Advanced tools and new analytical approaches promise to alter the shape of diagnostic medicine forever.

Interview by Steve Halasey

The way that digital tools are being deployed is changing forever the shape of diagnostic medicine, along the way creating the emerging field of digital pathology. According to field experts, applications based on the use of whole-slide imaging and digital repositories are making diagnoses faster and more accurate than ever.

To find out more about the current status of development and adoption in the field of digital pathology, CLP recently spoke to Michael C. Montalto, PhD, vice president and head of translation pathology and clinical biomarker technologies, translational medicine, at Bristol-Myers Squibb, and president of the Digital Pathology Association.

CLP: The emergence of digital pathology as a new diagnostic specialty area is heavily reliant on the development and adoption of advanced imaging systems and related analytical tools. How would you characterize the current state of such technologies for clinical lab applications?

Michael C. Montalto, PhD: Certainly, a lot of progress has been made when it comes to digital pathology product offerings designed for clinical use. There are several whole-slide imagers and associated software products designed specifically for the clinical laboratory, many of which were developed under stringent medical device regulations for both the US and global markets. There are also a wide range of options, including high- and low-volume scanners, image storage solutions and vendor-neutral archives, viewers with annotation, and real-time collaboration tools, as well as digital workflow software designed to interface with a laboratory information system (LIS) and to have the functionality needed to handle the demands of anatomic pathology laboratories. In fact, there are a few large academic medical centers in the US and EU that have recently made investments to digitize their entire anatomic pathology workflow using such products.

Even so, there is no ‘one size fits all’ solution for the varied range of use cases encountered in clinical labs. A laboratory interested in using such technologies for clinical applications may need to piece together different parts of the solution to fit their needs, or to work directly with vendors to help customize solutions.

CLP: A number of traditional instrument suppliers, such as Leica, PerkinElmer, and Roche, have begun to develop imaging and analytical systems for use in digital pathology, but only the Philips IntelliSite system has so far received FDA clearance for primary diagnostic use. Why have companies been slow to seek regulatory clearance or approval for their systems?

Montalto: Getting FDA market clearance or approval for a novel technology that does not have a ‘predicate device’ on the market is often challenging, because there is no defined playbook to follow. Most companies that want to register a similar whole-slide imaging device for primary diagnosis have been in a ‘watch and wait’ mode while awaiting clearance of the first device—the Philips IntelliSite system. Now, other companies can use the Philips device as a predicate and follow the same submission requirements. Theoretically, this should allow for more devices to become registered in the coming years. However, obtaining FDA premarket clearance requires a significant investment of resources, and can take at least a year from start to finish. Such difficulty is likely one reason that companies have been slow to submit their new products for regulatory approval.

Importantly, the technology in this space is evolving very quickly. If companies introduce new devices that have functionalities different from those of the IntelliSite system (for example, artificial intelligence-based algorithms, or multiplexed testing), then there is again no predicate device, and the regulatory path will be less clear. In this case, newer technologies may still have a long road ahead before they can get FDA clearance or approval.

CLP: Medical researchers appear to be very active in exploring new applications of digital pathology, but uptake in clinical diagnostic labs has been slow. What obstacles are hindering the adoption of digital pathology for diagnostic applications?

Montalto: There are many adoption barriers, from both a technical and a business perspective. On the technical side, the implementation of a digital workflow is very complex. Simple issues such as bar coding or network bandwidth limitations can become major hurdles for an institution. Linking the digital data to laboratory information systems is also a challenge—especially with many institutions having unique LIS solutions. Navigating such complexities requires a true commitment to going digital from the entire institution, including the information technology department, pathologists, and hospital administrators. Change management is a big consideration.

On the business end, a big adoption barrier is the simple value-cost equation. Much of the proposed value of digital pathology is aimed at improving workflow efficiency (using a machine to identify diagnostic features more quickly than a pathologist could do, for example). The healthcare system, on the other hand, is designed to pay for outcome improvements—and not necessarily for workflow efficiencies. Consequently, the return on investment in digital pathology remains somewhat theoretical. Nevertheless, with the continued introduction of new technologies at lower price points, and the fact that uptake is increasing, I am hopeful that we will start to see more studies that unequivocally demonstrate the value of digital pathology. Such studies will certainly help to accelerate adoption.

CLP: At this stage of systems development, how much work is being devoted to refining instrumentation, how much to developing novel detection and visualization technologies, and how much to exploring applications and analytical systems for specific diseases?

Montalto: This is hard to say. If I look at where venture capital is being directed in this space, as evidenced by the number of new startups, I would say more investment is happening in the artificial intelligence and automated analysis parts of the systems as opposed to hardware. We will still see some improvements in scanners—to make them less expensive with higher quality and speed, for example—but I think the majority of development now is in interoperability and automated image analysis applications.

CLP: In your view, how important are automated imaging analytics, deep learning algorithms, and artificial intelligence systems for making digital pathology useful in clinical settings?

Montalto: These technologies—and especially deep learning approaches—are extremely important for making digital pathology useful. Regardless of the use case, deep learning tools are the fastest and most reliable method for developing highly accurate detection algorithms, which in turn serve as the basis of visualization and feature identification tools. Several studies have started to show that deep learning-based detection methods identify specific pathological features more accurately than an individual human pathologist can do, especially when the ground truth is based on a consensus score among several readers.

CLP: What is the current state of play when it comes to such automated analysis systems? Are institutions pretty much on their own, or have telepathology applications begun to make regular digital pathology consultations a reality?

Montalto: Telepathology-based remote consultations are a current reality, and many institutions are using digital pathology systems for this purpose. However, the implementation of automated image analysis into routine workflows, even for teleconsultation, has not occurred widely. Telepathology technologies are still in their infancy in terms of clinical practice, and many institutions are choosing to explore them for research applications. Once the algorithms are introduced into clinical systems, and once large-scale validations can demonstrate accuracy even in the face of the widespread preanalytical variabilities that exist in pathology, then I think we will see increased adoption. Until then, pathology labs are likely to continue exploring and trying to devise methods for validations in their own labs.

CLP: Are reimbursement coverage and payor pricing major problems for digital pathology developers? How can companies overcome such challenges?

Montalto: As we discussed, the cost-value equation is very challenging, with current use cases aimed at workflow efficiencies, so reimbursement remains a challenge. In addition, there are currently no specific reimbursement codes for the use of digital pathology, outside of a unique code for the digital analysis of breast immunohistochemistry markers (CPT 88361), which is a very limited use case.

Once the technology advances to the point where we can measure enhanced patient outcomes, we may start to see some change in the reimbursement environment for digital pathology. For the time being, however, attempts to increase reimbursement for the use of current systems will likely be a challenge. In effect, this means that institutions will need to develop their own justifications for investing in the technology. Given the potential of digital pathology systems for achieving increased workflow efficiencies and reduced interreader errors, such justifications should not be difficult to find. In the meantime, however, implementing digital pathology systems will require that the institution absorb the upfront costs and find the savings to justify them.

CLP: Assessment of tissues for suspected cancer has been an early focus of digital pathology. How has the technology advanced clinical capabilities in this area?

Montalto: Deep learning and automated image analysis have demonstrated an impressive ability to detect certain types of cancers, although this has so far been done primarily for the purposes of academic research and not for clinical use. Specifically, automated analysis has shown promise for research applications that are particularly time-consuming or fraught with interreader variability, such as identifying metastatic disease in lymph nodes, or differentiating complex prognostic states, as in Gleason grading for prostate cancer. More recently, deep learning research has advanced to the point where deep learning tools can identify specific mutations with high accuracy from hemotoxylin and eosin stained slides.1 This is an amazing demonstration of the power of deep learning in pathology.

CLP: What other diseases are likely to benefit from advances in digital pathology?

Montalto: In my view, any disease that is identified using a microscope could potentially benefit from digital pathology. In particular, routine screening applications could benefit from having a computer do a first pass and quickly highlight areas of concern for a pathologist to confirm visually. Very time consuming diagnoses, such as rare event detections in infectious diseases or cancer detection, could benefit greatly from digital pathology.

CLP: For the future, how important will teleconsulting capabilities of digital pathology become? Do you foresee a time when digital pathology analyses might be performed 24/7 at overseas sites, as now occurs for x-ray and other imaging consultations?

Montalto: One of the main use cases today for telepathology-based systems is remote interoperative consultation, specifically in circumstances where having a pathologist cover a remote site full time is not possible. It is not clear whether 24/7 overseas services will become as important in pathology as they are in radiology. While possible, we have not seen this use case often.

CLP: What other aspects of digital pathology are likely to influence the development or adoption of such systems in clinical lab settings? Are we likely to see a sudden growth spurt in the next couple of years?

Montalto: Adoption of digital pathology is very hard to predict. Certainly, we are seeing growth in the number of institutions that are starting to explore digital pathology for clinical use. The alleviation of regulatory obstacles and the implementation of software interoperability-based standards will help accelerate adoption.

There are two emerging innovations that will also help to accelerate adoption over the next several years: namely, advances in precision medicine and deep learning technologies. For example, there has been amazing progress in the area of immunooncology for the treatment of many cancers, but we are not yet seeing consistent responses among patients, and even among those who do respond to treatment, many experience relapse. In the future, by using artificial intelligence coupled with digital pathology, we may be able to understand the complex interplay of the immune system in the context of the tumor microenvironment, which could lead to the development of advanced diagnostic methods to predict who will and who will not respond to immunooncology-based treatments. Digital pathology could have a big impact on the way we treat serious diseases, and I believe the real-world application of the technology in this way will ultimately be the catalyst for widespread adoption.

Steve Halasey is chief editor of CLP.

Reference

  1. Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non-small cell lung cancer histopathology images using deep learning. Nat Med. 2018;24(10):1559–1567; doi: 10.1038/s41591-018-0177-5.