Digital pathology is changing the ways in which pathologists work, driving a shift away from traditional diagnostic workflows that rely on glass slides and microscopes. However, change can be challenging in a field with over 100 years of history. Despite the benefits of digitization, most pathology practices have remained analog due to significant cost and technological barriers.

This could soon change as the past two decades have seen a significant increase in the image quality and resolution, enabled by whole slide imaging (WSI) scanners, and accompanying decreases in the costs of obtaining, storing, and managing digital slide data. Together with rapid progress in artificial intelligence (AI), these advances could unlock new applications, make diagnostic workflows more efficient, and address a worldwide pathologist shortage that is compounded by rising workload volumes and complexity.

“We are reaching an inflection point where we are going to see much broader adoption of digital pathology over the next couple of years,” said Andrew P. Norgan, MD, PhD, chief medical officer at Mayo Clinic Digital Pathology and consultant at the department of laboratory medicine and pathology at Mayo Clinic.

The Mayo Clinic has been an early adopter of digital pathology. Over the past few years, the non-profit medical group has undertaken the task of scanning its extensive archive of pathology slides, as well as slides from current patients, with the help of automated robotic scanning, leveraging more than 20 million digital slide images to date.

“We made the decision to digitize our practice because we fundamentally believe this is the way to advance medicine broadly for complex diseases,” said Norgan. “Pathology slides are a rich source of information that has previously been unable to be accessed digitally and combined with all the other patient data we store.”

One of the major benefits that the implementation of digital pathology has brought to the Mayo Clinic is logistics and workflow improvements, especially when it comes to sharing information remotely. This allows pathologists to quickly access expertise across the hospital network. “It has really revolutionized what used to take days,” said Norgan. “No more sending slides back and forth. We now do that instantaneously, and patients are getting expert answers a lot faster than we were previously able to provide them.”

But the best may yet be to come. Rapid advances in AI technology are enabling new diagnostic capabilities that were previously unthinkable. Through partnerships with Google, Microsoft, and NVIDIA, the Mayo Clinic has undertaken the development of AI models that can accelerate medical discoveries and further improve the efficiency of pathology practices. In collaboration with Aignostics, the organization recently built an AI foundation model to analyze histopathology slides using data from 1.2 million digitized slides provided by the Mayo Clinic and Charité – Universitätsmedizin Berlin, Europe’s largest university hospital.

Trained on vast datasets, foundation models can serve as the building blocks for more specialized applications down the line. Norgan emphasized their potential to break the log jam that currently exists in digital pathology, allowing AI to increasingly bring value to the pathology practice and ultimately, to patients.

AI drives innovation

With AI breaking new ground in digital pathology, a number of companies have been developing algorithms that serve digital pathology applications. One of them is Owkin, a French-American enterprise that has been working with big pharma partners like Merck and AstraZeneca to develop AI-powered digital pathology tests.

“Digital pathology really comes to power when it is combined with AI and new developments in precision medicine,” said Katharina Von Loga, MD, PhD, head of pathology at Owkin. She sees huge potential in digital pathology applications that focus on the prediction of patient outcomes, where AI can go far beyond the human eye.

For instance, Owkin has developed and validated an AI diagnostic for breast cancer patients that can assess an individual’s risk of relapse within five years by analyzing WSI and clinical data, helping doctors choose the best course of action. Another AI model pre-screens digitized slides of colorectal cancer samples for microsatellite instability, a genomic biomarker that is predictive of immunotherapy response in solid tumors.

Von Loga expects AI technology to soon enable breakthroughs in patient selection for next-generation precision treatments. Increasingly, detecting whether a biomarker is present or absent in a sample is no longer enough to make optimal treatment decisions. That is the case, for instance, of immune checkpoint inhibitor therapies targeting the programmed death ligand (PD-L1), for which it has been established that not all patients with PD-L1 tumor expression will respond to the treatment.

A much deeper level of biomarker analysis will be required going forward, including but not limited to expression levels, heterogeneity, and spatial distribution across a tissue sample. Here is where AI can step in to make better outcome predictions that bring together all information available, from histopathology to molecular diagnostic techniques and electronic patient records.

Furthermore, the capacity of AI to sort through large amounts of data will prove invaluable as the amount of tests and drugs available to patients continue to increase, pushing physicians to keep up with constant changes in the standard of care. “We are getting much more detailed in our reporting for all the different biomarkers and drugs available,” said Von Loga. “That would be impossible to do in a timely fashion with a standard workflow for a patient who may have multiple drugs available for their condition.”

Quest Diagnostics, a provider of diagnostic information services, has been scaling its efforts in the digital pathology space through an ongoing collaboration with PathAI. “Beyond assisting the pathologist in looking at and identifying problem areas on the slide, there are a lot of AI tools that can help the pathologist become more efficient and work more quickly,” said Chad Richards, president of pathology and medical services at Quest Diagnostics.

In this context, AI tools can assist with routine tasks that take valuable time away from pathologists. For instance, one of the AI tools developed by PathAI can automatically detect issues with a slide image and can flag the sample to be re-scanned or re-cut before a pathologist looks at it. Another tool for oncologists can identify which cases are most likely to be malignant, prioritizing them to ensure the pathologist looks at them first and orders any additional tests within the same day.

In addition to productivity gains, AI can help pathologists be more objective in their analysis, especially when it comes to borderline cases where a certain biomarker is near the threshold required to benefit from a given therapy. “If you showed a borderline case to ten pathologists, you will probably get half of them on one side and half of them on the other,” said Richards. “AI can do that much more accurately and much faster.”

Challenges to implementation

Across the board, the number one challenge facing the adoption of digital pathology remains the costs of digitization. Slide scanners are expensive, and the cloud storage, image management systems, and AI tools all carry additional costs. To make matters worse, there are currently no direct reimbursement programs in place for the digitization of a pathology practice, meaning that the return on investment has to be realized through improvements in efficiency and diagnostic accuracy.

There is still room for improvement in scanning technology, which cannot yet cover every use case within a pathology practice. The available technology may also present some interoperability issues, meaning customization efforts are often required to successfully implement digital pathology, driving costs further.

For tool developers, the challenges go beyond purely technological development. Any digital pathology tools need to be seamlessly integrated in an organization’s workflows and be as simple and straightforward as possible for users. “Every hospital has their own equipment and protocols, and we need to make sure that these systems work equally well in all of them,” said Von Loga.

Even within the same organization, different medical specialties can have very different needs. A dermatologist and an oncologist, for instance, may work very differently based on the number of cases they take on every day and the speed with which they need to establish a diagnosis. “Being sensitive to those differences and developing tools that are helpful rather than a hindrance is key to success in digital pathology,” said Richards. “If the tool does not integrate well in the daily workflow of the pathologist, it will only slow them down.”

Navigating regulations also remains an obstacle. “The biggest hurdle for tool makers is working with regulators,” said Richards. He explained that one of the challenges of working with AI is that the algorithms continue to change and improve as they look at more and more cases. This can clash with current FDA requirements for medical devices, which have to go through the whole approval process every time anything is changed. “That is something we still have to work out with the regulatory bodies.”

When it comes to implementing new tools and workflows within an organization, change management has to be considered throughout the whole process to ensure success. “Change is hard for everyone, and it takes time to get comfortable with a new way of doing things after years of doing them a certain way,” said Richards.

On the bright side, giving pathologists enough time to adapt to new tools can allow them to fully come on board with digital pathology. “Some of the pathologists who used to be diehard microscope people are now the most digitally oriented, because they have seen the intangible benefits of digitization,” said Norgan.

Ultimately, these challenges are not that different from what other fields, like radiology, have experienced through digitization in recent years. James Rogers, chief executive officer of digital pathology and senior administrator for generative AI at Mayo Clinic, is confident that all these challenges can be overcome, and that as technological costs decrease and the value generated by AI integration continues to increase, more and more pathology practices will start investing in digitization.

Preparing for a digital future

Experts have already noticed a shift in the digital pathology market, signaling the growing maturity of the field. “We see some of the bigger players, such as Roche Diagnostics and Leica Biosystems, starting to really ramp up their efforts in digital pathology, whereas three or four years ago the space was dominated by small startup companies,” said Richards.

With more players entering the market, there will be more choices available to cover a wider variety of use cases and address the unique needs of each organization. In the coming years, the number of approvals for drugs with AI companion diagnostics will grow, said Von Loga, and those who do not undergo digitization will not be able to diagnose and administer these treatments without relying on external services.

As organizations compete for an ever-scarcer talent pool, Rogers believes that digital pathology capabilities will also become a valuable recruitment tool. Digitization can be particularly attractive to prospective candidates because it offers pathologists more flexibility, allowing them to work remotely without being tied to the physical location of their laboratory.

Finally, “Digital pathology is opening up an avenue of research that has not been available before,” said Rogers. He sees immense potential in layering information from digital pathology together with omics data and electronic health records to build a detailed picture of each patient’s case, enabling breakthroughs on both the diagnostic and therapeutic fronts.

“We can now use AI in a way that is unprecedented,” said Rogers. “As more and more of these larger data sets become available, we will find solutions we had no idea had been sitting in front of our face. Now all that data is being unlocked, the best minds in the world can be put to work against today’s problems. This is a once in a lifetime opportunity to truly transform medicine.”

Clara Rodríguez Fernández is a science journalist specializing in biotechnology, medicine, deeptech, and startup innovation. She previously worked as a reporter at Sifted and editor at Labiotech, and she holds an MRes degree in bioengineering from Imperial College London.