banner of LMP department's DP&AI initiative

Khalifa imagines the AI agent as a “good fellow” trainee at the pathologist’s side

|
William Hoffman
7 minutes
Patients will benefit as pathology enters a new era

Digital pathology and artificial intelligence signature graphicThe fields of anatomy and pathology have had their change agents through two centuries of cell theory: investigators and clinicians who seek to improve their science and practice through exploration, experimentation, and determination. 

In today’s emerging world of digital pathology and artificial intelligence (AI), agency has taken on a new meaning.  “Intelligence agents” or digital assistants are change agents in their own right.  In pathology, they can now be used as aides to patient diagnosis and therapeutic response, clinical workflow, and research at the frontiers of the field.

 

Mahmoud Khalifa photo

Mahmoud Khalifa, LMP Gleason professor and Vice Chair for Anatomic Pathology, imagines an AI agent as a sharp and dependable trainee stand-in. “The way I envision it is, AI is like a good fellow, like a good resident with us,” he said. “The fellow is supposed to help me, for example in drawing my attention to a particular area of the slide and saying ‘please look at this area because I think this area could be x, y, or z.’ The final determination is going to stay with me, but with the aid of a good fellow, instead of spending five minutes analyzing the whole slide I can just zoom in to the area of concern in two seconds and make a determination.” 

 

“It is not about who or what is better,” Khalifa said.  Computer vision is complementary to human vision.  “It facilitates the process and makes humans more efficient.” AI “can be in everything we do.  Whether we like it or not, AI is going to be next to the pathologist as a ‘good fellow.’”

 

The authors of a major study by leading biomedical institutions and technology firms entitled “From tool to teammate: A randomized controlled trial of clinician-AI collaborative workflows for diagnosis” concluded:  “These results underscore the value of collaborative AI systems that complement clinician expertise and foster effective coordination between human and machine reasoning in diagnostic decision-making.”  The term “teammate” will have to move beyond Google’s AI definition as “a person who is a member of the same team or group….”

Visionary institutions and laboratories will recognize that investments in digital pathology and AI today will pay dividends down the line including in marketing their services to patients and their insurers.  “People are going to choose the institution and the lab that can provide the work,” Khalifa said.  “Right now 15 percent of our cases are viewed by more than one pathologist.  I have the data to support me. Imagine if I could say, ‘Come, give us your business because 70 percent of our cases are screened in advance by AI.’  That will be a marketing tool.”

Advocacy and marketing will be needed to advance pathology into the digital age.  The comparison with radiology isn’t quite right.  “Pathology has a huge disadvantage,” Khalifa said.  “When radiology said ‘we will move to digital,’ so you have a patient, and instead of looking at a negative against the light box, all you had to do is take a picture—like taking a picture with your cell phone – and the picture is in the computer.  Unfortunately, pathology is not like that because we still have to use the slide.  That is where the reluctance of the hospitals and those who write the checks.  They say ‘you are adding costs, you are not replacing costs.’” 

With radiology, costs were saved, for example, by eliminating the chemicals used for processing photographic negatives.  “For us, digital pathology was always seen as an added cost,” Khalifa said, adding that there is no case to be made for immediate financial benefit.  Put yourself in the shoes of Congress or CMS [U.S. Center for Medicare and Medicaid Services].  Why would I pay more for the same outcome?  I’m not one of those people who thinks the CPT [Current Procedural Terminology] code will change and the reimbursement is going to change anytime soon. I don’t think this is the right approach to the patient.  I doubt anyone would be willing to pay us more for a digitized slide.” 

The benefits of digital pathology are going to be felt years from now “once we start having efficiencies, once we start building quality assurance, once we start applying all these research results to clinical practice,” Khalifa said.  “As far as I am concerned, if I have my own kingdom to build, I will make the investment up front, and I will have all the financial and quality metrics monitor the work over time, and I will say, in 10 years, I’m going to produce X, Y, and Z outcomes.  One outcome would be clinician satisfaction by answering all their questions.  Yet another metric would be – I say this openly – maybe I don’t need that many pathologists.  Maybe I don’t need to invest in microscopes anymore.  It will take time for people to realize that digital pathology was a good investment.” 

Pathology practices and workflow can indeed experience rapid changes under stressful conditions. A case in point was how pathology adapted to the COVID pandemic, Khalifa said.  “COVID has changed a lot in our lives and told us how much can be done electronically,” he said.  “We are having this [Zoom] meeting now.  This would not have happened before COVID.” 

Khalifa compared COVID to a pressure cooker that forced pathology to embrace a digital world. “Pathologists who were reluctant to look at slides with the computer were forced to do it.  And guess what?  They are happier.”   

The pandemic experience also accelerated the networking of pathologists, which is now widespread.  “If I have consultants who are in different cities – because now there are networks that function this way – specimens from multiple states scanned in a central lab and they are in the IMS [image management system] and viewed by consultants in different cities,” Khalifa said.  “That increases accessibility to consultants and people with experience.  They are not sitting next to the lab that cut the section.  We are moving away from the classic model—a pathologist sitting next to the lab, the slide is prepared, they bring it to me, and I make my diagnosis.  We are moving away from that.  We are moving to the place where lab cuts the sections – that could be anywhere – and the slides are scanned, the scans put in the cloud, the consultants are in Rochester and New York, everywhere.  They have access to the network and can give their diagnosis. That improves the quality.”

As for the so-called challenges posed by digital technology workflow, they are not really challenges, Khalifa said.  “They are nuances.  I envision a different workflow when we don’t have microscopes and everything is done electronically.  Instead of the residents walking into my office with a pile of slides, in Epic [electronic health record system] you will get a notice that there are slides scanned for you, for example.  We learn to adopt digital technology in quality assurance, digital technology in sharing our work, and, look at the consensus meeting we do now, we have people sitting in Hibbing and showing their cases to us.  So I wouldn’t say workflow challenges, I’d say workflow nuances, and new ones need to be developed.”

Artificial intelligence “can be in everything we do,” Khalifa said.  “AI is going to be next to the pathologist as a good fellow and efficiently highlight areas of the lesion to zoom in on.  I have seen some tools that can work the other way around.  We let the pathologist look at the slide first, then the AI will scan the slide sort of as the QA [quality assurance] of the pathologist’s work to see if the pathologist arrived at the right conclusion.  There’s so much room for improving quality.”

In the onrush of digital devices, data, and intelligence agents, pathology is undergoing a sea change.  “And we are happy to be part of it.”

Here are the stories in our DP & AI series: