Rapid Advancements in AI Taking Center Stage at RSNA
“Leading through Change,” this year’s RSNA theme, is evident in the meeting’s extensive coverage of artificial intelligence (AI) applications. What we learn today will be out of date tomorrow, predicts Lawrence Tanenbaum, MD, FACR, neuroradiologist and chief technology officer at RadNet, of the rapidly moving field. At the plenary, scientific, and poster sessions, as well as the AI Showcase, AI Theater, RSNAI Resource Center, and in exhibit booths, attendees will try to stay on top of these important changes.
While AI is revolutionizing medicine as a whole, it’s doing so most profoundly in technology-based subspecialties like radiology, explains Suzie Bash, MD, neuroradiologist and a medical director at RadNet, who says the AI healthcare market is worth about $20B today, projected at $188B by 2030.
“There are over 10,000 healthcare AI patents right now in the marketplace and there's been 30% growth in FDA approval for AI products in 2023 alone. Of all the approvals…79% of those are in radiology,” Dr. Bash reports. “It’s really having a dramatic impact on patient care and on radiologists’ ability to diagnose in a timely manner, and helping imaging enterprises, as well. This is a positive trend,” she says, noting several promising applications, including in cancer screening whereby AI can detect breast cancer one to two years before expert radiologists.
Generative AI, in particular, is experiencing rapid growth in radiology as machine learning models generate new data that provides ways to exceed human capabilities, says Dr. Tanenbaum, who says this will have a large impact on reporting via natural language generation. Large language models are also becoming more robust.
“When you think of generative AI in the healthcare enterprise and foundational models which have multiple capabilities, the impact is going to be almost unfathomable,” says Dr. Tanenbaum, who states that AI’s current applications in triaging urgent cases, like stroke, are significantly impacting clinical outcomes.
“AI tools for triage…are quickly becoming standard of care,” agrees Dr. Bash. “Stroke is the number one cause of disability in the world, and number two cause of death, and so time is brain. This AI technology detects the large vessel occlusion even before the radiologist sees the images, and then conveys the imaging findings to the entire stroke team through a HIPAA-secured mobile phone app, thereby reducing the time from when the patient hits the ER to the time of treatment by 87 minutes, which results in a 40% improved neurologic outcome and two and a half less days in the hospital. AI triage solutions can also detect cervical spine fractures, brain bleeds, cerebral aneurysms, aortic aneurysms, and pulmonary embolism, then prioritize these studies to the top of the radiologist’s worklist.”
Deep learning for image reconstruction is also transforming patient care by enabling 50% to 75% faster scans, with increased image quality, Dr. Bash reports. Patients experience less anxiety with shorter scans, and imaging enterprises benefit from workflow optimization as more patients can be scanned daily. Dr. Bash also states that the superior image quality through AI-enabled denoising and sharpness enhancement results in higher diagnostic confidence for radiologists.
“AI can be integrally involved with the imaging process, so much so that these tools can actually make sure the exam is adequate or complete or free of artifact before the patient leaves. And there is a new branch of tools which we refer to as synthetically generated MR imaging…where if there’s something missing, these tools can restore that image series, avoiding the patient having to come back, and in other cases, can create novel contrast,” says Dr. Tanenbaum. “So not only are we making images faster, which gives us the opportunity to make them better, we’re actually using the same types of tools to make the exams complete, more reliable, more patient-centric, as well as adding additional value.”
One of the most significant developments to impact radiology was July’s FDA approval and CMS coverage decision for Leqembi®. In addition to increased scan volumes, Dr. Bash believes that AI tools will play an increasingly important role in the dementia imaging pathway.
The reimbursement situation, which Dr. Tanenbaum refers to as “one of the biggest headwinds we have” for AI application is noteworthy. Certainly, it’s been prohibitive for practices to use AI for applications without dedicated reimbursement.
“We're right at the cusp of a circumstance where we’re going to have these tools…Hopefully [their] value will translate not just to patients and treating physicians, but to the payers and CMS that they will go ahead and follow through and carry these preliminary tracking CPT III codes to the point where they will be full CPT codes and can be integrated into practices in a financially sound and feasible way over the next months to years,” Dr. Tanenbaum says.
RSNA will give radiologists an opportunity to explore what’s happening at the cutting edge of AI, he concludes. “This is a rapidly advancing field. Where AI capabilities may have had limits a couple of years ago, we’re wildly exceeding those limits today.”