Contributed Commentary by Eric Walk, MD, FCAP, Chief Medical Officer, PathAI
May 31, 2024 | After decades of false starts, digital pathology adoption is finally happening. Over the last two years, I’ve had the pleasure to travel around the world to dozens of pathology labs and directly observe this technology transition. I’ve visited the world’s largest reference labs as well as hospitals and smaller practices, but regardless of size, it’s truly amazing to witness a fully digital pathology lab. Pathologists in these labs don’t have glass slides anywhere in their offices and have completely adapted their workflow and mindset to a digital world. They experience ‘faster than glass’ workflow advantages and tangible benefits associated with AI tools, but also have learned important lessons along the path to adoption. These lessons learned can inform the rest of the field as we all think about bringing digital and AI pathology into mainstream practice.
Digital and AI Pathology Basics
Digital pathology (DP) involves the digitization of pathology glass slides into digital whole slides images (WSI) via the use of whole slide scanners. Basic DP unlocks applications such as primary digital diagnosis, remote diagnosis, consultations, virtual tumor boards, and education.
AI or computational pathology builds on DP and utilizes machine learning (ML) algorithms to extract tissue and/or cellular information from a WSI that is helpful for pathology workflows. AI pathology algorithms unlock additional advanced applications beyond DP that span three main application categories: 1) Workflow, 2) Diagnostic, and 3) Precision Medicine, as detailed in Figure 1.
The Image Management System (IMS) and the Advantages of AI-Nativity
In the era of digital and AI pathology, the Image Management System (IMS) is a necessary piece of laboratory software that serves as the digital repository for digital slides, but also the portal for case management, storage, and retrieval, in addition to being the hub for AI pathology. The IMS is ideally bidirectionally integrated into the Laboratory Information System (LIS) to enable importing of case metadata and delivery of data back to the LIS and pathology report.
A misconception about digital and AI pathology is that they are completely modular and interoperable. While universal interoperability is the ultimate goal and vision in the DP field, we are not there yet. Pathologists and laboratory professionals should be aware of the advantages and disadvantages of the different IMS and AI scenarios currently available:
AI Native IMS (preferred scenario): Either a single company develops both the IMS and AI, or the IMS developer works very closely with the AI developer to add integration via application programming interfaces (APIs) – true examples of the latter are currently rare due to cost and time.
IMS contains fully integrated AI algorithms and all necessary workflows, data fields and overlay functionality the AI algorithm requires.
Visualization of AI algorithm output is seamlessly displayed in the viewer itself and not in a separate window.
AI output is brought directly into the case-list view, enabling advanced workflow applications such as intelligent case distribution, load balancing, case prioritization, and reflex test ordering.
AI data (e.g. biomarker scores) can directly populate the appropriate fields in the LIS and pathology report.
IMS with partial integration of Third Party AI: Many IMS developers do not develop AI algorithms and many AI developers do not maintain an IMS, requiring partnerships between companies.
Typical integration is partial and contextual.
AI applications open a separate viewer and AI output is returned to the IMS to then be returned to the LIS.
Disadvantages: 1) if the IMS does not contain enough data fields (e.g. data field limit) to support AI algorithm output, data loss can occur. 2) the IMS may not contain core functionality (e.g. heatmap visualization) required by the AI algorithm. 3) spawning of a separate AI window from the IMS creates desktop clutter, potentially leading to diagnostic confusion.
Standalone AI application without IMS functionality:
AI application contains a simple viewer but lacks full IMS functionality necessary for comprehensive case management and workflow.
Least desirable scenario since a separate IMS would be required
Need to create a custom interface between the AI application and IMS to manage data flow and LIS connectivity to create scenario #2.
Pathologists and laboratory professionals need to be aware of the various DP and AI solution scenarios so they can make informed decisions about which option best meets the needs of their laboratories, both now and in the future. Laboratory decision makers need to realize that the IMS decision they make now has the potential to enable or limit their AI application options and functionality in the future. A detailed understanding of how all digital and AI solution elements interact with each other will become even more important as the field continues to evolve and progress.
Dr. Eric Walk is Chief Medical Officer at PathAI. He has 20+ years of experience in precision medicine, oncology drug development and IVD companion diagnostics development. He graduated from Johns Hopkins University and holds an MD from the University of Virginia School of Medicine. He is board certified in Anatomic and Clinical Pathology and is a Fellow of the College of American Pathologists (CAP). He can be contacted at eric.walk@pathai.com.