Year in Review: Evidence Around AI in Cardiology Grows

Artificial intelligence (AI), as in the world more broadly, continued to advance across cardiology over the course of 2025, with research and guidance from professional societies grappling with how the technology is going to help clinicians in their day-to-day practice.

TCTMD spoke with Thomas Maddox, MD (Washington University School of Medicine, St. Louis, MO), who serves on the US Food and Drug Administration’s digital health advisory committee, to get his thoughts on the biggest AI-related news from the past year and what the field can expect moving forward.

“It’s been a pretty good year for AI research. I think we’re starting to get more of an evidence base,” he said. “Because at first it was just pretty hype-y. And as with everything in medicine, we do need the evidence before we can responsibly recommend it as part of our care.”

AI in Imaging

Maddox said there has been mounting evidence around the use of AI in imaging. “Our radiology colleagues have been first over the hill in terms of incorporating AI into their work, and now I think we’re starting to see it in cardiology,” he commented.

He pointed to research showing that the EchoNet AI algorithm applied to echocardiographic measurements can automate much of the process around image acquisition, image interpretation, and report generation, increasing accuracy and saving physicians time. Another study showed that the AI-powered PanEcho system automatically interprets transthoracic echocardiograms (TTEs) with high accuracy. EchoPrime, the largest AI-echo model, has shown promise for increasing workflow efficiency, with additional clinical evaluation underway.

“Guess I’ll look for something else to do,” Maddox joked about his career path. He said it’s probably a stretch to envision that these AI-echo models will be able to perform autonomously, but they will provide significant assistance. “More people can get access to both the image acquisition and its interpretation in a pretty quick fashion, and I think it’ll provide a level of standardization around accuracy of diagnosis that we don’t currently have with some of the variability that comes with human reads.”

There also has been continuing development of AI-related systems involving coronary CT angiography and fractional flow reserve, he noted.

Enhancing ECG Interpretation

The evidence base around the use of AI in ECG interpretation has grown over the past year as well, with Maddox pointing to a US registry study showing that the Queen of Hearts platform (PMcardio) outperformed standard decision-making when it came to detecting angiographically confirmed STEMI across three PCI networks. FDA approval for the Queen of Hearts AI-ECG model is expected in the first quarter of 2026.

“I’ve started to hear more and more uptake of that in practice and people using that as a validity check on their own interpretation of incoming MI EKGs,” Maddox said.

Earlier in 2025, another AI tool applied to ECG data outperformed expert clinicians for detecting type 1 MI, with similar accuracy compared with high-sensitivity troponin testing.

Our radiology colleagues have been first over the hill in terms of incorporating AI into their work. Thomas Maddox

This part year also saw studies evaluating the use of smartwatches with ECG functionality to detect subclinical structural heart disease, including one presented at the American Heart Association (AHA) 2025 Scientific Sessions. In that study, an AI algorithm applied to single-lead ECGs identified structural heart diseases with high sensitivity (86%) and a high negative predictive value (99%).

In addition, research demonstrated the potential for AI to enhance care in resource-limited parts of the world. A study conducted in Kenya, for example, showed that an AI-ECG algorithm performed well for identifying patients with LV systolic dysfunction, using echo as the reference.

Many Potential Applications and Much Uncertainty

TCTMD covered a number of AI-related stories throughout 2025, touching on topics like AI-guided catheter ablation for persistent atrial fibrillation (TAILORED-AF), glucose control after cardiac surgery, preventive care for patients at risk for atherosclerotic cardiovascular disease, CV pharmacology, AI-supported coaching to improve glycemic control in patients with type 2 diabetes, piggybacking on mammograms to screen women for CV risk, adjudication of clinical events in trials, and digital stethoscopes that detect low ejection fractions in areas with limited resources.

Though the wide range of topics covered by these stories indicates AI’s broad potential in cardiology, there is still uncertainty about how it will play out. In November, the AHA issued a scientific advisory outlining pragmatic approaches for evaluating and monitoring AI in healthcare in the face of rapid development and adoption.

And there might be some skepticism when it comes to relying too heavily on AI, with one study published in September showing that even though clinicians saw value in generative AI, they perceived their colleagues who used the technology as less skilled.

That skepticism “is going to be driven a little bit by people’s familiarity and just overall philosophy around technology and its role in clinical practice,” Maddox said, adding that younger physicians are questioning the impact of AI on their career prospects. “I think there’s more than enough work for a cardiologist. I tell them not to worry too much about that. But I also say to them, ‘If you’re not getting familiar with and taking full advantage of these tools, I think you’re going to be a step behind and it’s going to be a problem.’”

There are questions about how the emergence of AI in clinical practice will affect training programs, Maddox said. “I’m mid-career, so I trained without these tools and now I have a fund of knowledge and experience and frameworks about how this field works to where I know how to put the tool into context and the information it gives me,” he said. “But if I don’t yet have those frameworks or the experience, I’m just in my training, there may be a tendency to over-rely on [these AI tools].”

AI is going to be part of practice, Maddox said, “but you also need to make sure you don’t shortchange the knowledge that you need to be an effective cardiologist.”

Moving forward, AI is likely to take on more of a role for ambient notetaking during physician-patient discussions to ease the burden around clinical documentation, for summarizing patient data, and—perhaps with agentic AI—for gathering information from patients ahead of an appointment, Maddox said.

“I just wonder about ways to start to configure these [large language models] so they can help with that information gathering and synthesis that really is core to a lot of our interaction with the healthcare system,” he said.

Issues involving security, privacy, and regulation related to AI still need to be worked out, according to Maddox, who noted that these algorithms are particularly challenging to regulate since they’re constantly changing and the investment needed to do so is massive.

“It’s a little tough to know how we underwrite that without bankrupting the system, but it is going to be something we need to have a coherent approach to,” Maddox said.