Interest in artificial intelligence (AI) and machine learning (ML) applications in laboratory medicine is at an all-time high. Laboratorians need to have a basic grasp of these systems as they inevitably make their way into clinical practice. Today’s Scientific Session, “Artificial Intelligence and Machine Learning in the Clinical Laboratories: Fundamental Concepts, Clinical Use Cases, and Future Considerations,” will give attendees an opportunity to learn more about the role that AI and ML can play in the clinical laboratory.
Given the large volume of data generated by laboratories, there is tremendous potential for applying AI algorithms at all stages of the testing process. However, to make the most of these algorithms, it is crucial to be able to look at a model and appreciate how it works. In the first part of the session, attendees will learn some of the foundational concepts of AI and ML from Christopher Williams, MD.
Williams will introduce various types of ML, a subset of AI that includes supervised and unsupervised learning. An understanding of the core concepts laid out by Williams can empower laboratorians to start asking questions about why a model makes certain decisions—a skill that will be useful for those evaluating their own models but also for those being approached by vendors with AI-enabled products.
In the second part of the session, David McClintock, MD, will discuss the tools and skills that laboratory professionals need to integrate AI/ML applications in clinical laboratories. If you are wondering whether laboratorians will have to become coding experts, the answer is no. McClintock mentions that there are simple “low-code and no-code” test systems that can help non-experts generate AI/ML models. However, he cautions that it is still imperative to think about datasets critically from the laboratory perspective.
“This is where it is more important not to understand the coding but to understand what the model is looking for,” says McClintock. Basic questions a laboratorian should be able to answer when developing and using a model include: “What am I looking for? How do I structure my data? What are the inputs? What are the outputs?”
McClintock will also describe some of the obstacles that prevent AI/ML from being distributed more broadly. “We don’t expect one group to figure all this out,” he says. Collaboration is key. He believes it will be important to get more people with backgrounds in data science into the laboratory so that the right support is in place from both the lab side and the IT side.
Thomas Durant, MD, will discuss several important issues surrounding the implementation of AI/ML applications, including ethics, interoperability, and regulatory considerations. For Durant, patient safety is a particular concern. He emphasizes that lab professionals need to do their due diligence before integrating AI tools because there can be biases within a model that lead to potential harm.
“We need to collectively as a field develop some degree of technology literacy around these new applications,” says Durant. From there, an acceptable framework can be developed for validating and verifying the performance of these tools.
While AI has the power to transform laboratory medicine, many questions remain unanswered. “There is a time and place for AI, and we’re going to learn what that is,” says McClintock. Laboratorians must be prepared and start to understand where and how AI/ML can fit in. One thing that McClintock makes clear is that “AI and ML are coming to your laboratory.”