Labiotech recently interviewed Nathan Buchbinder, chief product officer at Proscia, about digital pathology.
AI has been a hot topic in drug discovery. Can you outline how it’s being applied to pathology data?
Sure, it’s always great to start with such an exciting topic! There are two broad applications of AI that I want to highlight. But before I do, I’d like to provide some important context.
You mentioned pathology data in your question. Pathology data plays a big role in life sciences R&D. In fact, pathology data factors into the discovery and development process for almost every drug brought to market. While scientists traditionally viewed and analyzed this data on glass slides under a microscope, they have more recently adopted digital pathology, which centers around whole slide images of glass slides, to realize a host of benefits – including improved data management, streamlined collaboration, and those that I’ll now unpack related to AI.
To that end, one way in which scientists and research teams are leveraging AI is to unlock new insights from pathology data. AI is capable of recognizing patterns that enable scientists to draw correlations between what’s in the tissue and some of the underlying mechanisms of disease from genomics to proteomics and beyond. In doing so, AI is giving us new information to consider when deciding who to treat, what pathways to explore, and when those treatments are most likely to succeed. We see this playing out today in target identification, novel biomarker discovery, and clinical trials stratification among other areas of the R&D process.
The other broad application of AI is automating and optimizing routine processes. Quality control (QC) is one good example. There are dozens of quality issues – from pen marks to air bubbles – that can render whole slide images unusable, and checking for them can take hours per day. At some organizations, there are entire teams dedicated to QC. An AI solution like Proscia’s Automated Quality Control can minimize this burden for technicians, enabling them to focus on adding value elsewhere and reducing the time that it takes for high-quality data to make its way into the hands of researchers. Biomarker quantification, specifically PD-L1 quantification, is another often-cited example. PD-L1 can play a big role in the diagnosis and treatment of breast cancer; however, it can be very difficult to manually assess. As new drugs for PD-L1 positive patients have been brought to market, so, too, have algorithms that are capable of accurately and reproducibly assessing PD-L1 levels in patients’ biopsies. These algorithms help the practitioner to consistently deliver accurate results and the patient to be matched with the best treatment.
Can you share a bit more about the impact of digital pathology on drug development?
We’re experiencing a new wave of life sciences research brought on by an influx of data along with technologies that are helping to realize its full value. This is incredibly exciting, as it’s leading to innovations that we probably couldn’t have imagined a few decades ago. But with it comes increasingly complex processes and a growing number of partnerships that are necessary for seeing these advancements through.
Digital pathology is empowering research teams to tap into the full potential of data that has historically been trapped in glass slides, as evidenced by the AI examples above. It is also enabling them to overcome their challenges to drive efficiency and scale their operations.
Importantly, while digital pathology starts with the whole slide image, the digital pathology platform plays an essential role. A digital pathology platform like Proscia’s Concentriq for Research connects teams, data, and insights. It’s used to store and access images, eliminating the challenges of managing millions of glass slides. By extension, it also facilitates sharing and collaboration since scientists can share images with internal and external collaborators around the world at the click of a button.
Additionally, the platform unifies diverse technology ecosystems, including the LIMS, image analysis applications, and, of course, AI applications. From here, it becomes even easier to see how AI-enabled digital pathology is optimizing clinical trials stratification, novel biomarker discovery, and target identification, for example. We can also look at the more tactical benefits of digital pathology, such as streamlining the exchange of data and results between a CRO and its sponsor during drug discovery, or enabling more rapid peer review with any number of collaborators.
What impact could the growing use of digital pathology and AI have on the advancement of precision medicine?
Digital pathology is uniquely positioned to advance precision medicine because it benefits both drug discovery and development and patient diagnosis. I can’t overstate the synergistic impact of this.
Since we just covered the role of digital pathology in R&D, let’s take a quick look at how digital pathology is transforming the diagnostic laboratory. This story also starts with shifting away from glass slides and the microscope to whole slide images. And with the help of a digital pathology platform like Proscia’s Concentriq Dx*, laboratories can drive efficiencies in primary diagnostic workflows and increasingly access subspecialist expertise to deliver higher quality services. Concentriq Dx is also designed to incorporate AI applications directly into the workflow to unlock new insights about a patient’s condition and better inform treatment decisions.
What we can now see is that digital pathology and AI are helping to accelerate breakthroughs that can bring new drugs to market and ensure that there’s a more personalized treatment for each patient. In parallel, digital pathology and AI are also enabling pathologists to get a more precise look at each patient’s condition so that he or she can be matched with an individualized course of care. It’s expected that as research breakthroughs increasingly translate into the clinic and help to improve patient outcomes, this will create a flywheel effect that propels even more innovation.
What limitations need to be addressed in the rollout of AI in pathology?
It may surprise some, but resistance from scientists and pathologists is rarely a big blocker when it comes to the adoption of AI in pathology – for both research and diagnostic use cases. Scientists have been among the fastest adopters of AI for the reasons we explored above. Practicing pathologists are increasingly recognizing that AI can help them to work at the top of their license so to speak, equipping them with more information and giving them back time to make more informed diagnoses.
So then, what are the limitations? First, AI applications have historically required a lot of compute power to run. Organizations may be hesitant to implement them because they don’t want to invest in expensive infrastructure to deploy them at scale. While this may be the case for some algorithms, AI applications are increasingly being designed with compute requirements in mind. Many of these applications can leverage existing infrastructure, and some can be run in the cloud, reducing this barrier to adoption.
Second, AI applications can only deliver value when they are incorporated into routine operations. This means that the research organization or diagnostic laboratory must first adopt digital pathology at scale and do so in a way that enables it to seamlessly integrate AI into its digitized processes. The right platform will address these needs; however, the shift from microscope-based pathology to digital pathology is a transformation, and I would be remiss to gloss over the multi-stakeholder commitment required to see it through.
Finally, as in other domains more generally, AI solutions for pathology must be generalizable. They must generalize across sites as well as overcome the variation of data seen in real-world settings, which is especially challenging when it comes to pathology given the number of diagnoses and differences in tissue staining practices and scanning processes. AI researchers and developers are increasingly demonstrating the ability to achieve broad generalizability; however, organizations typically feel most confident adopting AI applications when they perform additional validation first.
Do you see AI completely replacing the role of pathologists in the next decades?
Quite the contrary. AI will only continue to augment the role of the scientist or pathologist, much as we are already seeing today. In fact, I’d go as far as to say that their role will only grow more important as they have more information to guide decisions.
*Concentriq Dx is CE-marked under IVDR and is available for primary diagnosis in the U.S. during the COVID-19 public health emergency.