourced from the publicly available TCGA database and were stained using the H&E method before being scanned with an Aperio scanner.
For more details, please check https://lunit-io.github.io/research/ocelot_dataset/.
Before downloading the dataset, please make sure to carefully read and agree to the Terms and Conditions at (https://lunit-io.github.io/research/ocelot_tc/).
Also, please provide 1. name, 2. e-mail address, 3. organization/company name.
Release note.
In version 1.0.1, we exclude four test cases (586, 589, 609, 615) due to under-annotated issue. In version 1.0.0, we include images and annotations of validation and test splits. In version 0.1.2, we modified the coordinates of cell labels to range from 0 to 1023 (-1 from the previous coordinates). In version 0.1.1, we removed non-H&E stained patches from the dataset.
Tags: Ai-Ready
Content type: Data
https://zenodo.org/records/8417503
https://doi.org/10.5281/zenodo.8417503