Artificial intelligence could substantially reduce the time cardiologists spend interpreting echocardiograms without compromising the quality of their reports. In a randomized study involving 1,200 heart ultrasound examinations, AI-generated preliminary interpretations performed comparably to reports prepared by healthcare professionals. Cardiologists also completed their reviews approximately 46% faster when assisted by AI. The findings suggest that automated reporting could help address growing pressure on cardiovascular services, although its impact on patient outcomes remains unproven.
Echocardiography is one of the most widely used diagnostic imaging techniques in cardiology. Ultrasound images provide information about the heart's structure, pumping function and potential abnormalities, helping physicians diagnose cardiovascular conditions and determine appropriate treatment.
Traditionally, a sonographer acquires the images before a cardiologist reviews them and prepares or finalizes a diagnostic report. The process requires specialist expertise and can be time-consuming, particularly in healthcare systems facing increasing demand for cardiac examinations. Researchers at Kaiser Permanente Northern California investigated whether AI could support this workflow by analysing echocardiographic images and generating preliminary reports for subsequent review by cardiologists.
Randomized comparison of AI and human reports
The study included 1,200 echocardiograms acquired during routine clinical care at hospitals and outpatient facilities across 21 Kaiser Permanente medical centres in Northern California. The examinations were performed over two weeks in March 2025. The patients had an average age of 69 years, and 47% were women. For each examination, an AI system, a sonographer and a cardiologist independently prepared preliminary interpretations. These reports were randomly assigned to reviewing cardiologists, who completed the final assessments without knowing whether the initial interpretation had been generated by AI or a healthcare professional. Nine sonographers and 20 cardiologists participated in the study.
The researchers assessed performance by measuring how frequently the reviewing cardiologists made substantial changes to the preliminary interpretations. They also recorded the time required to complete reports and examined consistency with previous findings documented in patients' medical records. According to the researchers, this was the first randomized trial evaluating an AI system capable of producing complete preliminary echocardiography reports, including measurements and clinical interpretations.
Comparable accuracy, substantially less time
The AI-generated reports met the study's predefined criteria for noninferiority compared with human-generated preliminary interpretations. Reviewing cardiologists made substantial changes to 50.6% of the AI reports, compared with 49.7% of reports initially prepared by sonographers or cardiologists. Although approximately half of the preliminary reports required meaningful adjustments, the frequency was similar across the AI and human groups. This suggests that AI could provide a comparable starting point for specialist interpretation, rather than independently producing definitive clinical reports. The most notable difference concerned efficiency.
Cardiologists spent an average of 4.6 minutes reviewing and finalizing AI-generated interpretations, compared with 8.5 minutes when completing reports without AI assistance. This represented a time reduction of approximately 46%. AI-assisted reports also showed greater consistency with historical reports in patients' medical records. Furthermore, reviewing cardiologists were unable to reliably distinguish AI-generated preliminary interpretations from those prepared by healthcare professionals.
Co-senior author Keane K. Lee, a cardiologist at Kaiser Permanente Santa Clara Homestead Medical Center, said the magnitude of the time savings exceeded the researchers' expectations. Given the large number of echocardiograms performed in clinical practice, reducing interpretation time by several minutes per examination could potentially free up substantial specialist capacity.
AI in cardiovascular diagnostics
The findings add to growing research into AI applications across cardiovascular diagnostics. As we previously reported, AI is also being investigated for identifying patients at increased risk of heart failure through electrocardiogram analysis. Unlike echocardiography, which uses ultrasound to assess cardiac structures and function, an electrocardiogram records the electrical activity of the heart. AI models can analyse ECG patterns to identify subtle abnormalities that may indicate underlying cardiac dysfunction or an elevated risk of developing heart failure.
These applications address different clinical questions. AI-assisted ECG analysis aims to support earlier identification of potential cardiovascular problems, while automated echocardiography reporting focuses on interpreting imaging findings and reducing the workload associated with diagnostic reporting. Together, they illustrate how AI could support different stages of cardiovascular assessment, from risk identification to specialist imaging. However, identifying potential abnormalities or generating preliminary reports does not automatically translate into improved patient outcomes. Both approaches require clinical validation and appropriate integration into established diagnostic pathways.
Clinical responsibility
Despite the promising results, the researchers emphasise that AI-generated interpretations cannot replace clinical expertise. Cardiologists remain responsible for reviewing the images, verifying the findings and approving the final report. They must also interpret the results in the context of each patient's medical history, symptoms and treatment needs. The study has several limitations. It evaluated previously acquired echocardiograms rather than the full diagnostic workflow during routine clinical operations. Cardiologists were also aware that they were participating in a study, potentially influencing their working practices.
Importantly, the research did not investigate whether AI-assisted reporting improves long-term patient outcomes, reduces diagnostic errors or shortens the time patients wait for treatment. Further studies are therefore needed to assess performance across larger and more diverse clinical environments. Nevertheless, the results suggest that AI-generated preliminary reports could offer a practical way to reduce the administrative and interpretative workload associated with cardiac imaging. With cardiovascular services facing increasing demand and projected shortages of specialist physicians, such applications could help healthcare organisations use available expertise more efficiently, provided that clinical oversight and diagnostic quality are maintained.
Add ICT&health on Google
Show more content from ICT&health in Google Search.
