TumorQuantAI colon cancer CD3, CD8, and CK20 whole-slide image dataset

This public research dataset contains 30 pseudonymized Motic MDS whole-slide images from 11 colorectal-cancer source cases: 10 CD3, 10 CD8, and 10 CK20 slides acquired at 0.261780 micrometres per pixel. Nine cases have complete CD3/CD8/CK20 serial-section sets; two incomplete cases are retained and explicitly marked unavailable.

De-identification. Public filenames and case/slide identifiers are non-semantic pseudonyms. Original archives, source filenames, scanner sidecars, identifiers, and original label/macro content are excluded. DSI0 tissue pixels are preserved byte-for-byte; non-pixel streams receive deterministic, generic, same-size neutral replacements. The release passed structural checks, multi-encoding private-marker scans, pixel-fingerprint verification, and controlled visual-review rendering.

Analysis artifacts. The record includes TumorQuantAI CK20-guided CD3/CD8 positive-cell counts, analysed areas and densities, descriptive summaries, numeric and visual registration QC, 36 paper-ready 300-dpi case/slide figures in PNG and PDF form, exact manifests and checksums, unavailable-case auditing, and an offline pathologist accept/flag/exclude adjudication dashboard with blank review CSV and codebook.

Scientific limitation. The separately named pI0-pI4 output is an exploratory within-cohort CK20-proxy rank based on six automatic-QC-pass reference cases. It is not the clinically validated consensus Immunoscore. Pathologist-reviewed tumour-core and invasive-margin regions and the validated external reference distribution were not supplied; the consensus Immunoscore field is therefore blank. These outputs are for research and technical review only and must not guide patient care.

Access and rights. Record metadata and all 57 files are publicly accessible. Public visibility does not waive copyright, imply a separate open-content license, or establish clinical validation. Users remain responsible for lawful and ethical reuse, including avoiding re-identification attempts and clinical decision-making. Copyright (C) 2026 The Authors.