WSI Format Conversion

Rewriting a slide file into a format the tool you need can actually open — what survives the trip, what silently does not, and why the standard everyone is told to adopt is the one most viewers cannot read.

Rewriting a slide file into a format the tool you need can actually open — what survives the trip, what silently does not, and why the standard everyone is told to adopt is the one most viewers cannot read.

What it is

Whole Slide Imaging describes the file. This page is what happens when the file is in the wrong format for the thing you want to do with it, which in practice is often.

The landscape is a standard plus a set of vendor formats that predate it. DICOM is the interoperability standard, and departments digitising for routine diagnosis are urged to adopt it. SVS (Aperio/Leica), NDPI (Hamamatsu), MRXS (3DHistech), VSI (Olympus) are what scanners actually emit and what the free vendor viewers actually read.

The gap between those two sentences is the whole subject. Surveying freely available viewers, Chauveau 2025 — Converting WSI from DICOM to SVS-like TIFF found only one of six vendors’ free viewers reads DICOM — 3DHistech’s. Aperio ImageScope does not. NDP.view2 does not. OlyVia does not. Collaborative-diagnosis and teaching platforms are the same, and on the compute side RAPIDS cuCIM accelerates only SVS and generic tiled TIFF, so a DICOM cohort silently falls back to the slower OpenSlide path.

So conversion is not a niche concern. It is the tax on adopting the standard.

What survives a conversion, and what does not

The useful distinction is not “lossy versus lossless” but which of the several things inside a slide file survive. A slide is a pyramid of tiles plus a label image, a macro image, an ICC profile, and a pile of metadata; each can be preserved, dropped or silently mangled independently.

A conversion worth trusting does this:

• Copies encoded tiles without decompressing them. If JPEG tiles are decoded and re-encoded, the result is generationally degraded even though both ends are “JPEG”. Copying the compressed bytes through is what makes a conversion genuinely lossless, and it is also why it is fast — roughly 18 s per GB on an ordinary workstation, 3 s per GB on current Apple silicon.
• Preserves every pyramid level rather than regenerating them by downsampling, which would change what a reader at 10× sees.
• Carries the ICC profile across. See below; this is the one most often lost.
• Keeps tile size unchanged, since downstream tiling code may assume it.

And these are the traps:

Sparse tiling does not survive into every format. DICOM can store only the tiles containing tissue and record their positions as metadata. SVS cannot express that, so the background has to be filled in. On slides that are mostly background the file grows — measured at +75% mean and up to +260% on a slide over 90% background, against +2.2% for normally tiled slides. At archive scale that is the number that decides whether a conversion strategy is affordable, and it depends on a property of the source file most people never look at.

Unsupported cases should be skipped loudly, not converted badly. Multiplexed immunofluorescence, multiplanar/z-stacked brightfield and unusual compression schemes are cases where a converter should refuse. A converter that silently flattens a z-stack has destroyed the thing that made the slide worth scanning — which matters directly for Cervical Cytology Screening, where the z-stack is the point.

The ICC profile, and the QuPath finding

An ICC profile is a third mechanism for handling colour, distinct from the two on Colour Calibration: not a physical correction at the scanner and not a software normalisation afterwards, but a colour-space mapping embedded in the file that travels with it, telling a viewer how to render the scanner’s native colours faithfully.

It only works if the software reads it. Chauveau 2025 records, in a table footnote, that QuPath has no out-of-the-box ICC profile support — which is why the author calls it software “dedicated more to WSI analysis rather than WSI viewing”. The same paper illustrates the difference on an Aperio GT450 DX image, a scanner for which ICC support is described as essential.

For a group whose tooling runs through QuPath Annotation Workflow on Aperio scanners, the consequence is concrete: the colour on screen in QuPath is not the colour the scanner intended. For drawing annotations and running analysis that is tolerable. For any judgement a human makes about colour, for teaching images, and for anything touching Colour Calibration or Stain Normalisation, it is a confounder that sits upstream of the method being studied.

Conversion is the natural place to anonymise

A converter is already rewriting every byte and re-deciding which sub-images to include, so dropping the label and macro images and stripping acquisition metadata costs nothing extra there. Chauveau’s tool takes exactly that option — rename, drop label and macro, strip sensitive tags, in the same pass.

This is worth knowing because it is the cheapest correctly-ordered de-identification available: De-identification describes three surfaces that each fail differently, and conversion touches all three at once. The caveat from De-identification Release Gating still applies in full — the option existing is not evidence it worked, and nothing in that paper measures the released bytes.

Why it matters for my work

Whole Slide Imaging carries the standing question of whether the group adopts DICOM or stays on vendor formats. This page is the practical half of that answer: DICOM adoption is affordable with a converter in the workflow and awkward without one. Choosing DICOM for the archive does not commit you to fighting every viewer, provided a conversion step is a normal part of getting slides to collaborators and teaching platforms.

It is also the same operation as DZI Web Publication seen from a different angle — that page converts a slide into web tiles for publication, this one converts between working formats. Both are “re-tile the pyramid into something the consumer can read”, and both are irreversible in practice once distributed.

How it connects

Whole Slide Imaging — the formats and the pyramid this operates on, and the page carrying the DICOM adoption question.

Colour Calibration — the ICC profile is the third colour mechanism, and the QuPath gap means in-house viewing does not honour it.

De-identification — conversion is the cheapest moment to strip the label, macro and metadata, because the file is being rewritten anyway.

De-identification Release Gating — and the reason that is not sufficient: an anonymise option is a claim, not a verified outcome.

DZI Web Publication — the publication-facing sibling; same re-tiling operation, different consumer.

Slide Metadata Extraction — conversion is exactly when metadata is at risk, so extract before converting rather than after.

QuPath Annotation Workflow — the tool most affected by the ICC finding here.

Chauveau 2025 — Converting WSI from DICOM to SVS-like TIFF — the source for every specific claim on this page: the viewer survey, the lossless-copy mechanics, the sparse-tiling penalty, and the QuPath ICC footnote.

Open questions

• Does the group’s archive standardise on any format, and is DICOM the intended destination? Still the open question on Whole Slide Imaging; this page prices it but does not answer it.
• Has anyone verified a conversion here at pixel or hash level rather than by opening the result? No conversion tool this wiki records reports one. [unverified]
• Does the ICC gap in QuPath affect any judgement currently being made from QuPath screenshots — teaching images, published figures, or colour-based scoring? Worth one check before it matters. [unverified]
• What proportion of the group’s slides are sparsely tiled? It decides whether the conversion size penalty is negligible or prohibitive, and it is a one-command answer on the existing archive.