For more than a century, pathologists have relied on a microscope and a glass slide to diagnose cancer. At the Kimmel Cancer Center, this powerful tool has guided care for generations of patients. Now, however, with the help of digitization and artificial intelligence (AI), the familiar glass slide is being transformed into a digital treasure trove, one that promises faster, more precise diagnoses, better training for future pathologists and new opportunities for discovery.
“Every year, we generate close to a million slides at Hopkins,” says Angelo De Marzo, M.D., Ph.D., associate director of cancer research pathology. “Each one contains gigabytes of information. Together, they represent an immense archive of knowledge, but they are stored in drawers and warehouses, accessible only by pulling the glass and putting it under a microscope.”
A Virtual Microscope
A robust effort to scan these slides and digitize them so they can be easily accessed and viewed on a computer screen opens up a world of possibility.
The idea of scanning pathology slides into a “virtual microscope” has been envisioned for decades, but now, the technology has caught up with our imagination.
At the Kimmel Cancer Center, Alex Baras, M.D., Ph.D., director of digital pathology and precision medicine informatics, is helping bring it to fruition, developing plans to ramp up from scanning tens of thousands of slides to hundreds of thousands of slides each year, and eventually scaling up to 1 million annually in five years. Together with Lisa Rooper, M.D., director of the Division of Surgical Pathology, and other colleagues across surgical pathology, Dr. Baras is building momentum to digitize the enormous Johns Hopkins pathology archive.
Although the task is an onerous one, the benefits are immediate and, ultimately, immensely more efficient. Pathologists would no longer need to request slides from an off-site warehouse. Instead, these digitized slides could be pulled up on a computer instantly, and experts could confer with one another from as close as the office next door to the far reaches of the globe. Slides from other hospitals, which often must be returned after a short time, could now be preserved indefinitely. Samples stored at Johns Hopkins dating back to the 1918 influenza pandemic could now be digitized and studied, Dr. De Marzo says.
For patients with cancer, that means greater efficiency and better care. If a cancer recurs years after an initial diagnosis, doctors can compare the new biopsy with the original digital record in seconds, a process that often required shipping fragile slides across the country. For research, this provides a gold mine of samples that can be analyzed and incorporated into research data to enhance cancer discovery.
AI at the Microscope’s Side
Scanning slides is the key first step to employing AI in cancer pathology, but it is only the beginning. Once digitized, pathologists can begin to use AI tools to assist in cancer detection, precision diagnosis, and treatment guidance. In breast cancer, for example, Dr. De Marzo says AI can help quantify key biomarkers that reveal what is fueling the cancer, such as estrogen and HER2 receptor levels, more consistently and quickly than humans alone, helping ensure patients quickly receive the targeted therapies that will combat the cancer.
When reviewing slides for the presence of prostate cancer, Dr. De Marzo says he often examines 50 slides for a single patient. In one day, he may review these for 10 different patients. It requires a highly trained eye to find cancer, precisely grade the cancer and — just as importantly — correctly assess when no cancer is present.
Experts like Dr. De Marzo are in short supply and high demand, often sought out by hospitals around the world to review cancer pathology slides. His colleague and prostate cancer expert Tamara Lotan, M.D., the Rose-Lee and Keith Reinhard Professor of Urologic Pathology, is leading efforts employing novel AI systems that can screen trays of biopsy slides and highlight areas most likely to contain cancer, so the pathologist can immediately zero in on concerning areas. Dr. Lotan’s team has found that these algorithms can also accurately grade the cancers and predict outcomes.
“Instead of spending 20 minutes per tray, AI could cut that to just a few minutes,” says
Dr. De Marzo. “It frees us to take on more cases and to collaborate and spend more time on the most challenging cases.”
He is quick to point out that AI tools will be used to augment and amplify, not replace, the pathologist’s judgment. He says studies at Johns Hopkins and elsewhere found that AI can match the accuracy of top pathologists, while also lifting the quality of diagnoses across the board. Considering the significant shortage of highly qualified pathologists, AI can help expand the capabilities of top pathologists like Drs. De Marzo, Lotan and Rooper, and help ensure all patients have access to this level of review.
“Every year, we generate close to a million slides at Hopkins, Each one contains gigabytes of information. Together, they represent an immense archive of knowledge, but they are stored in drawers and warehouses, accessible only by pulling the glass and putting it under a microscope.”
Angelo De Marzo, M.D., Ph.D., associate director of cancer research pathology.
3D and Molecular Prediction
In addition, researchers are now taking pathology from the two-dimensional images on a slide into the realm of 3D. Johns Hopkins faculty members — including Laura Wood, M.D., Ph.D., Ralph Hruban, M.D., Denis Wirtz, Ph.D., Ashley Kieman, Ph.D., and others — are exploring how AI and new imaging technologies can reconstruct tissue in 3D, offering a more complete picture of diseases such as pancreatic cancer.
Meanwhile, the Cancer Center’s co-associate director of precision oncology, Srinivasan Yegnasubramanian, M.D., Ph.D., who is the director of inHealth Precision Medicine at Johns Hopkins, along with Dr. Baras and computational collaborators, are working rapidly to marry digital pathology with genomics and big data. The goal broadens the reach of pathology beyond diagnostics to forecast molecular changes, and predict treatment response — including a response to immunotherapy — all based on the subtle features contained in the standard H&E pathology slide.
“Imagine guiding therapy without having to order additional expensive tests. We see that as part of the promise of AI,” says Dr. De Marzo.
Training the Next Generation
These advances are changing pathology now, but it is the next generations of pathologists who will uncover the full promise of this new landscape.
Traditionally, residents learn and sharpen their diagnostic skills by spending hours at a multiheaded microscope under the mentorship of a senior faculty member. As a school of medicine faculty member who trains the next generation of pathologists, Dr. De Marzo, along with departmental leadership and several up-and-coming young pathology faculty members with a passion for teaching, are working to ensure that young pathologists learn to master the fundamentals, with AI serving as a tool, not a crutch.
“With the right design, AI could actually enhance training,” he says. “Imagine asking the system to show you 20 examples of a rare cancer, all verified by experts. That kind of resource could accelerate learning in ways we’ve never had before.”
Challenges and Opportunities
Still, the path ahead isn’t without obstacles. Storing millions of digital slides requires immense amounts of space. Implementing AI tools clinically will also demand FDA approvals, careful validation, and new safeguards for privacy. Training programs will need to adapt to ensure young pathologists still learn core skills. The investment needed to realize these goals, particularly in light of the current economic pressures on academic medicine, is challenging.
“In the short term, it costs more,” Dr. De Marzo acknowledges, “but over time, it will save money, improve efficiency, and make us better.”
Johns Hopkins leadership is committed to meeting the challenge, and has already begun addressing these issues.
“Johns Hopkins’ unique strength lies in its ecosystem. With collaborations across medicine, engineering, and computer science, and the launch of a $400 million AI Institute, the pieces are in place to connect digital pathology with genomics, large language models, and big data analysis,” says Dr. De Marzo.
Looking Ahead
Within five years, Dr. De Marzo predicts, most Johns Hopkins diagnostic slides will be scanned prospectively, and AI tools will be embedded into daily practice, improving speed, accuracy, and patient outcomes.
Dr. De Marzo says the promise of AI in pathology reflects a long-held Johns Hopkins tradition, in which mentors — such as Johns Hopkins cancer pioneer Dr. Donald Coffey and leaders such as Kimmel Cancer Center Director William Nelson and school of medicine Dean Ted DeWeese — foster a culture in which collegiality thrives and bold ideas take root.
“This place has always respected every discipline — mathematicians, engineers, biologists, clinicians, all of them,” he says. “That culture of respect is what lets us solve problems together. Now, AI is becoming part of that story.”