Slides to the Cloud, AI Reading Them: Digital Pathology in 2026

一、From impressive papers to partial use

AI-assisted pathology has been discussed for years. Earlier coverage focused on accuracy, competition rankings and publications. Over the last two years the centre of the conversation has clearly moved: can it be approved, can it be reimbursed, and will pathologists actually look at what it outputs.

That shift itself shows that the technology is no longer the main obstacle.

二、What is actually working in three layers

1. Already stable: scan, store, remote consult

These three are technically mature. Whole-slide scanning resolution and speed now support routine diagnosis, and remote consultation runs routinely on most provincial platforms. The bottleneck is cost and process, not technology.

2. Landing now: AI-assisted screening

Cervical cytology, tuberculosis and localisation support for certain tumours are the areas with the most solid uptake. Their positioning is explicit: help pathologists miss less, not replace them in issuing reports.

Doctors accept these tools for a practical reason. The repetitive work is taken over while the judgement stays with the person.

3. Early stage: multimodal and prognostic prediction

Combining slide images with molecular tests and clinical data is still at the research and small-scale validation stage. There is distance to routine use, but the direction is clear.

三、Four real barriers in practice

1. Data compliance

Whole-slide images are medical data. Cross-institution use, model training and cloud upload all fall under specific regulations. This shapes the system architecture, not just the feature list.

2. Compute and cost

Scanning, storage and compute together are a serious outlay for a county-level hospital. Most institutions scan and store first and add AI later.

3. Clinical trust

Retrospective validation is easy to make look good; prospective validation is what settles the question. Doctors care less about published accuracy than about the miss rate on this department's slides.

4. Reimbursement and liability

After an AI prompt, who signs and who is responsible still varies by region. Until that is settled, most departments will look at the output but not act on it.

四、What this means for procurement

1. Keep the interface open, skip the AI premium

When buying a scanner today, check whether it can connect to external AI platforms and export standard formats. For the analysis module bundled with the device, whether it gets used matters more than how long the feature list is.

2. Size storage once, add compute later

Expanding storage is far more painful than adding compute. Plan three to five years of capacity up front and avoid adding drives every year.

3. Make it usable before making it smart

If consultation cannot be sent out, teaching cannot be annotated and case numbers do not match, adding AI only adds another unused feature.

五、Closing

Digital pathology is not a single upgrade but a staged workflow change. The reality in 2026 is that scanning and remote consultation can be adopted with confidence, AI-assisted screening works in specific directions, and full-process intelligence is still on the way.

For most institutions the sensible move is to lay a solid base in scanning and storage, tidy the workflow, and connect AI when it matures.