TumorQuantAI breast cancer immunohistochemistry patch dataset for marker quantification and reproducible paper figures

Version 1.0 is an open research dataset of 1,901 metadata-sanitized, pseudonymized, single-page 8-bit RGB TIFF fields from 51 breast cancer cases. It contains 385 H&E, 379 estrogen-receptor (ER), 409 progesterone-receptor (PR), 355 HER2, and 373 Ki-67 fields. Each TIFF has audited per-image micrometres-per-pixel calibration embedded in standard resolution tags; the public manifest records calibration provenance, geometry, whole-file SHA-256 and MD5, and a recomputed decoded-RGB SHA-256. Non-sequential case and patch aliases replace source identifiers. Together with the linked TumorQuantAI repository, the collection supports reproducible research marker quantification and paper/QC figure generation.

H&E and the four IHC markers were acquired from separate sections or fields. They are not spatially registered measurements of the same cells and must not be interpreted as same-cell co-expression. All 51 cases include H&E, ER, PR, and Ki-67 images. HER2 images are present for 50 cases; one case is explicitly not assessed for HER2 rather than inferred negative.

Research use only. These are selected fields of view, not whole-slide measurements. No pathologist-verified invasive-tumour ROI/classifier, laboratory controls, clinical outcome reference, diagnostic ground truth, or independent clinical validation is supplied. Segmented cells are measurement proxies and may include non-tumour tissue. H&E is morphology/QC only and has no expression percentage. Computational marker measurements and receptor-profile pre-scores are not diagnoses, intrinsic molecular subtypes, prognostic claims, treatment recommendations, or pathologist sign-out. HER2 membrane-proxy measurements are not clinical HER2 positivity; an equivocal 2+ pre-score requires independent ISH or another guideline-acceptable test. Selected-field sampling, staining/acquisition differences, thresholds, artifacts, and physical-scale provenance must be considered in reuse.

CC BY 4.0 applies to this dataset. TumorQuantAI software and the gated HistoPLUS model have separate terms.