Key takeaways
- A 2026 randomized trial in iScience found AI assistance cut echocardiogram review time by 24.1% across five cardiologists (Playford et al.).
- Automated measurement cut analysis and report creation time by 70% in a 2024 comparative study: 159 vs 325 seconds for measurement, 71 vs 429 for reports (Hirata et al., Journal of Echocardiography).
- On real lab days, AI assistance raised daily exam volume from 14.1 to 16.7 and lowered sonographer fatigue scores (AI-ECHO randomized trial, presented at AHA 2024; preprint).
- PROTEUS, the largest stress-echo AI trial (2,341 patients, 20 UK hospitals), improved reader consistency but did not beat standard care on referral outcomes.
Yes. In randomized trials, AI assistance cut echocardiography reporting time by roughly one-quarter to two-thirds, depending on which slice of the workflow was measured. Echocardiography reporting time is the full span from image acquisition through measurement, interpretation, and a signed report, and it is the bottleneck most echo labs name first. The honest answer is narrower than the marketing: the gains are real, they concentrate in measurement and draft reporting, and they disappear when image quality is poor. Keep that last point in mind; it comes back in every section below.
Does AI actually cut echo reporting time?
Yes, and the effect holds up across different study designs. A 2026 randomized crossover trial measured a 24.1% cut in review time. A 2024 comparative study measured a 70% cut in measurement and report creation time. A real-world lab trial measured shorter exams and more patients per day. The range looks wide because each study measured a different slice of the workflow, not because the results disagree.
Think of echo reporting as three jobs: measuring structures on the images, assembling those measurements into a draft report, and the cardiologist's interpretive review. AI is excellent at the first two and merely helpful at the third. Every credible time saving in the literature comes from automating measurement and drafting. Nobody has shown that AI shortens the thinking part, and you should distrust anyone who claims it does. The trials assume a PACS that feeds the AI clean DICOM data; that is what the MedIQ cloud PACS was designed to do.
What did the iScience randomized trial find?
Playford and colleagues randomized 200 echocardiograms across five cardiologists, each reading with and without AI assistance, and published the results in iScience in October 2026. AI cut review time by 24.1% overall, with every cardiologist improving somewhere between 11.8% and 32.0%. The AI, working from measurement data alone with no image analysis, correctly identified all 100 independently adjudicated cases of severe aortic stenosis.
Two details matter more than the headline number. First, reader concordance improved: with AI, cardiologists agreed with each other more and referred more uniformly for further investigation. Second, and less flattering, cardiologists were frequently reluctant to follow the AI's advice even when it was right. The trial's own conclusion is blunt: accuracy gains stalled because humans did not trust the machine. That trust gap is the real implementation problem, and no software update fixes it. It gets fixed with training, transparent AI outputs, and a review workflow that shows clinicians what the AI saw.
There is a third detail worth your attention: the AI in this trial worked from measurement data alone. No image analysis, no fancy vision model, just automated numbers fed to the reporting cardiologist. And it still cut review time by nearly a quarter while catching every severe case. That tells you where the current value sits. You do not need the most advanced model on the market to get the time savings. You need reliable automated measurements presented clearly at the point of review. This trial is one example of a wider shift; our overview of AI in cardiac imaging covers acquisition, triage, and reporting together.
How much time does automated measurement save?
In Hirata and colleagues' 2024 study, 23 patients were analyzed both manually and with the US2.ai automated platform (Journal of Echocardiography, 2024). Measurement time fell from 325 seconds to 159. Report creation fell from 429 seconds to 71. Total: a 70% reduction in analysis and reporting time, with statistical significance on both halves.
| Task | Manual | With automated measurement (US2.ai) |
|---|---|---|
| Measurement | 325 seconds | 159 seconds |
| Report creation | 429 seconds | 71 seconds |
This is the cleanest number in the literature because the comparison is so direct: same studies, same measurements, stopwatch on both methods. The software processed 2D and Doppler data in real time and produced guideline-aligned draft reports with findings and comments. One caveat from the authors: in cases with fair or poor image quality, the AI needed more corrections and measurement took longer. Good acquisition remains the price of admission, a theme we will return to.
The accuracy question has its own evidence base. A multicenter validation of 602 echocardiographic exams found that variability between the AI and human readers was lower than variability between expert humans themselves. The platform studied, Us2.ai, is CE-marked and vendor-neutral: it takes standard DICOM input, quantifies 68 guideline-recommended parameters, and typically produces a structured draft report within two minutes of upload. Erasmus Medical Center Rotterdam and Wilhelmina Ziekenhuis Assen have deployed it in routine practice, and Fujifilm has embedded the same technology in its LISENDO ultrasound systems. In other words, this is not lab-only software. It is running in real hospitals, on real schedules. Those draft reports only help if they land in structured reporting templates your cardiologists will actually sign.
What changes on a real lab day with AI?
The AI-ECHO randomized crossover trial put the question to a working lab: 585 patients scanned over 38 study days, with AI assistance randomly assigned per day (Kagiyama et al., presented at the American Heart Association Scientific Sessions 2024; published as a preprint, not yet peer-reviewed). On AI days, exams averaged 13.0 minutes versus 14.3 manual. Daily volume rose from 14.1 to 16.7 exams. Sonographers reported lower mental fatigue scores. And the AI measured 3.4 times more parameters per exam than the manual workflow.
That last number deserves attention. AI did not just do the same work faster; it did more work: 85 parameters per exam versus 25. Ninety percent of AI-generated measurements landed within acceptable clinical limits of the final expert-endorsed values. So the lab got faster, fuller studies with less tired sonographers. The limitation is the usual one for preprints: treat the effect sizes as promising rather than settled until peer review lands. The platform underneath matters as much as the algorithm: a cardiology PACS built for echo workflows keeps measurements, priors, and reports in one record.
Does AI help with stress echo too?
The evidence here is thinner, and more honest than the marketing. PROTEUS randomized 2,341 patients with suspected coronary artery disease across 20 UK hospitals to standard stress echocardiography or AI-assisted interpretation using EchoGo Pro (Woodward et al., American Heart Journal, 2023; trial registered as NCT05028179). On the primary endpoint, appropriate referral for invasive angiography, AI did not beat standard care. The trial was underpowered for that endpoint, with only 85 patients referred, but the result is the result.
What PROTEUS did show is arguably more useful for a lab director. AI support improved decision-making consistency, most clearly for less-experienced clinicians and complex cases. It leveled the field between readers. An embedded economic analysis found the intervention cost-effective at an AI cost of up to 35.35 pounds per case (Chauhan et al., Value in Health Regional Issues, 2026). And the trial surfaced the image-quality problem at scale: the AI rejected 26.1% of scans for insufficient quality.
My read: PROTEUS is the trial that keeps everyone honest. It says AI stress echo is safe, consistent, and economically sane, but not a magic upgrade on hard outcomes. If a vendor shows you this trial as proof of superiority, ask them to show you the primary endpoint table. Stress echo is one piece of a larger automation story; see the seven ways practices are automating cardiovascular imaging workflows.
Where does AI fall short?
Three places, and you should budget for all of them. First, image quality. Every study in this article shows the same pattern: good images, big savings; poor images, manual corrections that eat the savings. PROTEUS rejected more than a quarter of scans outright. Your acquisition quality program is not separate from your AI program. It is your AI program.
Second, trust and workflow. The iScience cardiologists ignored correct AI advice. Sonographers in early rollouts double-check everything, which temporarily makes things slower, not faster. Plan for a 4 to 8 week trust-building period where measured throughput dips before it climbs. The labs that get through this phase fastest do two things: they show staff the AI's work (measurements visible alongside images, not a black-box verdict), and they let skeptics keep their old workflow in parallel until the numbers convince them. Forcing adoption before trust exists just creates quiet workarounds.
Third, scope. Current validated tools cover adult transthoracic echocardiography. Pediatric echo, transesophageal echo, and congenital heart disease are outside the validated scope of platforms like US2.ai. Do not let a vendor demo on clean adult outpatients set expectations for your whole lab.
What should your lab do first?
Start with measurement automation, not full report automation. It has the strongest trial evidence, the clearest return, and the smallest disruption to how your cardiologists work. Here is the sequence that matches what the trials actually tested:
- Baseline your current state. Time ten studies end to end: acquisition, measurement, drafting, sign-off. You cannot prove savings without a before number.
- Pilot on your highest-volume study type, usually routine adult transthoracic echo. That is where the validated tools live.
- Fix acquisition quality first. Standardize views and train sonographers on what the AI needs. This is the step everyone skips and the trials say matters most.
- Keep a human-in-the-loop review protocol. Every AI draft gets cardiologist review and sign-off, full stop.
- Measure at 30 and 90 days: time per study, studies per day, correction rate, and sonographer fatigue. Expand only what the numbers support.
If you want the implementation detail behind this evidence, our companion guide walks through automating your echo reporting workflow step by step, from RIS integration to structured templates. And keep one number taped to your monitor during the pilot: correction rate. It tells you faster than any dashboard whether the AI is earning trust or just generating rework.
Frequently asked questions
Between 24% and 70%, depending on what you measure. The iScience randomized trial (2026) found a 24.1% cut in cardiologist review time. Hirata et al. (2024) found a 70% cut in measurement and report creation time with US2.ai. In a real lab, the AI-ECHO trial found exams about 9% shorter but daily volume up 18%. Quote the slice, not just the biggest number.
The validation data says yes for adult transthoracic echo. A 602-exam multicenter validation found AI-to-human variability was lower than expert inter-reader variability. In the AI-ECHO trial, 90% of AI measurements fell within acceptable clinical limits of expert-endorsed values. Accuracy is not the weak link; image quality and clinician trust are.
No, and no credible trial suggests it should. Every study in this article used a human-in-the-loop design: AI measures and drafts, the cardiologist reviews, edits, and signs. In the iScience trial, the interesting finding was not that AI beat cardiologists on classification, it did, but that cardiologists often did not act on correct AI advice. The workflow that works is AI assistance with accountable human sign-off.
PROTEUS was a prospective randomized trial of AI-assisted stress echocardiography across 20 UK hospitals, enrolling 2,341 patients with suspected coronary artery disease (Woodward et al., American Heart Journal, 2023). Patients received standard stress echo or AI-assisted interpretation with EchoGo Pro. AI did not beat standard care on the primary referral endpoint, but it improved reader consistency, especially for less-experienced clinicians, and an economic analysis found it cost-effective.
High-volume labs with standardized acquisition protocols see the biggest gains, because the trials show savings concentrate in repetitive measurement. Labs with less-experienced readers gain disproportionately on consistency, per PROTEUS. Labs struggling with image quality should fix acquisition first; AI amplifies good input and chokes on bad input. If your lab does mostly complex congenital or transesophageal work, current tools are not validated for you yet.


