Expert Q&A: Harnessing the Power of AI

Laboratory medicine has witnessed a transformative shift with the advent of artificial intelligence (AI). Its rise has sparked considerable interest and debate among laboratory professionals—and although clear benefits are emerging, many pathologists are wary of the ethical and logistical hurdles that accompany AI.¹ One prevalent concern surrounds the risk of cyber threats to sensitive patient data, which can hinder pathologists’ willingness to embrace AI and adopt it into their workflow. To find out more about AI’s emerging place in the lab and its impact on cybersecurity practices, we spoke with Anil Parwani, MD, PhD, MBA, professor of pathology and director of pathology informatics at The Ohio State University.

Q: How is AI currently used in laboratory medicine?

A: AI is used to assist with a diagnosis or parts of a diagnosis. For example, there are algorithms that can carry out parts of the tumor grading process and then present the findings to the pathologist. They can also augment a diagnosis by putting the grade together for you or identify features you might have missed. Furthermore, they can screen cases and place the most abnormal cases at the top of the stack so they are prioritized and reviewed first by the pathologist.

AI tools can also help create pathology reports and support quality assurance by checking work and alerting the pathologist to potential cancer they might have missed. Finally, it can risk-stratify patients by predicting those who will have the worst diagnosis and those who will benefit from certain treatments versus others.

In most cases, the pathologist will review all of the AI’s work, but we do have some more autonomous systems. For example, pathologists used to review every cervical smear, but then cytotechnologists came along and they began to screen the cases instead. Now, AI can screen the smear before the cytotechnologist reviews and signs off—but, in cases where AI picks up something abnormal, the pathologist also reviews it.

Q: What do labs need to consider when using AI?

A: Transparency is key. These systems are designed based on the use of patient material, so AI has learned from patients—but how involved are those patients? Did they give consent to this research? Similarly, how transparent should you be with the patient about the fact that a portion of their diagnosis was provided by AI? When a specimen goes into the lab, it is assumed that a pathologist has reviewed it completely; however, if AI was used, you need to be transparent and let them know or put a disclaimer on the report.

Q: What barriers do labs typically encounter when implementing AI?

A: Cost is a big barrier because AI-based diagnostic assistance is often not reimbursed. These systems require a tremendous amount of computing power and infrastructure, which are not easy to build or buy, and there are no billing codes to provide additional revenue to offset the costs.

Integration into the laboratory information system workflow can also be a barrier. There is a concern from pathologists that these systems will not be HIPAA-compliant or safe because, often, you have to send your analysis out into the cloud for processing because the hospital doesn’t have the computing power or infrastructure to do it on-site. Some labs are reluctant to do that because they fear that somebody will break into the system and obtain or distribute patient information. These systems are generally very safe when they are on HIPAA-compliant servers with built-in safety features, but it’s the fear of the unknown—the fear of losing control. Pathologists are still concerned that AI is not safe or accurate and will take away their jobs.

Q: How can labs protect patient data from cybersecurity threats?

A: Cybersecurity practices are becoming more relevant because we are doing more work that uses external systems. It is good that we are protecting patient data, but the burden is now on us to ensure its safety. Pathologists have to be very careful about where they send the data, what’s in it, who sees it, and where it is stored. Also, consider where the reports are being generated—is it inside or outside the firewall? They need to look at every loophole in the equation.

Labs must invest more in cybersecurity by dedicating a portion of their capital or overall operating budget to it. There are pathologists who say they want to use AI, but that the cost and time to have the tools cleared are not worth it. That’s not going to go away. It may become easier over time, but it will always be a big part of implementing AI.

Q: What excites you most about AI in laboratory medicine?

A: It’s an exciting time because we are at the crossroads of precision medicine and new therapeutics—and pathologists have a critical role to play. Two patients may have similar-looking cancers, but the tumors will behave differently and need personalized treatments. AI will help us recognize those differences faster. It’s also very exciting to train the next generation of pathologists with these tools and technologies.

We also have a new area of AI that allows us to predict which patients will benefit from a specific treatment. Therefore, instead of giving a $30,000 treatment to every patient who might not need it, we can give it to only those who are most likely to respond to it. That will then reduce overall healthcare costs, benefiting everyone.

Q: What message do you have for pathologists who are reluctant to adopt AI?

A: Think of AI as a friend that is going to assist with your daily tasks and augment your diagnostic capabilities. We are dealing with increased workloads and staff shortages while learning more about new diseases and treatments than ever. Pathologists with heavy workloads cannot keep up with all of this information, but AI can help. Investing in AI will help prevent burnout and free up your time to devote to patient care, so don’t be afraid of it; instead, learn about it so that it can empower you and make you a better pathologist.

Reference:

1. Shafi S, Parwani AV. Artificial intelligence in diagnostic pathology. Diagn Pathol. 2023;18(1):109. doi:10.1186/s13000-023-01375-z.

______________________________________________________________________________________________________

Anil Parwani, MD, PhD, MBA, is a professor of pathology and biomedical informatics at The Ohio State University. He serves as the Donald A. Senhauser Chair of the Department of Pathology and the chief of pathology services for OSU’s health system. His research is focused on diagnostic and prognostic markers in bladder, prostate, and renal cell carcinoma. Parwani has expertise in surgical pathology and pathology informatics including biobanking, whole slide imaging, digital imaging, telepathology, image analysis, artificial intelligence, and lab automation. He has authored over four hundred peer-reviewed articles in major scientific journals and several books and book chapters. Parwani is the editor-in-chief of Diagnostic Pathology and co-editor of the Journal of Pathology Informatics.