A lab digitizing glass pathology slides for research needs to decide six things: how to scan, how to remove identifiers from the slide label, how to prove provenance, how to check image quality, whether to compute AI-ready representations, and how to catalog the result. Scanning produces an image. The other five steps decide whether a researcher can find, trust, and use it.
What is a whole-slide image?
A whole-slide image (WSI) is a high-resolution digital scan of an entire glass slide. A single image is often gigapixel scale, stored as a pyramid of tiles at several magnifications so a viewer can zoom without loading the full file.
Scanners write proprietary formats as well as open ones. The DICOM standard includes a whole-slide imaging object, which gives labs a vendor-neutral option for storage and exchange. The DICOM standard defines it alongside other medical imaging types. Recording the scanner, magnification, and file format for each image makes later processing predictable.
Which slides should be scanned first?
Scan the slides that answer known research demand. Archives are large, and scanning time is finite. A useful first batch has a clear diagnosis, available associated clinical information, and, where possible, a matching FFPE block. FFPE (formalin-fixed, paraffin-embedded) tissue is the standard way pathology specimens are preserved and stored.
A slide with linked context is worth more to a researcher than an isolated image. Our guide to linking FFPE blocks, slides, and clinical data covers how those records are matched to the same patient.
How should slide labels be handled?
Slide labels often carry an accession number, a barcode, or a patient name. Many scanners capture the label as part of the image or as a separate label image. Before an image is shared for research, the label must be redacted or excluded so it no longer shows identifying information.
Redaction is one part of de-identification, the process of removing information that could identify a patient. The HHS guidance on de-identification under the HIPAA Privacy Rule describes the two recognized methods: Safe Harbor and Expert Determination. File metadata needs the same review as the visible label, because identifiers can sit in image headers.
How can a lab prove where an image came from?
A provenance watermark is an invisible mark embedded in an image that identifies its source. It does not change what a pathologist sees. It lets the originating institution show that a copy came from its archive, which matters when images are licensed for research or model training.
Provenance also covers the record around the image: which slide it came from, when it was scanned, and what processing it has undergone. Keep that history with the image rather than in a separate spreadsheet.
What quality checks should run on every scan?
Automated quality checks should flag images that would mislead an analysis. Three checks are common:
- Tissue detection marks the regions of the image that contain tissue, so blank glass is excluded from analysis.
- Blur detection marks regions that are out of focus.
- Fold detection marks regions where the tissue section folded over itself during preparation.
These checks produce masks, which are image overlays that label each region as usable or not. Masks let a researcher filter images before analysis, and they tell the lab which slides may need rescanning. Record the quality status with each image, so a researcher can see it before requesting the file.
What are foundation-model embeddings, and why compute them?
A pathology foundation model is an AI model trained on large numbers of slide images to recognize general tissue patterns. An embedding is the numeric summary such a model produces for a small tile of the image.
Computing embeddings once, at digitization, makes the archive searchable by visual content and ready for model development. A researcher can look for slides with similar morphology without the lab sharing every raw image. Embeddings are a research aid, not a diagnosis.
What makes a digitized archive discoverable?
A catalog makes images findable. Each catalog entry should describe the image at a level a researcher can search: tissue site, diagnosis, stain, scan year, quality status, and whether linked clinical data or an FFPE block exists.
Each entry describes one slide, so it carries slide-level details but no identifiers: no names, record numbers, or accession numbers. The institution decides what is discoverable, to whom, and under what terms. Researchers can then ask whether a cohort exists before anyone requests a file. Our article on what a feasibility count is explains that step.
How Biobase approaches digitization
Biobase runs digitization as one sequence: scan, redact, watermark, quality-check, embed, and catalog. The output is a set of slides that are scanned, quality-checked, and ready to use. The images remain under the institution’s control, and the institution approves what is shared.
For the full path from archive to research use, see how a clinical archive becomes a research-ready cohort. To discuss a slide archive, tell us what your institution holds.