By - November 14, 2023
Artificial intelligence (AI) tools are already in use in pathology and laboratories, and they are definitely here to stay. While their current usage is relatively small now, largely confined to digital pathology and automation tools, we’re likely to see a massive evolution of their sophistication, improving many functions within laboratories and for pathologists and laboratory professionals. Experts explain some of the ways that AI will transform the laboratory over the next decade.
“The next five years are going to be really exciting as we see increasing utilization of these tools that really aim to support pathologists, making them more accurate, more reproducible, and also in the future have access to new diagnostics that can't be done any other way,” says Andy Beck, MD, PhD, CEO of PathAI, an AI-powered pathology technology company.
PathAI has been working on AI algorithms to help with the interpretation of liver biopsies, for example. Currently, to diagnose severe nonalcoholic steatohepatitis (NASH), pathologists determine a score based on factors such as the amount of inflammation in the liver, the number of ballooning hepatocytes (diseased liver cells), and the degree of fibrosis in the liver, Dr. Beck explains. They then assess these criteria to deliver a score from zero to four. This is a difficult manual task without the assistance of an AI tool helping the pathologist to count each of these things, Dr. Beck says.
“Instead of a pathologist acting alone and looking at it under the microscope without any assistance, the next step after staining is to put these slides into a whole slide imager to create large image files, and then essentially process those slides with an AI system whose job it is to do very similar to what you might ask a resident or trainee to do in theory.” However, the AI will be “far more exhaustive and accurate, and exact, than what any trainee could do,” Dr. Beck says.
The AI systems will be able to label all the relevant pathologic components and even offer a preliminary score. “The pathologist will no longer have to count the cells individually, but take all the data, synthesize it, and either agree with the preliminary read from the AI system, or change it if they think that's incorrect.”
AI along these lines will take the manual work of cell counting out of pathologists’ hands, making their work more efficient and more accurate.
Predictive AI algorithms are also in development in a wide variety of use cases in pathology, according to Dr. Naveed Rabbani, MD, Fellow Physician in Clinical Informatics at Lucile Packard Children’s Hospital, Stanford Medicine.
A pediatrician who works with pathologists, Dr. Rabbani is part of a research team at Stanford developing AI tools for “laboratory stewardship,” namely ordering the right test at the right time, and not over-ordering tests.
They are developing machine learning algorithms that can predict the results of a blood test that hasn’t been done yet based on information in a patient’s medical record to identify if a test is worth doing or not.
“This will help cut down on unnecessary blood tests, which is good for the health system, because it means less waste, and good for the patients who don’t have to be awoken in the morning for a blood test that might not be necessary,” Dr. Rabbani says.
Given that many tests can be expensive, and that supplies to collect blood are in shortage, an algorithm like this can offer big savings and save patients from tests they don’t need.
In similar research, Dr. Rabbani mentioned that algorithms are in the works that will one day offer guidance to physicians to suggest blood tests, as well, referred to as recommender algorithms. “Consider the scenario where a physician has a patient with a suspected diagnosis for which they have referred to a specialist. A recommender algorithm can help the physician order the necessary blood tests ahead of time, so that patient can have the tests done before they see the specialist and take full advantage of the visit the first time around. No time wasted.”
AI will also likely become more integrated into anatomic pathology, along the lines of region of tumor identification and classification, able to alert the pathologist when something is abnormal and that they should pay attention to a specific area of tissue. Many of these tools are already in the works.
According to a study in The Lancet reporting on the results of a new AI-based digital pathology tool for pathological assessment of cancer resection specimens, “Clinical validation showed the tool could have substantial benefits when used by pathologists, including increased diagnostic accuracy for detecting tumour and regression tissue and for regression grading, reproducibility…better prognostic information, and reduced time expended on analysis.”1
Another area where AI may improve workload for pathologists and pathology professionals is through the use of large language models (like ChatGPT), which are getting very good at language creation. “You could picture an AI system writing the first version of a report,” Dr. Beck says.
Moreover, since these generative tools are very powerful at adapting styles to specific audiences, he suggests, “You could have a style adapted for pathologist, a style adapted for a general practitioner, for an oncologist, for a patient, for patients of different ages.” Ultimately, AI might be able to help communicate the same information in different styles, optimized for a number of different stakeholders.
Even more exciting, AI may one day be used to create an interactive pathology report, Dr. Beck says. “So, imagine if you could ask questions of your pathology report, and get true answers back that are actually helpful in real time. You wouldn’t have to focus on sending an email to your doctor but have a lot of that interaction be done directly with the AI system that is super knowledgeable about your health history as well as what's going on in your particular case.”
In short, AI technology is experiencing “astronomical growth,” according to Joseph Rudolf, MD. While that does threaten some laboratory professionals, he suggests that this kind of growth is in sync with changes in the laboratories themselves. Dr. Rudolf is Assistant Professor of Clinical Pathology at the University of Utah School of Medicine, Medical Director of the Automated Core Laboratory, and Medical Director of Clinical Informatics at ARUP Laboratories.
“When I look back over the history of the laboratories, it’s always been a place of constant change and innovation and adaptation. New jobs come online, and some other jobs cease to exist or in their current form. But we’ve always been able to make these pivots with our people, our labor force, and our practices, to meet the needs of the modern day,” Dr. Rudolf says.
He sees AI tools as “an add,” saying, “We don’t know what they’re going to look like exactly, but in the long run I think they’re going to improve healthcare and improve our practices. I just think we should be very thoughtful about how we deploy them.”
Tolkach, Yuri, et. al. Artificial intelligence for tumour tssue detection. The Lancet. May 2023. Retrieved from: https://www.thelancet.com/journals/landig/article/PIIS2589-7500(23)00027-4/fulltext