AI in Digital Pathology: Whole Slide Imaging, Cleared Tools and Costs

AI in digital pathology is one of the most talked-about areas of clinical AI, and one of the most misunderstood. The models are impressive, and a handful of tools are authorised for clinical use. But pathology AI depends on something most labs, in India and the US, still have not done: digitising glass slides at scale.

This guide explains how whole slide imaging works, which AI tools are actually authorised, what the new pathology foundation models change, and what a hospital lab in India should do first.

The short answer: digitise a focused workflow first, prove the scanning and storage, and add AI where it saves pathologist time on high-volume, well-validated tasks such as prostate biopsies.

What digital pathology and whole slide imaging are

In a traditional lab, a pathologist reads a stained glass slide under a microscope. In digital pathology, a whole slide imaging (WSI) scanner photographs the entire slide at high magnification and stitches it into one very large image, often several gigabytes per slide. The pathologist then reads it on a screen.

Once slides are digital, three things become possible:

• Remote reading. Pathologists can report from another site, which matters for hospital groups and for districts without a resident pathologist.
• Sharing and second opinions. Slides can be sent for review without couriering glass.
• AI analysis. Algorithms can scan every square millimetre for suspicious regions, count cells or grade tumours.

Which pathology AI tools are authorised

The list is shorter than many expect. An April 2026 analysis by regulatory consultancy Innolitics counted 51 AI-flagged FDA authorisations in pathology-related panels, but only seven that analyse whole slide images, and four of those are older image-analysis systems for HER2 scoring.

Since then, more has happened. Artera received clearance for a breast cancer risk-stratification tool in May 2026, PathAI's AISight Dx platform is cleared for primary diagnosis, and in February 2026 Labcorp announced it would deploy AISight Dx across its anatomic pathology labs.

The pattern: prostate is the most mature use case, and many newer tools are prognostic (predicting outcomes or treatment benefit) rather than diagnostic.

What the evidence looks like

The clearest early evidence came from Paige's FDA submission: 16 pathologists read 527 prostate biopsy slides, and their sensitivity for cancer rose from 89.5% to 96.8% when assisted by the AI. That is the typical value of pathology AI: it helps pathologists miss less on tedious, high-volume reads.

Two things to keep in mind:

• Authorised tools are tied to specific scanners and staining conditions. Change the scanner, and the evidence may not carry over.
• The Innolitics analysis notes that most US labs still use glass for primary diagnosis. The bottleneck is digitisation, not algorithms.

Pathology foundation models: what they change

The biggest technical shift is foundation models trained on millions of slides. Paige built Virchow, described as the first million-slide foundation model for cancer. In August 2025, Tempus acquired Paige for about $81 million, gaining a dataset of almost 7 million digitised slides, and launched Paige Predict, which estimates biomarker presence from standard H&E slides.

Why this matters:

• One model, many tasks. Instead of building a new algorithm per cancer type, developers fine-tune a foundation model with far fewer labelled examples.
• Biomarkers from routine slides. Predicting molecular features from H&E could, in time, help decide who needs expensive molecular testing.
• Lower cost to build new tools. Useful for academic centres and startups, including in India.

What has not happened yet: as of late 2026 we found no FDA clearance for a pathology foundation model as a general diagnostic. These models power research and specific products; each clinical claim still needs its own validation.

The Indian context

Indian hospitals are starting to digitise pathology, mostly at large centres. KIMS in Bhubaneswar commissioned a high-throughput scanner in November 2025, AIG Hospitals in Hyderabad installed one in 2023, and Bengaluru-based Morphle makes whole slide scanners locally. Most district hospitals and smaller labs remain fully glass-based.

The case for digital pathology in India is strong for specific reasons:

• Pathologist shortage outside cities. Remote reading lets a hub pathologist report for spoke hospitals.
• Cancer volume. Oncology centres handle large biopsy workloads where AI assistance helps.
• Hospital groups. Multi-site groups can centralise subspecialty reporting.

Regulation applies too. Under CDSCO's guidance on medical device software, AI that analyses slides to support diagnosis is device software, and IVD software has its own licensing route. Our post on AI medical device regulation under FDA and CDSCO explains the classification.

Costs and infrastructure

Digital pathology is infrastructure-heavy. Indicative cost drivers:

Storage surprises most buyers. A lab scanning a few hundred slides a day generates terabytes a month. Decide early what you keep online, what you archive, and for how long, in line with your record retention policy.

Integration with your LIS and HIS

Digital slides only help if they connect to the case. That means:

• Barcoded slides linked to the specimen and patient ID in your laboratory information system.
• The viewer opening the right case from the LIS worklist.
• AI results (flags, scores, heatmaps) returned to the case and visible at reporting.
• The signed report flowing back to the HIS and EMR, and on to ABDM-linked records where applicable.

Our guide to laboratory information systems for hospitals covers barcoding and analyser integration, and HL7 FHIR integration covers how results move between systems.

How to start: a phased plan

1. Pick one workflow. Remote reading for one spoke hospital, or prostate biopsies at your main lab.
2. Validate digital against glass. Have pathologists read the same cases both ways and compare agreement before relying on digital reads.
3. Build the plumbing. Barcodes, LIS integration, storage tiers, user access controls and audit logs.
4. Add AI on one task. Choose an authorised or well-validated tool for that workflow, run it in shadow mode, measure time per case and discordance.
5. Expand by evidence. Add sites or tissue types only when the first workflow is stable.

Where AI helps pathologists most today

Pathology AI earns its keep on tasks that are high-volume, tedious and well defined. In practice that means four kinds of work.

• Screening and second reads. Flagging suspicious regions on prostate biopsies so the pathologist looks there first, or acting as a safety net after the pathologist's own read.
• Counting and quantifying. Mitotic counts, Ki-67 and HER2 scoring, tumour percentage for molecular testing. Manual counting is slow and varies between readers.
• Quality control. Detecting blur, folds, pen marks and poor staining before a slide reaches the pathologist. Leica received FDA clearance for an AI quality-control tool for digital pathology in 2026.
• Case triage. Ordering the worklist so likely-malignant cases are read first.

What AI does not do well yet: rare tumours, unusual stains, frozen sections and the integrated judgement that combines morphology with clinical history. Those stay firmly with the pathologist.

Questions to ask a digital pathology vendor

1. Which scanners and file formats does your viewer support, and can we export slides in an open format if we change vendors?
2. Which AI modules are authorised, in which countries, and on which scanner and stain combinations?
3. Has the AI been validated on Indian patient samples and on slides prepared in Indian labs?
4. How does the system integrate with our LIS: barcodes, worklists, result return?
5. What are the storage requirements per slide at our magnification, and what compression is used?
6. How are model updates handled, and will we be notified before an update changes outputs?
7. What audit logs exist for who viewed, annotated and signed each case?

Proprietary slide formats are a real lock-in risk. Insist on export paths before you sign.

Mistakes to avoid

• Treating digital pathology as a scanner purchase rather than a workflow and data project.
• Assuming an AI tool validated on one scanner works on another.
• Ignoring storage growth until disks fill up.
• Locking slides into a proprietary format with no export path.
• Forgetting pathologist training on screen-based reading, which takes time to feel natural.
• Skipping the digital-versus-glass validation step.
• Running AI outputs in a separate system that pathologists do not open during sign-out.

Conclusion

AI in digital pathology is real, but narrower than the headlines. A small number of tools are authorised, prostate biopsy is the most proven use, and foundation models are changing how quickly new tools can be built. For most Indian hospitals the first job is not AI at all: it is digitising a focused workflow, connecting it to the LIS and proving that remote reading works.

At Algon we build the integration and health data layers around lab systems; the cleared AI models come from specialist vendors. Get the foundation right and adding validated AI later becomes a manageable step.