Unlocking the potential of AI in radiology with real-life applications
Dr. Peter Lafferty, Chief Physician Integration Officer at LucidHealth is interviewed by Ariella Shoham, VP of Marketing at Aidoc about the value of artificial intelligence (AI) in radiology workflows, best practice in using AI in radiology, and how AI can provide better outcomes for patients.
LucidHealth has already implemented AI, using the technology as a safety net to ensure that turnaround times are reduced and that high clinical quality is maintained. In this Q&A, Dr. Lafferty explains how his practice has leveraged AI to achieve measurable results and some of the best practice processes that have been undertaken to improve workflows. We speak to Dr. Lafferty about how AI is changing the shape of radiology today and what this means for the patient and the practice.
Ariella Shoham (AS): Dr. Lafferty, can you tell us a bit more about your organization and the work that you do?
Dr Lafferty (DL): We’re a community radiology practice that’s committed to being a leader in community radiology – it may sound odd, but our company culture is built around patient care, not just scale. We’re constantly looking at ways in which we can scale to deliver care to more people but without sacrificing our commitment to patients. We specialize in digital so we have the ability to do things in patient care that can’t happen in more traditional facilities. We started out as traditional, but then we became a platform for digital because we realized that we had to change our stance if we wanted to remain competitive. There were too many companies wanting a slice of our pie, so this meant we had to invest into a sub-specialty model and technology to create sophisticated workflows that have transformed our patient care. This commitment to people and digital has become ingrained in our culture today. Everything we are doing is about bringing together the best tools in information technology to support our human intelligence.
AS: How do you drive your patient-centered vision?
DL: We bring the right patients to the right radiologists at the right time. We recognized that patients’ images have been placed in multiple silos – moving from one independent physician and imaging center to the next. This makes it complex for both patient and practitioner. We look over the top of these silos and have developed workflows that can present all the data connected with a single patient to a single radiologist. We centralize the data to improve both access and efficiency. Our goal is to empower both the physician and the patient. One way we do this is to change our radiologist’s roles around. One year I may be reading mammograms, the next year I am reading PEs. This minimizes the observability bias that can occur if a radiologist is looking at the same mammograms year after year.
AS: What about your AI investment?
DL: Thanks to our unique structure and culture we were able to inject AI meaningfully into our system. This allowed us to improve patient care and outcomes at the enterprise level, not just the facility level and I’m extremely excited about its potential. This is already up and running in Ohio and will soon be in our Wisconsin and Iowa practices. We are using the Aidoc Intracranial Hemorrhage algorithm and have already demonstrated how well it works. We’re often presented with lots of cases from emergency departments late in the day and the algorithm notifies that there is an acute case in the list so that it’s accessible to the next available radiologist that has the relevant qualification.
AS: Do you believe that the AI can transform workflow and better support the radiologist?
DL: We realized that for AI to be effective we had to understand how it really works in practice. As we have such a diverse hospital population, it presented us with an opportunity to learn how well AI performs in this setting. We designed a peer-reviewed study led by the clinical team at LucidHealth that identified the need to focus more on ED exams as those patients have a higher potential clinical significance. We looked at 2, 000 exams in the ED, then we ran an NLP tool and classified them by either negative or positive exams based on the reports. Then all the negative exams were reviewed by the algorithm to see if it could detect anything and it flagged 14 cases which we then reviewed to assess if they were false positives or accurate. The initial data was extraordinary. Initially I thought this study was going to be a waste of time, but now I feel exactly the opposite. I thought there would be a lot of false positives because we are a high functioning practice with extensive experience. Yet this combination of radiologist and machine has the potential to allowed us to detect additional true Intracranial hemorrhages. This has the potential to set the bar on patient outcomes and patient safety – the combination of people and machines is so much better than people on their own and the use of AI can truly transform how we deliver patient care and radiologist support.