In 2019, Mayo Clinic researchers published a study in Nature Medicine showing that an AI model could detect asymptomatic left ventricular dysfunction, a serious heart condition, from a standard 12-lead ECG that looked entirely normal to a trained eye. Tested against nearly 53,000 independent patients, the model achieved an AUC of 0.93 with 86.3% sensitivity. Patients it flagged as positive were four times more likely to develop the condition later. This was not a promising demo. It was a validated, FDA-adjacent capability that a human reading the same tracing could not reliably match.

Cardiologists are rated Moderate risk, with a horizon toward 2046, and this profession's story is different from the ones covered in AI Doomsday's coverage of anesthesiologists and nurses. Those pieces found working technology that the market or the institution ultimately resisted, or a burnout crisis AI was recruited to soften. Cardiology is the case in this series where the diagnostic AI genuinely, repeatedly, matches or beats specialist accuracy, in products that are FDA-cleared and already deployed.

And the profession is still short thousands of doctors, with wait times getting longer, not shorter. The AI solved a harder diagnostic problem than most people expected it could. It did not solve the problem of not having enough cardiologists.

Key Points

  • Cardiologists are rated Moderate risk with a horizon toward 2046, one of the longer healthcare horizons in this series, because the profession's diagnostic AI genuinely matches specialist accuracy while its procedural and judgment work does not automate.
  • FDA-cleared AI tools now read ECGs and echocardiograms at or above specialist accuracy: Mayo Clinic's AI-ECG model reached an AUC of 0.93 for ventricular dysfunction, and HeartFlow's non-invasive coronary imaging reached 86% diagnostic accuracy versus 65% for standard invasive angiography.
  • The American College of Cardiology projects a shortage of roughly 3,010 cardiologists by 2026, with 62% of U.S. counties having no cardiologist at all and demand projected to grow 15.7% against a 2% supply increase through 2037.
  • AI diagnostic tools are reaching population scale through consumer devices, Apple Watch and AliveCor's Kardia among them, which increases the volume of flagged patients needing a cardiologist's judgment rather than reducing it.
  • Interventional procedures, catheter placement, stenting, remain dependent on human tactile feedback and manual dexterity that AI-assisted imaging supports but does not replace.

What a Cardiologist Actually Does

The role spans diagnosis, reading ECGs, echocardiograms, and advanced imaging to identify heart and vascular disease, ongoing management of chronic conditions like heart failure and arrhythmia, and for interventional cardiologists, physically performing procedures like catheterization, angioplasty, and stent placement. It requires synthesizing imaging data, patient history, and comorbidities into a treatment plan, and communicating risk and prognosis to patients making decisions about their own care.

What AI Is Already Doing

The ECG interpretation results extend well beyond the original Mayo study. A follow-up prospective trial validated similar detection accuracy, an AUC of 0.885, using ECGs captured on consumer smartwatches rather than clinical equipment. iRhythm's Zio monitor carries an FDA-cleared arrhythmia-classification algorithm performing comparably to cardiologist review, and AliveCor's newest Kardia 12L system brings full 12-lead ECG capability, paired with AI interpretation across 35 distinct cardiac determinations, to a handheld consumer device.

Imaging analysis has moved just as far. Ultromics' EchoGo Pro, FDA- cleared and deployed across the NHS and at Mayo Clinic, automates detection of coronary artery disease from echocardiograms, addressing what the company estimates is a one-in-five rate of underdiagnosis in standard review. Its companion EchoGo Heart Failure tool, an FDA Breakthrough Device, detects heart failure with preserved ejection fraction from routine echo images. On the interventional side, HeartFlow's non-invasive FFR-CT technology reached 86% diagnostic accuracy against invasive angiography's 65% in pivotal trials, was added to American College of Cardiology chest pain guidelines in 2021, and now carries its own reimbursement code. CathWorks' comparable FFRangio technology, deriving flow measurements from standard angiography images at reported accuracy above 90%, was valuable enough that Medtronic acquired the company for $585 million.

THE SCREENING PARADOX

AI-enabled ECG interpretation is now reaching population scale through consumer hardware: roughly a billion Apple Watches capable of on-device ECG, plus AliveCor's handheld devices, are screening people who would never otherwise see a cardiologist for anything. That expands, rather than shrinks, the number of patients a cardiologist needs to evaluate, the same inversion AI Doomsday has documented in plumbing and nursing: a detection tool that works well generates more referrals, not fewer specialists.

The Shortage AI Can't Solve

The American College of Cardiology's 2025 workforce analysis projects a shortfall of roughly 3,010 cardiologists by 2026, with 62% of the country's counties having no cardiologist at all, a gap associated with measurably higher mortality in those areas. A separate supply-and- demand model projects demand growing 15.7% between 2025 and 2037 while supply grows only about 2%, pushing overall workforce adequacy from 92.2% down to roughly 81.4%, with rural and non-metropolitan areas falling far further, to under 30% adequacy in the worst-projected states. None of the diagnostic AI reviewed here adds a single practicing cardiologist to an underserved county. It makes the cardiologists who do exist more efficient at the part of the job that was never the bottleneck.

Where AI Still Can't Go

Interventional cardiology remains stubbornly manual. A systematic review of AI in interventional practice describes the field as still in its infancy specifically because catheter navigation through complex or unusual vascular anatomy depends on tactile feedback current robotic systems don't reliably replicate, alongside real dexterity in physically placing a stent. AI's documented limitations compound this: black-box decision-making that clinicians can't audit, performance that degrades on patient populations underrepresented in training data, and a persistent inability to weigh personalized risk and benefit for patients with multiple, interacting conditions the way a treating physician does. Explaining a prognosis to a frightened patient and helping them decide between treatment paths is not a task any of these systems attempt.

The Labor Market Reality

The U.S. Bureau of Labor Statistics projects roughly 5% employment growth for cardiologists, faster than the average for physicians overall, with about 800 openings a year against a base of approximately 16,400 practicing cardiologists. Compensation reflects the same demand signal rather than automation anxiety: Medscape's 2026 report put average cardiologist compensation at $575,000, up 10% year over year, the fastest growth of any specialty the survey tracked. Nothing in the labor data resembles a profession being displaced.

How to Use AI as a Cardiologist Now

For screening and triage: AI-ECG and echo tools are reliable enough to trust as a first read, especially at population scale. The added value is deciding what to do with a flagged result, not re-verifying that the algorithm read the tracing correctly.

For imaging-guided procedures: tools like HeartFlow and CathWorks reduce the need for invasive diagnostic catheterization before treatment decisions. Use them to focus procedural time on patients who actually need intervention.

For workforce planning: given the documented shortage, the practical lever is coverage, using AI-enabled remote monitoring and consumer device screening to extend a limited number of cardiologists across more patients, particularly in the underserved counties the ACC data identifies.

What I Think

The 2046 horizon looks generous rather than tight, and that's a genuine compliment to the technology. Cardiology is the strongest evidence in this entire series that AI diagnostic tools can reach and exceed specialist accuracy on well-defined tasks. What keeps this from translating into job losses is that the well-defined diagnostic task was never the actual constraint on cardiac care. The constraint is having enough trained physicians to see patients, perform procedures, and make judgment calls in counties that currently have none, and no AI model reads an ECG for a patient who never gets to see anyone.

What I find most significant here is the contrast with the rest of this series. In most professions, AI's limits are what protect the job. In cardiology, the AI's demonstrated competence is what's driving adoption, and the job is protected instead by a shortage of the humans needed to act on what the AI finds. That's a healthier position for a profession to be in than most.

"The algorithm can read a heart's electrical signature better than most humans now. It still can't be the person who explains what that means to the patient sitting across from them, or the one who isn't there because the county doesn't have one."