This article is part of a VB special issue. Read the full series: AI and the future of health care
Traditionally, when patients undergo a biopsy, their sample tissues or cells have been sent to a pathology lab, put on blocks, cut into sections, stained with tint, then analyzed under a microscope.
Even in this era of digitization, that process remains relatively unchanged.
“Pathology still lives on glass,” said Brian Rubin, chair of the Pathology and Laboratory Medicine Institute at Cleveland Clinic.
“One of the really surprising things about pathology today is, it is pathologists with microscopes and glass slides,” said Dr. Andy Beck, CEO of AI platform developer PathAI. “People are amazed it isn’t digital.”
But PathAI and Cleveland Clinic aim to change that. The northeastern Ohio medical system and the Boston-based developer of AI and deep learning medical pathology tools have embarked on a five-year research collaboration. This partnership will involve digitizing pathology specimens and linking clinical data with digitized pathology data. PathAI will also develop algorithms based on analysis of workflows and pathology use in the lab.
Using AI for a cure
The end-goal is to help Cleveland Clinic leverage AI to more quickly identify disease and match patients to the best therapies unique to their conditions, Beck said.
“There is tremendous potential for creating new algorithms in AI-powered pathology that are going to be helpful for research and patient care,” he said.
PathAI’s tools use deep convolutional neural networks (CNNs), which consist of millions of layered, densely interconnected processing nodes. Using parameters in the tens of millions, these complicated networks can identify patterns in images and video.
These tools will be applied at Cleveland Clinic as it executes a plan to scan 1.5 million slides over five years. The medical system has one of the largest glass databases in the world – what Rubin described as a “fantastic, well-curated library of well-annotated, well-diagnosed slides.” As he explained, the operator of 10 medical facilities in northeastern Ohio has been using digital slide technology on a limited basis for years.
As part of the endeavor, the clinic will purchase several new scanners and add roughly a dozen specialists – including analysts to develop methods to transfer de-identified data – to its team of 100 sub-specialized pathologists.
“The partnership is really to launch development at scale that we couldn’t do on our own,” Rubin said.
“Our commitment is to provide the best possible care for our patients and it is increasingly clear that AI-powered pathology can radically enhance diagnostic accuracy and treatment selection,” he added. “By doing this work, we’re able to maximize the value of machine learning for our patients and fuel deeper innovation that can result in better outcomes.”
Providing insights at deeper molecular levels can lead to more accurate diagnoses, Beck agreed.
“In a research setting, it can predict things that you could not predict at all by eye,” he said. “At the end of the day, deep learning-powered pathology can really broadly impact research as well as clinical care.”
Enhanced diagnosis at scale
It sounds like a win-win, Beck and Rubin agreed. Still, widespread digitization has been limited. The biggest hurdle, Rubin said: Added debt. Pathology is a low-margin business and creating digital slides requires scanners averaging in the $500,000 range, as well as upfront software and people costs. The benefits – the perceived ones, at least – haven’t yet outweighed those costs.
“Whether you look at something under a microscope or on a desktop, unless there’s an added benefit and it’s cost neutral, most people are not going to try to embark on that at scale,” Rubin said. “They have to have an incentive.”
Cleveland Clinic sees a major one in the portability of data and the ability to share images. There is also immense opportunity in terms of storage – physical slides are large and the intensive physical space they require could be reduced with digital slides.
And of course, there are the broad-reaching implications, Rubin said: more accurate predictions, better patient outcomes, increased insight into the molecular underpinnings of disease, insights for research and education.
“Pathology is a very subjective practice,” Rubin said. Although highly trained, pathologists can make mistakes or miss things – which is why patients are always told to get a second opinion. “It’s nice to be able to add just layers of safety, bring additional layers of quality,” he said.
Over just six years, PathAI has excelled in its niche: The company closed a $165 million series C round in 2021 and also recently acquired Poplar Healthcare Management, one of the country’s largest pathology labs.
The company homes in on the drug development cycle, aiming to provide new insights into how pathology relates to drug response, Beck said. PathAI is deploying its systems in clinical trials with various healthcare providers and it’s also building new types of tests that are based wholly on digital images.
The Cleveland Clinic partnership promises many new and different opportunities.
For example, Rubin pointed out, pathology requires robust methods of cataloging, annotation and retrieval. Pathologists pull slides out of physical archives all the time – when a patient has a recurrence of cancer, for instance, new and old slides are compared. Samples must also be securely stored for varying sorts of time based on different regulations.
As Rubin posited: What are the required steps to realize all this in the digital realm?
The partnership aims to answer such questions, he said, while ultimately establishing algorithms and models for diagnoses and workflow to take pathology to the next level.
“It’s all really new territory,” Rubin said. “It’s exciting for us, being academic pathologists, to now be working with AI/ML developers.”
Beck estimated that only about 5% of labs have gone fully digital. But he said he’s confident that the field will be transformed over the next five to 10 years, “Labs will look very different.”
Getting there, though, will only happen through partnerships, which requires a significant time and monetary commitment for providers and tech companies. To break tradition takes a good reason and a motivator and medical institutions need help recognizing the value in increased quality and reproducibility, Beck said. Those in the tech industry also must help healthcare stand up digitization efforts and make them less cumbersome.
“With just digitization alone, the convenience hasn’t been enough to switch from glass slides and microscopes,” Beck said. “But once you layer on AI, the value that these digital images can bring increases tremendously.”
“We see an incredible opportunity to accelerate innovation in precision pathology and to use our strengths to bridge communities in the healthcare ecosystem including patients, biopharma and academic research,” Beck said.