Tissues with Issues Benchmark Dataset

Tissues with Issues Benchmark Dataset The Tissues with Issues Benchmark Dataset is a multi-center, multi-organ, and multi-stain collection of whole-slide images (WSIs). The dataset consists of 555 WSIs collected from 10 publicly available datasets: Acrobat Challenge, DLBCL-Morph, HISTAI, Human Protein Atlas, Monkey Challenge, Mostwiedzy, MSKCC, NADT, TCGA, and TG-GATEs. The dataset covers 29 organs and 36 staining protocols, including different specimen types (biopsies, resections, and tissue microarrays), tissue categories (normal and tumor), tissue preparation methods (FFPE and frozen tissue sections), scanning devices, and challenging imaging conditions such as pen marks, out-of-focus regions, and background artifacts. The selected slides were previously published under open-access licenses permitting redistribution. The specific license information for each slide is documented in the accompanying information file. Data Source Slide Count Organs Stains Acrobat-challenge 199 Breast ER, HE, HER2, KI67, PGR TCGA 160 Kidney, Brain, Head & Neck, Liver, Lung, Uterus, Adrenal & Paraganglia, Adrenal Gland, Bile Duct, Bladder, Breast, Colon, Connective Tissue, Esophagus, Eye (Uvea), Lymph Node, Ovary, Pancreas, Pleura, Prostate, Stomach, Testis, Thymus, Thyroid, Rectum, Skin HE Monkey-challenge 55 Kidney PAS DLBCL-Morph 42 Lymph Node BCL2, BCL6, CD10, HE, MUM1, MYC MSKCC 32 Lung PDL1 Mostwiedzy 32 Aorta, Breast, Colon, Kidney, Liver, Lung, Lymph Node, Mesentry, Ovary, Skin, Thyroid Alcian Blue+PAS, BCL2, BCL6, CD10, CD20, CD3, CD31, CD34, CD68, CMYC, ER, HER2, KI67, MUM1, Masson Goldner, Masson Trichrome, PAS, PDL1, SMA, SOX10, Sirius Red, Van Gieson, Ziehl-Neelsen NADT 21 Prostate AR, ERG, GR, Ki67, P53, PIN4, PSA, PTEN, SYP TG-GATEs 10 Kidney, Liver HE HISTAI 3 Skin, Stomach CD34, CD68, HP Human Protein Atlas 1 Multi-organ SOX10 Data contents and annotations The dataset provides TIF-formatted WSIs at a standardized starting resolution of 0.5 or 1.0 μm/pixel, together with corresponding tissue–background annotations in XML format generated at the initial image resolution. Each XML annotation file contains two annotation groups ("tissue" and "hole") represented using the Polygon or Spline annotation types. Ground-truth annotations were generated through an AI-assisted annotation workflow. Initial tissue masks were created using a previously developed in-house model (not publicly released) and converted into XML format. These annotations were subsequently refined by two annotators using the open-source annotation tool ASAP to ensure annotation quality and accuracy. Tissue–background annotations were generated following predefined annotation guidelines. Disconnected artifacts were excluded, and tissue-associated holes were annotated only when their shortest diameter exceeded 1 mm. Dataset organization Due to the large size of the dataset, the complete dataset is distributed across multiple Zenodo records. Each record contains a subset of WSIs, corresponding XML annotations, and information. The dataset is organized as follows: Part 1 of 8: [https://doi.org/10.5281/zenodo.21644817] Part 2 of 8: [https://doi.org/10.5281/zenodo.21643924] Part 3 of 8: [https://doi.org/10.5281/zenodo.21646335] Part 4 of 8: [https://doi.org/10.5281/zenodo.21648086] Part 5 of 8: [https://doi.org/10.5281/zenodo.21651695] Part 6 of 8: [https://doi.org/10.5281/zenodo.21652253] Part 7 of 8: [https://doi.org/10.5281/zenodo.21652661] Part 8 of 8: [https://doi.org/10.5281/zenodo.21652772] A information file is provided with each record, containing slide-level information including source dataset, license information, and associated metadata. Because the dataset combines images originating from multiple sources, individual files may be subject to different licensing conditions. Users should consult the information file and comply with the original license terms associated with each source dataset.