Digital pathology AI adoption is slow because most labs still read glass slides. AI tools such as Paige Prostate have been FDA-authorised since 2021, but an algorithm can only read slides that have been scanned. Scanners, storage, lab-system integration and workflow redesign cost far more than the software, and industry estimates put full digital adoption near 10%.
Walk into most hospital pathology departments and you will see a scene that has barely changed in fifty years. Trays of glass slides move from the histology lab to a pathologist’s desk. The pathologist sits at a microscope, reads, dictates, and the trays move on to an archive room that is always running out of space. Somewhere in a vendor’s demo, meanwhile, an algorithm is outlining prostate cancer on a whole-slide image in seconds.
Both scenes are real, and the distance between them explains almost everything about AI in pathology. The algorithms are, for several narrow tasks, ready. The labs are not. And the thing standing between them is not model accuracy, regulatory approval or pathologist scepticism. It is the unglamorous business of turning every glass slide into a very large image file, every day, forever.
That leads to a conclusion vendors rarely put on a slide: AI will not pay for digital pathology. Digital pathology has to pay for itself, and AI arrives as a dividend afterwards.
How far has digital pathology adoption actually got?
Not very far. The FDA cleared the first whole-slide imaging system for primary diagnosis, Philips’ IntelliSite Pathology Solution, in April 2017. More than eight years later, a CAP TODAY industry roundtable in December 2024 put adoption at roughly 10%, with many organisations saying digitisation was “on their plan.” That estimate was a speaker’s judgment rather than a formal survey, but nobody in the field seriously argues the figure is much higher for fully digital clinical workflows.
Europe has more fully digital departments, often in public systems that can fund a regional build-out. Even there, a March 2025 study in Virchows Archiv of eight leading European laboratories found that high upfront costs for scanners, training and system integration remained a significant barrier, and that larger institutions were better placed to absorb them.
The AI, meanwhile, has been on the shelf for years. In September 2021 the FDA granted de novo authorisation to Paige Prostate, the first AI product authorised in digital pathology, designed to flag areas suspicious for cancer on scanned prostate needle biopsies. The validation work behind it, later published in Archives of Pathology & Laboratory Medicine, had 18 pathologists read 610 biopsy images drawn from 218 institutions, first alone and then with the AI. Their sensitivity and specificity both improved with assistance.
So the software is authorised, validated and useful. The slides mostly are not scanned.
Why is digital pathology so expensive to implement?
A radiology department went digital because the images were born digital: CT and MRI scanners produce files. Pathology is different. The primary object is still a piece of tissue on glass, and every digital image has to be created by an extra step that did not exist before.
That step carries several costs, and the scanner price is only the first.
Scanners and throughput. High-throughput scanners are expensive, and labs need enough of them to scan every slide the same day it is made, or the digital workflow becomes slower than the glass one. Vendor throughput figures can mislead. A 2025 study from Memorial Sloan Kettering in the Journal of Pathology Informatics compared scanners from seven vendors and noted that quoted speeds are often calculated for a theoretical 15 by 15 millimetre tissue area, not the thick, uneven, multi-level slides a real lab produces, and exclude the operator time around each scan.
Storage. Whole-slide images are enormous, often around a gigabyte each and sometimes more. A lab producing hundreds of thousands of slides a year is generating hundreds of terabytes, which must be stored for as long as regulations require retention and remain quickly retrievable.
Integration. The images must attach to the right case in the laboratory information system, open in the right viewer, and reach the pathologist in the right queue. The CAP TODAY roundtable described labs that bought scanners but did not digitise their workflow, precisely because the integration work was so heavy.
The histology lab itself. Scanning exposes every weakness in slide preparation: tissue near the edge, air bubbles, pen marks, cracked coverslips, inconsistent staining. A microscope forgives what a scanner rejects. Labs going digital typically find they must tighten processes upstream, add barcoding and build a quality-control step for rescans.
Memorial Sloan Kettering, one of the earliest large adopters, published a detailed breakdown of these infrastructure and operating costs in the Journal of Pathology Informatics in 2023, noting that cost is cited as a major barrier to adoption by many organisations. Its experience, built up over more than 15 years of incremental investment, is not something a typical community hospital lab can replicate in one budget cycle.
Does digital pathology pay for itself?
It can, over years, and mostly for reasons that have nothing to do with AI.
The Virchows Archiv study built a net present value model for European labs. It estimated cost recovery within about six years and a modestly positive seven-year net present value of around €0.21 million, driven by productivity and higher diagnostic volumes. The benefits pathologists reported were practical: easier access to archived slides, faster sharing with colleagues and multidisciplinary teams, remote reporting, better teaching.
That is a reasonable business case. It is also a slow, infrastructure-shaped one, closer to building a new wing than buying a software licence. And it tells you where AI fits. The AI is not the reason the numbers work. It is an option that becomes cheap to exercise once the lab has paid for everything else.
A worked example: why the algorithm cannot carry the scanner
Hypothetical, with illustrative numbers. Imagine a regional lab producing 250,000 slides a year. A vendor offers an AI tool for prostate biopsies. Prostate biopsies, say, account for 4% of the lab’s slides, about 10,000 a year.
The AI’s value case might be real: fewer missed cancers, faster sign-out, less need for second opinions on ambiguous cores. But to use it, the lab has two options.
Option one is to scan only the prostate biopsies. The lab buys a scanner, builds a parallel workflow for one specimen type, and asks pathologists to read some cases on screen and most on glass. The AI works. The lab now runs two processes, gets none of the wider productivity benefits of digital, and carries a scanner and storage bill justified by a single use case. Most finance committees will reject that.
Option two is to digitise everything. That means scanning 250,000 slides a year, storing perhaps a couple of hundred terabytes annually if images average around a gigabyte, integrating with the lab system, retraining staff and redesigning histology. The prostate AI is now a small add-on to a large programme. Its value is real but it is a rounding error in the business case, which has to stand on productivity, remote working and archive access.
Either way, the AI does not decide the investment. The digitisation decision does, and it is made on different grounds, by different people, on a longer timeline.
Wouldn’t cheaper storage and cloud AI change this?
The strongest counterargument is that the bottleneck is temporary. Storage keeps getting cheaper. Compression is improving. Cloud platforms let labs pay per slide rather than build data centres, and vendors increasingly bundle scanner, storage and AI into a subscription. Pathologist shortages in many countries add pressure that did not exist a decade ago. On this view, the tipping point is close, and AI will pull labs over it.
Some of that is right. Falling storage costs and subscription models lower the upfront barrier, and workforce shortages do change the calculation, especially for labs that struggle to recruit subspecialists and could use digital slides to share work across sites.
But the argument underrates the parts of the cost that do not fall with technology prices. Scanning adds a physical step to every specimen. Histology quality has to rise. Integration work is bespoke to each lab’s systems. Change management among pathologists who have used microscopes for decades is slow and human. None of these gets cheaper because a cloud provider cut its price per terabyte. The tipping point will come, but it will come lab by lab, driven by operational leaders, not by the arrival of a better algorithm.
What this means for lab leaders, AI vendors and health systems
For lab directors and hospital leaders, the practical lesson is to evaluate digital pathology as infrastructure first. Build the case on turnaround time, remote reporting, consultations and archive retrieval. Pilot scanners on real slides from your own lab, not vendor test sets, and measure total scan time including rescans. Only then ask which AI tools become affordable on top. Our guide to the questions hospitals ask too late when buying digital health applies with extra force when the purchase is a decade-long platform.
For AI vendors, the market is not every pathology lab. It is every digitised pathology lab, which is a much smaller number. Vendors who help customers digitise, through partnerships with scanner makers, financing or integration services, will grow faster than those who wait for labs to arrive. The pattern is familiar from other health technology markets, where startups die in procurement, not in the pilot.
For health systems and ministries, regional scale matters. The European evidence suggests larger institutions capture the economics more easily. Shared scanning hubs, shared storage and networked reporting across several hospitals may be the only realistic route for smaller labs, and that is a planning decision, not a lab-level one.
For pathologists, the change is a working-life change, not a software update. Training time, ergonomic screens and a say in workflow design determine whether digital reading feels like progress or like a slower microscope.
Frequently asked questions
What is digital pathology?
Digital pathology is the practice of scanning glass microscope slides into high-resolution whole-slide images that pathologists review on screen. It enables remote reporting, easier sharing with colleagues, digital archives and the use of AI tools that analyse the images. The FDA cleared the first whole-slide imaging system for primary diagnosis in the US in April 2017.
Which AI pathology tools are FDA-authorised?
The first was Paige Prostate, granted de novo authorisation in September 2021 to help pathologists detect areas suspicious for cancer on scanned prostate needle biopsies. It is an assistive tool; the pathologist still makes the diagnosis. Other AI pathology products have followed through FDA pathways, and more are available in Europe under CE marking.
How much storage does digital pathology need?
A whole-slide image is often around a gigabyte, sometimes more, depending on tissue size, magnification and compression. A lab scanning hundreds of thousands of slides a year can generate hundreds of terabytes annually, and images may need to be retained for years. Storage design, including tiering older images to cheaper storage, is a major part of the cost.
Is digital pathology as accurate as a microscope?
Validation studies submitted to regulators, including those behind the FDA clearance of whole-slide imaging for primary diagnosis, found diagnostic performance on screen comparable to glass for most surgical pathology. Some tasks, such as certain small organisms or fine nuclear detail, can still be harder on screen, and many labs keep a microscope available for these cases.
How long does it take for digital pathology to pay off?
It depends on lab size and how fully the workflow is digitised. A 2025 study of eight European laboratories in Virchows Archiv estimated cost recovery within about six years, with benefits driven by productivity and higher diagnostic volume rather than AI. Smaller labs often find the economics harder without regional collaboration.
Pathology AI is not waiting for a better model. It is waiting for the scanner, the storage bill and the integration project, and for someone in the lab with the budget and patience to own them.
Related on the GHC blog: Why Universal Health Coverage Stalls After the Pilot Phase and What Health Ministries Get Wrong About Airline Safety Culture.
Is your lab digital yet, and if not, what would actually tip the decision: cost, staffing or something else?
