A Multi-Center Breast FNAC Cytology Dataset for AI-Assisted Patch-wise Classification Using C1-C5 Reporting Categories

Breast fine needle aspiration cytology (FNAC) is a rapid, minimally invasive, and cost-effective method for evaluating breast lesions. Despite its clinical utility, FNAC interpretation remains dependent on expert cytopathological review and can be affected by variation in smear preparation, staining, cellularity, and overlapping cytomorphological features. Development of reliable artificial intelligence (AI)-assisted cytology systems requires large, diverse, and well-annotated datasets. We present a multi-center breast FNAC cytology dataset for AI-assisted patch-wise classification using C1--C5 reporting categories. The dataset was collected prospectively from participating tertiary medical centers in India between May 2023 and March 2026, with ethical approval from the participating centers and the Indian Council of Medical Research (ICMR). It contains 321 patients and 470 whole slide images (WSIs), including 190 Papanicolaou (PAP)-stained and 280 May--Grunwald--Giemsa (MGG)-stained slides. The WSIs were scanned using a Hamamatsu whole slide scanner at 40$\times$ magnification and 0.25 microns per pixel. The dataset includes 7,393 manually extracted diagnostically relevant image patches labeled as C1 insufficient, C2 benign, C3 atypical, C4 suspicious for malignancy, or C5 malignant. Patch labels were assigned by a pathologist and verified by a senior pathologist. The released data will include WSIs, GeoJSON annotation files with patch coordinates and C1--C5 labels, extracted patch images, anonymised patient-level metadata, and patch-level metadata. The complete dataset is approximately 950 GB and will be made available as an open-access resource. This dataset provides a resource for developing and benchmarking AI methods for breast cytology classification and future slide-level decision-support workflows.

WSIs are also available for download.
Set 1 (26 WSIs): https://zenodo.org/record/20701935 
Set 2 (19 WSIs): https://zenodo.org/record/20701937
Set 3 (25 WSIs): https://zenodo.org/record/20701939
Set 4 (22 WSIs): https://zenodo.org/record/20701943
Set 5 (18 WSIs): https://zenodo.org/record/20701945
Set 6 (17 WSIs): https://zenodo.org/record/20701947
Set 7 (28 WSIs): https://zenodo.org/record/20701949
Set 8 (26 WSIs): https://zenodo.org/record/20701951
Set 9 (36 WSIs): https://zenodo.org/record/20701953
Set 10 (24 WSIs): https://zenodo.org/record/20701955
Set 11 (26 WSIs): https://zenodo.org/record/20701958
Set 12 (21 WSIs): https://zenodo.org/record/20701960
Set 13 (18 WSIs): https://zenodo.org/record/20701962
Set 14 (20 WSIs): https://zenodo.org/record/20701964
Set 15 (18 WSIs): https://zenodo.org/record/20761031
Set 16 (18 WSIs): https://zenodo.org/record/20701968
Set 17 (28 WSIs): https://zenodo.org/record/20701970
Set 18 (20 WSIs): https://zenodo.org/record/20701972
Set 19 (26 WSIs): https://zenodo.org/record/20701976
Set 20 (28 WSIs): https://zenodo.org/record/20701978
Set 21 (6 WSIs): https://zenodo.org/record/20701980