The adoption of digital pathology and artificial intelligence (AI), particularly among large health systems and academic medical labs, continues to increase. Though many challenges remain with going digital, these technologies hold great potential for labs and other healthcare providers. Juan Antonio Retamero, MD, an anatomical pathologist and medical vice president of AI solution provider Paige, shares his insights on key developments in the industry and how these trends may progress in the future.
Q: What are some of the biggest recent advancements in digital pathology and AI?
A: Perhaps one of the biggest advancements in digital pathology is the increasing adoption for primary diagnostic use. There is increasing interest on the part of academic medical centers as well as commercial labs to digitize their operations. And as this happens, the use of digital pathology for cases diagnosed digitally worldwide will grow more quickly. These organizations are early adopters and trailblazers and others tend to follow suit. A reflection of this can be seen in the creation of temporary CPT [Current Procedural Terminology] codes for digital pathology, generated to document the adoption of this emerging technology. It means there is more interest from the medical community and FDA in the technology.
In terms of the application of AI to pathology (more formally known as computational pathology), Paige achieved FDA authorization for the first AI application for diagnostic support in pathology for use in 2021. Furthermore, the industry is now also seeing the implementation of large language or foundation models that Paige is training on a vast dataset to eventually be able to create diagnostic aids for all types of cancers. This was totally unthinkable only a few years ago, but as the industry starts to digitize glass slides, more data points become available, and technology is able to merge with medicine in ways that can be really impactful for patients.
Q: What are the key developments on the regulations and standards side of digital pathology?
A: From a regulatory standpoint, we have seen an increasing clinical use of digital pathology scanners approved by the FDA, since the first scanner gained authorization in 2017. However not many AI tools have been authorized to date, and, in fact, no other AI applications have gained authorization since Paige Prostate received approval over two years ago. The reason for this is largely because AI tools must generalize and perform well with data that is different than what the algorithm was trained on. For this reason, some AI tools require in situ retraining and calibration, which is not something regulators favor. Paige Prostate is different in this regard since it was trained using a very large dataset, which exposed it to variations in tumor morphology, histological variant, preanalytical factors, patient variables, et cetera. Therefore, it does not need to be retrained or calibrated with local data and will perform optimally out of the box. This sets it apart from any competitors trying to introduce AI in this space.
Q: Can you discuss the integration of digital pathology/AI into healthcare systems? How is it changing the way pathology services are delivered?
A: The adoption of digital pathology for routine diagnosis is still growing, but we are witnessing increasing demand for clinical-grade digital pathology scanners, which will ultimately translate into full-scale digitalization. Digitizing an entire lab workflow is challenging for a number of reasons, but if we think that pathology is one of the few medical disciplines that still operates in the analog world, we can see how the digital transformation of pathology is inevitable. This creates countless opportunities. Early adopters report an increase of efficiency together with diagnostic quality associated with digital pathology, and this has sparked significant interest from academic medical centers together with commercial labs to digitize their pathology operations. Furthermore, the application of AI in pathology relies on digital transformation, which is a condition sine qua non for computational pathology. Digital pathology in itself brings about significant advantages to lab operations, but it is the implementation of AI that, combined with the expertise of pathologists, elevates the performance of these pathologists well beyond what is achievable with a humble microscope.
Q: What are some challenges healthcare providers/labs may run into when adding digital capabilities?
A: First of all, to apply AI in pathology, digital pathology has to be implemented. This is challenging because in order to digitize the pathology lab operations, a series of steps must be taken. It is essential to have a level of integration of the IT infrastructure involving the laboratory information system, together with the scanner and the digital pathology viewer in order to assign the right digital slide to the right patient and to trigger the right AI algorithm on the right case on the right digitized image. Also, histological glass slides have to be prepared somewhat differently to facilitate the scanning process. Finally, there is a significant element of change management since not all lab personnel and pathologists are enthusiastic about the idea of going digital and implementing AI as a diagnostic aid. These difficulties must be addressed and managed before digital pathology and AI can be implemented.
Early digital pathology adopters have encountered difficulties when it comes to implementing AI on their digitized slides. This is because not all vendors can integrate AI tools in their workflow, and in some scenarios the integration with these AI tools is less than optimal, which severely diminishes any potential gains. For this reason, it’s important to consider solutions that offer the greatest flexibility and ability to integrate third-party applications seamlessly, in order to facilitate the pathologist’s job, so they can focus on their diagnostic work instead of contending with different IT elements that do not interact well with each other.
Q: What are some ways to overcome these challenges?
A: The process of digitization is challenging in itself, so a great deal of commitment is required on the part of the institution undergoing this transformation. But also, IT and digital pathology and AI vendors need to integrate more closely and realize that digital pathology, while beneficial, is not an end and merely a necessary step in the adoption of computational pathology.
Q: What recent innovations in digital pathology/AI are you most excited about or impressed by? Why?
A: High-throughput scanners that are capable of digitized routine histological slides without human supervision in increasingly short periods of time are essential for the digital transformation of pathology, and we see that these clinical-grade scanners achieve these goals more proficiently each day. In computational pathology, we are witnessing the development of AI tools that integrate with each other to assist pathologists in every step of their workflow. Tumor detection algorithms prioritize cases in the day’s worklist and assign them intelligently to the right pathologist, suggesting the request of ancillary techniques, and work together with large language models that help pathologists generate reports. Also, the advent of foundation models will revolutionize how AI is applied to histologic diagnosis, ranging from routine tasks such as cancer detection, grading and quantification, to the detection of biomarkers in plain H&E [hematoxylin and eosin] slides, that will help identify and prioritize slides for downstream molecular testing.
It is important to realize that AI is here to help pathologists do their jobs, and by no means to replace pathologists. Think of it as having another weapon in the diagnostic arsenal that, much like immunohistochemistry or molecular pathology, must be interpreted and integrated by the pathologist. For this reason, pathologists are far from being replaced by technology anytime soon.
Q: How do you expect the key trends in digital pathology to progress moving forward?
A: We will continue to witness an increasing trend of digitization, present not only amongst academia, but also in commercial labs, driven largely by the efficiency and diagnostic quality gains that are simply not realizable with a microscope on an analog workflow. This increasing digitization will be reflected in the creation of permanent CPT codes that will lead toward reimbursement associated to the use of these new technologies. The industry will further consolidate and integrate in order to facilitate this holistic AI-assisted workflow, rather than the current paradigm, largely centered around image analysis, which is just one of the many aspects in which AI can be applied to pathology. For these reasons, the number of samples, and, ultimately, patients that are diagnosed with these novel and exciting tools will only grow in the near to medium term.
Juan Antonio Retamero, MD, is an anatomical pathologist with extensive experience in the implementation and use of digital pathology and artificial intelligence in clinical use. He played a key role in the adoption of digital pathology for routine diagnosis in a pioneering group of hospitals in 2016. Since then, he has shared his experiences worldwide and has supported many centers in the adoption of digital pathology, including labs in the United States and many countries in Europe and Asia. He is a regular speaker at various digital and computational events globally and is an ardent advocate for the modernization of the profession.