FM-Bench

Pan-cancer histopathology linked to genomic and clinical data, covering all TCGA cohorts in the FM-Bench collection. Supports organ-specific and cross-cancer studies across the full TCGA disease spectrum.

Prostate biopsy histopathology for automated cancer grading. The benchmark predicts ISUP grade groups 0–5, distinguishing non-cancerous tissue from progressively higher-grade prostate cancer.

Clinical histopathology challenge cohort. See the official challenge page for organ coverage, task definitions, and participation or data-access details.

Colorectal polyp classification and adenoma dysplasia grading. Public labels distinguish normal tissue, hyperplastic polyps, and tubular or tubulovillous adenomas with low- or high-grade dysplasia.

Clinical histopathology challenge cohort. See the official challenge page for organ coverage, task definitions, and participation or data-access details.

Colorectal cancer histopathology cohort. See the original COBRA dataset page for cohort composition, annotations, and data-access details.

Histopathology linked to proteomic, genomic, and clinical profiling, covering all CPTAC cohorts in the FM-Bench collection. Supports cross-cancer studies of tumor morphology, molecular characteristics, and clinical outcomes.

Breast lesion subtyping across benign, atypical, in situ, and invasive disease. The supplied labels are B, PB, UDH, ADH, FEA, DCIS, and IC, supporting fine-grained and grouped slide-level classification.

Colorectal liver-metastasis histopathology for treatment-response assessment, as identified in the supplied inventory. See the original release for cohort details and annotation definitions.

Brain and central nervous system histopathology for neuropathological diagnosis. Public benchmark tasks include both detailed diagnoses and broader diagnostic groups.

Cutaneous melanoma histopathology for studying prognosis and relapse. The focus is the relationship between primary-tumor morphology and subsequent clinical outcome.

Breast-cancer metastasis detection in lymph-node histopathology. Supports identifying tumor-containing slides and localizing metastatic regions within otherwise non-tumor lymph-node tissue.

Rare kidney-tumor histopathology, identified in the supplied inventory as a pediatric cohort. See the dataset card for patient coverage, tumor classes, and access details.

Breast-cancer lymph-node assessment with patient-level pathological nodal staging. Combines evidence across a patient’s lymph-node slides to classify nodal involvement rather than only detecting an individual tumor region.

Normal, non-cancer human tissue histopathology spanning multiple organs. Provides a reference for tissue morphology and for comparing normal tissue representations across anatomical sites.

Cancer Moonshot Biobank histopathology covering all CMB cohorts in the FM-Bench collection. This multi-cancer biospecimen resource supports research on cancer treatment response and resistance.

Human Cancer Models Initiative histopathology associated with cancer-model development. Connects tumor-tissue morphology with the broader study of patient-derived cancer models and their molecular characteristics.

Histopathology of blood clots retrieved from ischemic-stroke patients. The supplied task distinguishes cardioembolic (CE) from large-artery atherosclerosis (LAA) stroke etiologies using clot morphology.

Dermatopathology of cutaneous spindle-cell neoplasms, covering lesions from benign to malignant. The public dataset provides multiclass annotations from expert and trainee pathologists for studying diagnostically challenging skin tumors.

Prostate adenocarcinoma histopathology cohort in the supplied inventory. The exact public release and any Gleason-grade, treatment, or outcome labels are not established by the identifier alone.

Ovarian carcinoma histotyping from whole-slide histology. The task distinguishes major ovarian cancer histotypes, supporting evaluation of tumor-subtype recognition across morphologically distinct ovarian cancers.

Renal cell carcinoma histopathology for subtype classification. Focuses on distinguishing morphologically different kidney-cancer categories rather than treating renal tumors as a single class.

Breast-tumor proliferation assessment from histopathology. The challenge connects tissue morphology and mitotic activity with tumor-proliferation scoring.

Prostate-cancer histopathology for predicting biochemical recurrence after prostatectomy. The LEOPARD study includes follow-up data from multiple cohorts; its project repository identifies the publicly released development data.

Head-and-neck squamous cell carcinoma histopathology in a multimodal clinical-research setting. Supports studying tumor morphology alongside associated clinical information.

Prostate-cancer histopathology associated with neoadjuvant androgen-deprivation therapy, supporting treatment-response assessment.

Breast histology classified into four categories: normal, benign, in situ carcinoma, and invasive carcinoma. Supports evaluating progression-aware diagnostic classification across benign and malignant tissue patterns.

Automated Gleason grading in prostate whole-slide histopathology. The challenge evaluates recognition of prostate-cancer growth patterns for grading.

Epithelial lesions in cervical biopsies. The challenge classifies the most severe lesion present into four categories: benign, low malignant potential, high malignant potential, or invasive cancer.

Lung adenocarcinoma histopathology from Lee Lab, Kunming Medical University. The public release contains 408 Aperio whole-slide images (.svs), scanned at 20×/40× (approximately 450–500 GB). Tasks follow WHO 2015/2021 guidelines: invasive adenocarcinoma (IA), minimally invasive adenocarcinoma (MIA), and adenocarcinoma in situ (AIS); predominant lepidic, acinar, papillary, micropapillary, and solid growth patterns; and mucinous versus non-mucinous status.

Lung histopathology associated with the National Lung Screening Trial. This inventory currently contains segmented slides only.

High-grade serous ovarian cancer histopathology from the OncoFusion multimodal risk-stratification study. The original study integrates whole-slide histology, CT imaging, genomic data, and clinical variables for overall-survival modeling.

Colorectal-cancer liver-metastasis histopathology in the supplied inventory. Intended for studying metastatic tumor morphology in a clinical-treatment setting; the exact public release and available response or survival labels need confirmation.

Ovarian-cancer histopathology associated with bevacizumab treatment response, supporting investigation of relationships between tumor morphology and therapeutic outcomes.

Lung histopathology for adenocarcinoma and squamous-cell carcinoma subtyping. See the original Dartmouth dataset page for cohort details and label definitions.

Breast-cancer histopathology focused on tumor-infiltrating lymphocytes (TILs). Supports studying the immune microenvironment and the relationship between immune-cell presence and tumor tissue.

Colorectal-cancer histopathology with survival and genetic-marker annotations. Public data include KRAS, NRAS, BRAF, mismatch-repair status, and five-year survival information, with primary colorectal tumors and metastatic sites represented.

Liver-cancer histopathology focused on hepatocellular carcinoma. Supports evaluating tumor-region recognition and segmentation within whole-slide tissue images.

Colorectal-cancer histopathology for microsatellite-instability (MSI) prediction. Relates tumor morphology to molecular instability status rather than only to a visual histological subtype.