"The critical part of a stethoscope," seasoned cardiologists advised as I pondered this medical school investment, "is between the eartips."

Artificial intelligence (AI) is maturing at a time when even the most dedicated cardiologists' patient care bandwidth and analytical capability "between eartips" have both hit critical bottlenecks. Beyond quiet growth in back-end applications like logistics, AI technologies are expanding on the front-end: testing, patient encounters and procedures.

AI in Cardiac Testing

Labs. Electrocardiograms. Imaging. Modern cardiology runs on data but faces accessibility and interpretability challenges. For example, cardiac MRI readers' absences could delay reports. Some of my pregnant (and also non-pregnant) patients had been previously misdiagnosed with cardiomyopathy due to outside echocardiogram quality and readers unfamiliar with expected changes in pregnancy. A randomized-controlled trial from the research group of Cedars-Sinai statistician-echocardiologist David Ouyang, MD, FACC, found AI better than sonographers at assessing left ventricular ejection fraction¹. Their paper, "Blinded, randomized trial of sonographer versus AI cardiac function assessment," notes, "AI tools can improve efficacy as well as efficiency in assessing cardiac function."

Technologies like Viz.ai evaluate pathologies such as strokes. Vascular surgeon and ultrasound company ButterflyNet co-founder, John D. Martin, MD, also noted "the power of normal,"² pertinent negative results for screening. Conversely, Jacob Agris, MD, PhD, interventional neuroradiologist and chief medical officer of coronary analysis company Artrya, argued that AI training sets "contain the knowledge of all the more than 80,000 cardiologists globally" for specific conditions, so AI-powered computed tomography algorithms can soon "assign the risk on an MI long before it occurs in order to prevent MIs."³ AI can recognize patterns that lack the physiologic explanations humans rely on and synthesize multiple data sources.

AI in Patient Encounters

Clinicians spend more time on electronic health records (EHRs) than direct patient care across settings. AI might help. Cleveland Clinic cardiologist Ashish Sarraju, MD, and his group showed that preventive cardiologists rated 84% of large language model ChatGPT's responses to preventive cardiology questions as appropriate.⁴ University of California San Diego behavioral scientist John W. Ayers, PhD, MA, found that health care professionals rated ChatGPT responses to online forum medical questions higher-quality and more empathetic than volunteer physician responses.⁵

Boston University endodontist and digital health entrepreneur Suneel Kandru, CAGS, MSD, Cert-Ed, explained, "AI plugins for EHRs can allow us to focus on patient care again."⁶ As in testing, patient care may benefit from AI synthesizing key trends across data sources. He warned that despite confident AI responses, no current systems provide references to literature or specific patient data. Easier applications include assisting screening, data gathering or data entry to optimize patient visits or prompt presentation. Indeed, patient engagement or even recruitment could begin even in the community. Mayo Clinic Chair of Cardiovascular Medicine Paul A. Friedman, MD, FACC, group built an IOS app to collect 125,610 Apple Watch single-lead, distal-limb electrocardiographic tracings from 2,454 patients in 46 U.S. states and 11 countries, from which 421 patients had an echocardiogram. In those 421 patients, an AI algorithm predicted left ventricular ejection fraction of 40% or below within 30 days with area under the curve of 0.885.⁷ User interface and ethical follow-up capability are critical.

AI in Procedures

Cardiac procedures entail significant complexity and time constraints. Pediatric electrophysiologist and SentiAR co-founder Jennifer N. Avari Silva, MD, describes three tiers of AI involvement:⁸

1. AI-assisted reporting and editing can "give us computational power in a hands-free way. In the moment, I know exactly what I want to tell my future self if I need to come back in the future to do another procedure."
2. AI-supported, intraprocedural data management could predict what data to display at a given moment or help annotate. Dr. Silva cautioned that, depending on design optimization and training, such technologies may or may not reduce cognitive burden, but can help improve spatial and situational awareness.
3. Decision support. Regulators and cardiologists would likely resist autonomous systems that apply radiofrequency or reposition C-arms, but AI software to assist measurements and help to visualize transcatheter valves in situ or simulated ablation lines are already available commercially.

"Building trust in AI is key," Silva concludes. "My personal bias is that we should never have a black box module be integrated into any system: we need to have some intuitive sense of how algorithms we work with behave and be able to override at any time."

Natural Intelligence

The critical part of a stethoscope – and test, clinic, procedure – is the cardiologist that translates it into patient care. Like any such technology paradigm – sterility, anesthesia, EHRs  – AI needs careful management of growth and integration. Ultimately, AI could augment the natural intelligence "between the eartips."

References:

1. He, Kwan AC, Cho JH, et al. Blinded, randomized trial of sonographer versus AI cardiac function assessment. Nature. 2023;616: 520-524.
2. Personal communication
3. Personal communication
4. Sarraju A, Bruemmer D, Van Iterson E, et al. Appropriateness of cardiovascular disease prevention recommendations obtained from a popular online chat-based artificial intelligence model. JAMA. 2023;329(10):842-844.
5. Ayers JW, Poliak A, Dredze M, et al. Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Intern Med. 2023;183(6):589-596.
6. Personal communication
7. Attia ZI, Harmon DM, Dugan J, et al. Prospective evaluation of smartwatch-enabled detection of left ventricular dysfunction. Nat Med. 2022;28: 2497-2503.
8. Personal communication