How to evaluate digital pathology AI
Most pathology AI works on whole-slide images, so it depends on scanners, storage and a viewer the lab may not have yet. The digital setup is often the larger cost, and it applies to every tool below.
Then use the section for what the tool does. The questions in the core guide apply to all of them.
Last reviewed November 2026 · Independent and vendor-neutral
What does the tool you’re evaluating do?
Jump to the matching checklist. Some answers lead to a different guide.
Pathologists usually ask
- Does it work with our scanner and viewer?
- Is it a second look or a primary read?
- Does it catch small foci without slowing sign-out?
Lab directors usually ask
- Total cost of going digital: scanners, storage, viewer
- Accreditation and validation requirements
- Turnaround time and subspecialty coverage
The digital foundation (applies to every tool)
Applies if Use this section for any tool that analyzes whole-slide images.
Questions to ask
- Which scanners, file formats and viewers are covered by the clearance or validation? Is it tied to one scanner?
- What does storage cost per year at your slide volume, and who pays for it?
- Have you validated whole-slide imaging for diagnosis in your lab? The College of American Pathologists guideline calls for at least 60 cases per application and a washout period between glass and digital reads.
Red flags
- Scanner-agnostic claims without data
- No storage cost estimate
Cancer detection and grading
Applies if the tool finds or grades cancer on slides, for example prostate biopsies or lymph nodes.
Questions to ask
- Is it authorized as an adjunct (a second look after the pathologist) or for primary diagnosis? Use it only as authorized.
- How much did it improve pathologist detection in a reader study, and on which tissue types and stains?
- How does it handle small foci, artifacts and unusual morphology?
- What is the effect on sign-out time in a prospective setting?
Test it locally
- Run it on a set of your own signed-out cases, including difficult ones, and compare with final diagnoses.
Red flags
- Use for primary diagnosis beyond the cleared indication
- Results only from the developer's own lab
Biomarker quantification
Applies if the tool scores stains or counts cells, for example HER2, PD-L1 or Ki-67.
Questions to ask
- How does agreement with expert pathologists compare with agreement between pathologists, especially near treatment cut-offs?
- Which antibody clones, stain platforms and protocols was it validated on?
- Is the score used to choose therapy? If so, what regulatory status does it have for that use?
Red flags
- No data near clinical cut-offs
- Validated on a different stain platform than yours
Case prioritization and quality control
Applies if the tool sorts the worklist or flags slide problems (blur, folds, missing tissue) rather than making a diagnosis.
Questions to ask
- How often does it place a cancer case low on the list?
- Does it reduce rescans and turnaround time in practice?
Measure
- Turnaround time for urgent cases and rescan rate before and after go-live
What published studies show
The first AI pathology device authorized by FDA (De Novo, September 2021) is an adjunct for prostate biopsies, and its authorization was tied to a specific scanner and viewer. In the pivotal reader study, pathologists using it improved slide-level cancer detection by 7.3% on average. Check that your equipment matches what was authorized.
FDA De Novo decision summary DEN200080Further reading
Browse pathology tools
Every listing shows regulatory status, published evidence and named health-system users where public.