Digital Pathology and AI: Now and in 2025

Digital and computational technologies saw a surge in 2024, with clinical labs expanding their digital workflows and adopting artificial intelligence (AI) tools for everything from diagnostic support to patient communication. How has the landscape of these technologies changed over the past year—and what’s the outlook for 2025? We spoke to computational pathology experts Nathan Buchbinder and Julianna Ianni to find out how digital pathology and AI are shaping the future of the lab.

Q: What have been the key trends in digital pathology and AI over the past year?

Nathan Buchbinder: I want to highlight two connected trends. The first is that digital pathology adoption is being driven by much more than just quality and efficiency. We’re starting to see real business drivers that create significant new value for labs, whether research or diagnostic. Digital pathology has become an integral part of business planning and strategy.

The second is the increasing relevance of precision medicine as a value proposition for digital pathology. In the life sciences, digital pathology is now deeply embedded in everything from biomarker discovery to clinical trials. On the diagnostic side, we’re seeing major technological advances in AI driving a surge in precision medicine applications. This allows clinical labs to think beyond just improving efficiency, productivity, and quality and consider how the information they provide can have an even greater impact on care decisions.

Julianna Ianni: From the AI perspective, labs are now realizing the value of foundation models and looking to adopt them—both because they see the power of these models for building applications and because they anticipate the ability to become more efficient in their AI development. In general, I think there’s a lot more recognition of the need for investment in AI. Labs are no longer just thinking about what applications to adopt, but also about how they can better build their AI functionality.

Q: What are the key developments driving the increased interest in AI?

NB: The most impactful one is the rise of powerful foundation models that can accelerate the time to develop new AI applications and open up a whole new world of problems that AI can solve. We’re starting to see that the technology has reached a point where most large enterprises have figured out their strategy and how it ties into their overall goals. In the diagnostic space, the question is no longer, “How do I adopt the best AI application for this use case?” It’s now, “How do I address as many use cases as possible and build a foundation to scale up in the future?”

We also saw the first public announcement of a digital biomarker—one that leverages digital pathology to identify patients who would benefit from a particular therapy.¹ That’s incredible because it takes digital pathology from a way to improve lab operations to an essential tool for delivering the next generation of medicine.

Q: What do you anticipate for digital pathology and AI in 2025?

JI: I think we’ll see more foundation models. I also think we’ll see more multimodal models—ones that can process language, image, and maybe even genomic or proteomic data simultaneously. We’re starting to understand the potential applications for both life sciences and diagnostics and I think these models are going to be really big.

NB: I have four discrete points:

1. Several things we saw in 2024 will be increasingly relevant in 2025. Digital biomarkers and digital companion diagnostics are one example that I expect will be really transformative.
2. The ability to iterate quickly on foundation models and bring that technology into routine workflows will make them an integral part of the way labs work.
3. I predict that 2025 will be the year a lot of labs achieve 100 percent digital routine workflows. A lot of organizations that are not yet fully digital will take a deep dive in the coming year.
4. We can expect a lot of surprises in terms of what digital and computational technologies can do. They’re developing so fast that I can hardly imagine what the landscape will look like a year from now.

Q: What digital and computational pathology challenges will labs face in the coming year?

JI: Because things are advancing so quickly, I think some organizations may struggle to maximize the benefits of these technologies. There will definitely be a learning curve regarding how to adopt and scale the use of AI technologies. Not only that, but as labs begin to build their own AI tools, they’ll need data to test and train them. That data—if it exists—may be in silos across the organization, which makes finding, organizing, and centralizing it a challenge.

NB: Early adopters who opted for on-site digital infrastructure will find that some of their goals are difficult to accomplish at scale on-site. We’re already seeing an increased push toward scalable cloud infrastructure and I think those who didn’t adopt it initially will likely need to make the transition as their digital workflows evolve from scanning a few slides a day to processing thousands of images and serving hundreds of users across multiple systems.

Q: What one action would you recommend labs take to prepare for digital pathology and AI in 2025?

NB: I encourage labs to stop thinking about digital pathology solely as a driver of productivity and quality and instead think of it as an opportunity to create significant new value in the shift toward precision medicine. Consider how technological advances are impacting the value digital pathology and AI can create and what that implies in terms of adoption strategies and timelines.

JI: Start building with foundation models if you haven’t already. I think that will allow labs to see both the opportunities and the challenges that they will face in response to the rise of this technology. You don’t need to create something that’s production-ready; just building something that might be helpful for your organization is a great learning experience in adopting and working with AI tools.

Reference:

1. Daiichi-Sankyo. Novel Computational Pathology-Based TROP2 Biomarker for Datopotamab Deruxtecan Was Predictive of Clinical Outcomes in Patients with Non-Small Cell Lung Cancer in TROPION-Lung01 Phase 3 Trial. September 8, 2024. https://www.daiichisankyo.com/files/news/pressrelease/pdf/202409/20240908_E.pdf.

Nathan Buchbinder is cofounder and chief strategy officer at Proscia. He holds a BS in biomedical engineering from Johns Hopkins University and an MS in biomedical innovation and development from the Georgia Institute of Technology and has driven digital pathology and AI adoption and advancement for over a decade.

Julianna Ianni is vice president of AI research and development at Proscia, where she has spent the past five years developing AI solutions for the lab. She holds a BS and a PhD in biomedical engineering, both from Vanderbilt University, and is responsible for AI strategy, objectives, and forecasting in her current role.