stMIDOG++ and LUNG-MITO

MIDOG++ is a large-scale annotated mitosis dataset designed for robust domain generalization in mitotic figure detection and classification. It includes 503 tumor cases spanning 10 distinct tumor domains, such as breast carcinoma, lung carcinoma, lymphoma, cutaneous mast cell tumor, neuroendocrine tumor, soft tissue sarcoma, and melanoma. The image resolution ranges from 0.23 µm/px to 0.25 µm/px, offering high-detail cellular visualization across a diverse set of histopathological samples. The original MIDOG dataset provided annotations for mitotic figures and morphologically similar hard negative instances, based on a three-expert consensus using bounding box annotations. MIDOG++ significantly extends this dataset by adding a more granular classification: each previously labeled mitotic figure is now assigned to one of five canonical mitotic subphases—prophase, prometaphase, metaphase, anaphase, or telophase—or to an atypical mitotic class for abnormal or unclassifiable figures. Additionally, manually refined segmentation masks have been added for each annotated cell to enable fine-grained morphological analysis. The mitotic figure dataset consists of human lung adenocarcinoma samples from 2 different patients and 12 slides. 1 slide from patient 1 was scanned with 3DHISTECH PANNORAMIC MIDI II, and 11 slides from patient 2 were scanned with 3DHISTECH PANNORAMIC 1000. The images had a resolution of 0.24 and 0.12 µm/px, respectively, and were tiled to 1024x1024 parts. The dataset includes two annotation files in COCO format: MIDOGpp_subtyping.json and Lung_mito.json. Each file contains detailed object-level annotations, including segmentation masks, bounding boxes, image metadata, and category definitions. All images are linked to annotations via a unique image_id field. In the MIDOG++ dataset, original filenames from the MIDOG++ repository were preserved. This file (MIDOGpp_subtyping.json) contains 26,283 annotations across 503 whole slide image crops.In the LUNG-MITO dataset, the images were anonymized and renamed using numeric identifiers (e.g., 00002.png). The Lung_mito.json file includes 5251 annotations for 2916 images Each annotation is assigned a category ID from 1 to 7, representing specific mitotic or non-mitotic classes: Category Definitions Prophase Prometaphase Metaphase Anaphase Telophase Non-mitotic Atypical mitotic figures Annotation Counts by Dataset LUNG-MITO Class 1 (Prophase): 532 Class 2 (Prometaphase): 811 Class 3 (Metaphase): 962 Class 4 (Anaphase): 112 Class 5 (Telophase): 304 Class 6 (Non-mitotic): 1113 Class 7 (Atypical mitotic figures): 1417 MIDOG++ Class 1 (Prophase): 2522 Class 2 (Prometaphase): 2458 Class 3 (Metaphase): 4112 Class 4 (Anaphase): 327 Class 5 (Telophase): 417 Class 6 (Non-mitotic): 14347 Class 7 (Atypical mitotic figures): 2100 Additional information:MIDOG++ GitHub Repository MIDOG++ Dataset Release Blog Post