AI Augments Cath Lab to Predict Outcomes
LONG BEACH, CALIFORNIA — Artificial intelligence (AI) has been able to extract functional and physiological data from routine coronary angiography and predict four key cardiovascular biomarkers with an accuracy or 80% or greater, according to researchers at the Mayo Clinic.
Mohamad Alkhouli, MD
“AI can be leveraged to identify clinically meaningful information from routinely collected angiograms,” Mohamad Alkhouli, MD, division chair of research and innovation at the Mayo Clinic Alix School of Medicine in Rochester, Minnesota, said here at the Society for Cardiovascular Angiography and Intervention (SCAI) 2024 scientific sessions.
While presenting results from the AI-ENCODE study, Alkhouli said that his team used advanced machine-learning techniques to extract data from 20,000 angiograms performed at the Mayo Clinic.
The team trained multiple AI algorithms to extract data on left and right ventricular functions, intracardiac filling pressures, and cardiac index using one or two angiographic videos. Echocardiography served as the comparator used to validate three of the algorithms; simultaneous right heart catheterization measures were used to validate cardiac index.
The models predicted left ventricular ejection fraction, left ventricular filling pressures, right ventricular dysfunction, and cardiac output with a high degree of accuracy; area under the curve was 0.87, 0.87, 0.80, and 0.82, respectively (a score of 1 indicates 100% accuracy), Alkhouli reported.
The team is still perfecting the algorithms before adopting them in the clinic. “We need to refine the model and make it more than a prediction model of one or two things,” Alkhouli explained after his presentation. “We want to predict multiple measures and install it into a dashboard.”
Multiple Measures on a Dash
The goal is to enable interventional cardiologists to access those data in real time in the catheterization lab, Alkhouli said.
Next, the researchers plan to develop algorithms that can predict heart valve calcium, pericardial restriction, transplant rejection, and regional wall motion abnormality. “That would include data cleaning, external validation, and then building back the IT infrastructure that can communicate with the cath lab,” he said. That stage will take at least a year, maybe 2, he added.
Clinicians have concerns about AI, Alkhouli acknowledged, but it can serve as a valuable tool.
“In this study, two things were obvious,” he said. “One is that AI can actually let you focus on your procedure, but it will also supplement you with all these other data so that you can simultaneously focus on the angiogram, the ejection fraction, the blood pressure, the cardiac intake.”
“As humans, we’re more intelligent than AI, but we have less capacity,” he continued. “AI will supplement us with the bandwidth so we can achieve bigger things. AI is not a threat. We should use it intelligently to supplement our talent and allow us to focus on the higher-yield things.”
AI Will Bump Up Against the System
Although this study demonstrates the potential for AI to improve diagnostic predictability, the widespread adoption of the technology will bump up against the realities of the American healthcare system, said Ian Gilchrist, MD, professor of medicine at the Milton S. Hershey Medical Center, Penn State Health, Hershey, Pennsylvania.
Ian Gilchrist, MD
“AI has some very interesting potential,” Gilchrist said. “We collect tons of data that could be integrated into these AI systems, but we don’t support the infrastructure. Other than local efforts or isolated efforts paid for by grants, the widespread application is still very unclear.”
Healthcare systems are selective with their capital spending on equipment, he said. “If we started to talk about adding this AI device that may have application, but we just need to use it for a while, that’s going to fall on deaf ears,” he explained “That’s always hurt medicine along the way. The interconnectivity that you really need for AI — you get a lot of information coming from different multiple sources — requires a system of interconnectivity and our machinery is not connected well.”
“There’s a lot of potential there. We just presently need to show how we’re going to save money doing that to offset the cost. And then maybe someone will invest in it,” he said.
