Artificial intelligence can now automatically detect signs of heart disease in lung cancer screening scans, potentially identifying millions of patients at risk for heart attacks before symptoms appear. Yet this promising capability faces a critical barrier: unclear payment structures and reporting gaps mean many patients never learn about their risk, even when AI flags it. The challenge reveals a deeper tension in clinical AI adoption: just because technology can detect disease doesn't mean the healthcare system is ready to act on it.

How Are Doctors Currently Missing Heart Disease in Routine Scans?

Every year, approximately 19 million patients receive chest CT scans for reasons unrelated to heart health, such as lung cancer screening or investigating a persistent cough . During these scans, coronary artery calcium, which appears as bright white spots in the heart tissue, often becomes visible to radiologists. This calcium buildup is a reliable marker of heart disease risk; the more calcium present, the higher the likelihood of a future heart attack.

The problem is straightforward but consequential: even though cardiologists can spot this calcium, many cases go unreported. An estimated 20 to 40 percent of incidental calcium findings are never documented in the patient's medical record . This means patients walk away from their scan with no knowledge of a significant health risk that could have prompted preventive treatment.

"We need to find more of these patients," said Ami Bhatt, chair of the Food and Drug Administration's Digital Health Advisory Committee and chief innovation officer of the American College of Cardiology.

Ami Bhatt, Chair of FDA's Digital Health Advisory Committee and Chief Innovation Officer of the American College of Cardiology

Heart disease remains the leading cause of death in the United States, making the missed opportunities particularly significant. Radiologists are human, and even experienced ones can overlook subtle findings when reviewing scans focused on a different clinical question, like lung nodules.

What Role Can AI Play in Catching These Cases?

This is where AI screening algorithms enter the picture. Machine learning models trained on thousands of CT images can be deployed to automatically scan every chest CT performed in a hospital or imaging center, flagging cases where coronary artery calcium is present. Because AI doesn't get fatigued and applies consistent criteria across every scan, it could theoretically catch many of those missed cases.

The technology represents what researchers call "opportunistic screening," meaning the system uses imaging already being done for other reasons to identify additional health risks. It's efficient in theory: no extra scans needed, no additional radiation exposure, just smarter analysis of data already collected.

The potential scale is enormous. If AI algorithms were deployed across the 19 million annual chest CTs performed in the United States, they could identify millions of patients with undetected coronary artery calcium, many of whom would benefit from preventive medications or lifestyle changes .

Why Payment and Workflow Uncertainty Threaten to Limit Impact?

Despite the clinical promise, a fundamental question remains unanswered: who pays for this AI-enabled screening, and how does it fit into existing healthcare workflows? This uncertainty is already shaping whether hospitals and imaging centers actually deploy these tools.

Several practical barriers complicate adoption:

• Billing Ambiguity: Radiologists and hospitals are unclear whether they can bill insurance companies for AI-assisted detection of incidental findings, or whether reimbursement codes even exist for this service.
• Follow-Up Care Costs: Identifying millions of new patients with coronary artery calcium could trigger a flood of follow-up appointments, imaging, and preventive medications, creating downstream costs that neither patients nor healthcare systems may be prepared to absorb.
• Liability and Responsibility: If an AI system flags coronary artery calcium but the radiologist misses it or fails to communicate it to the patient, questions arise about who bears responsibility for the missed diagnosis.

These aren't purely technical problems; they're economic and organizational ones. Healthcare systems must decide whether the long-term benefit of preventing heart attacks justifies the upfront investment in AI tools and the immediate costs of follow-up care for newly identified patients.

What Broader Lessons Does This Reveal About Clinical AI Adoption?

The coronary artery calcium screening case illustrates a pattern emerging across clinical AI: the technology often advances faster than the healthcare system's ability to integrate it. Algorithms can detect disease, but detection alone doesn't equal better patient outcomes if the system can't reliably act on the findings.

Similar challenges are appearing in other areas of AI-enabled diagnostics. For example, AI tools are being developed to detect minimal residual disease in cancer patients by analyzing whole-genome sequencing data across multiple time points, potentially identifying recurrence earlier than imaging alone . Yet these tools also require new workflows, new billing codes, and new conversations between pathologists, oncologists, and patients about what the results mean.

The broader healthcare AI landscape is shifting toward longitudinal, continuous monitoring rather than isolated snapshots of disease. Artificial intelligence is enabling laboratories to evaluate how disease biology changes over time, integrating molecular data with imaging and clinical information to create a more complete picture . But this shift requires infrastructure, payment models, and clinical workflows that don't yet exist at scale.

In liver disease, the FDA recently validated the first AI-enabled drug development tool, called AIM-NASH, which uses historical datasets to standardize clinical scoring of liver biopsy features . This represents progress in AI-assisted diagnosis, yet it also highlights how AI adoption often requires regulatory approval and clinical validation before widespread use, adding time and cost to implementation.

The path forward likely requires collaboration between technology developers, healthcare providers, payers, and regulators to establish clear payment models, liability frameworks, and clinical guidelines for AI-assisted screening. Without these structures, even highly accurate algorithms may remain underutilized, and patients may continue to miss opportunities for early intervention.