# PDR 010: Ingest the Zenodo skin frozen and FFPE preview

- HTML version: https://robbiepalmer.me/projects/pathologize/pdrs/010-ingest-skin-frozen-ffpe-preview
- Project: Pathologize
- Status: Rejected
- Opened: 2026-11-11
- Decided: 2026-11-11

## Context

Zenodo record `18780726`, DOI `10.5281/zenodo.18780726`, describes five
de-identified frozen and FFPE whole-slide image pairs from skin cancer resection
margins. The cases cover extramammary Paget disease, Bowen disease, invasive
squamous cell carcinoma, basal cell carcinoma, and melanoma. Unsuffixed names
identify frozen preparations. A `P` suffix identifies FFPE preparations.

The release has 30 files and a total size of 418,147 bytes. Each preparation has
an `.mrxs` file, a metadata XML, and an annotation XML. Zenodo applies CC BY 4.0
to the release. The related paper uses CC BY-NC-ND 4.0.

The paper reports a separate clinical cohort of 2,594 slides from 283 patients.
It contains 1,192 frozen and FFPE slide pairs and 210 additional main-tumour FFPE
slides. The paper also reports 60 frozen-to-FFPE diagnostic discrepancies, a
55-case expert reassessment, and an example where a deeper FFPE section contained
tumour that was barely present in the frozen section. These facts explain why the
source looked useful, but they do not describe image content in the five-case
Zenodo preview. The full cohort remains subject to institutional governance and
patient privacy. The authors offer access by reasonable request.

## Payload investigation

On 11 November 2026, the source-product investigation downloaded all 30 Zenodo
objects to temporary storage. Every object matched its published byte count and
MD5 checksum.

The ten `.mrxs`-named objects ranged from 25,838 to 86,312 bytes. Each object was
a valid 698 by 1,600 JFIF JPEG. None was a MIRAX descriptor. The record contained
no same-named companion directory with MIRAX indexes or image tiles. The ten
annotation XML files contained empty `annotations` elements.

OpenSlide Python 1.4.7 with OpenSlide 4.0.1 attempted to open every `.mrxs`-named
object. All ten attempts raised `OpenSlideUnsupportedFormatError`. The JPEGs can
act as small visual previews, but they cannot provide WSI regions, pyramid levels,
or diagnostic-resolution pixels.

## Decision

Reject the proposed source product for this release. Pathologize will not publish
a DVC product, native records, derived comparisons, or SOP inputs from these
files. Re-encoding a 698 by 1,600 JPEG as TIFF or another container would preserve
the same low-resolution preview and would not recover the missing pixels.

Keep the investigation in this rejected PDR and the associated Work Graph record.
The Zenodo pair names remain publisher claims. They do not establish spatial
registration or identical cutting planes.

## Consequences

The current release cannot support frozen-artifact analysis, diagnostic
comparison, section-depth analysis, model development, or whole-slide viewing.
The paper remains useful background evidence, but Pathologize will not treat paper
cohort counts or workflow observations as preview data.

Rejecting the product also avoids a misleading metadata-only node in the data
mesh. A source product should supply material that at least one intended consumer
can read and use.

## Reconsideration

Reconsider ingestion if the publisher releases either complete MIRAX descriptors
and companion directories or another full-resolution WSI format. A replacement
must pass checksum verification, format detection, reader open, pyramid geometry,
and bounded pixel-read tests. Review its rights and access terms again, and retain
the five source case and preparation identities without assuming registration.

Access to the governed full cohort would require a separate decision. It must not
be substituted silently for this public preview.

## Related records

- Project: [Pathologize](https://robbiepalmer.me/projects/pathologize.md)
- Evidence: [Zenodo skin frozen and FFPE preview record](https://zenodo.org/records/18780726)
- Evidence: [Translation of frozen sections into FFPE images for skin cancer resection margins using generative AI](https://doi.org/10.1038/s41746-026-02939-1)
- Evidence: [OpenSlide MIRAX format documentation](https://openslide.org/formats/mirax/)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt