Histo-DG-Benchmark: Domain Generalization Benchmarks for Histopathology Images Across Multiple Organs

Introduction This repository provides a collection of standardized benchmarks for domain generalization in histopathological image analysis. The resource is designed to facilitate robust evaluation across heterogeneous data sources originating from different medical centers. The collection is organized into four independent benchmarks, each corresponding to a specific organ. Within each organ-specific benchmark, four distinct domains are defined. Each domain is derived from a different publicly available dataset, which has been adapted to ensure consistency and usability in cross-domain evaluation scenarios. Due to the lack of consensus in annotation schemes across the original datasets, class labels have been harmonized. For most organs, a binary classification setup is adopted, grouping samples into: Normal/Benign Malignant (Cancerous) An exception is the prostate benchmark, where a more detailed labeling is preserved: Normal Gleason grade 3 Gleason grade 4 Gleason grade 5 This repository does not include any original data. Instead, it provides: Code for dataset adaptation and preprocessing Detailed descriptions of all modifications applied to each dataset Instructions to reproduce the processed benchmarks from the original sources The content is structured into folders, with each benchmark containing dataset-specific documentation and scripts to guide users through the reconstruction process. This resource aims to support reproducible research in domain generalization and to promote fair comparisons across methods in computational pathology. Original Datasets This work builds upon the following publicly available datasets: Breast BACH: 10.5281/zenodo.3632035 BRACS: 10.1093/database/baac093 BreaKHis: 10.21227/a6ss-5168 Burnasyan: 10.37690/1811-0193-2020-3-25-30 Colon Chaoyang dataset: 10.1109/TMI.2021.3125459 EBHI_Seg: 10.6084/m9.figshare.21540159 Kather2016: 10.5281/zenodo.53169 LC25000: 10.5281/zenodo.14998042 Lung LC25000: 10.5281/zenodo.14998042 LubLung: 10.1186/s12885-022-10081-w LungHist700: 10.6084/m9.figshare.25459174 UNN-LC: 10.18710/ZZASBA Prostate Gleason2019: 10.1016/j.media.2018.09.005 TMAGleason: 10.7910/DVN/OCYCMP CrowdGleason: 10.5281/zenodo.14178894 SICAPv2: 10.17632/9xxm58dvs3.1 All original datasets remain the property of their respective authors. Users must cite the original works when using the data.