AI and heart health: Machines do a better job of reading ultrasounds than sonographers do, says study

Artificial intelligence (AI) could potentially do a better job of screening for heart health than trained sonographers. This is the finding of a study from the Smidt Heart Institute and the Division of Artificial Intelligence in Medicine at Cedars-Sinai in Los Angeles, California.

In the study, published in the journal Nature, a total of 3,495 heart echocardiograms (ultrasounds) were assessed.

Roughly half the scans were assessed by AI; the other half were assessed by 25 cardiac sonographers with an average of 14 years of experience. 

The scans and assessments were then sent to 10 cardiologists for review. 

CAN HEART FAILURE RISK SHOW UP ON A SMARTWATCH? IT’S POSSIBLE, STUDY SAYS

For the scans assessed by AI, cardiologists found fewer discrepancies, the results showed.

“Cardiologists were not able to distinguish between preliminary assessments by AI versus sonographers, and in fact, the AI assessments needed less correction,” cardiologist David Ouyang, principal investigator of the clinical trial and senior author of the study at Cedars Sinai, told Fox News Digital in an email.

“This showed the AI was both faster and more precise than the sonographers in assessing heart function,” he added. 

For the scans assessed by AI and then a cardiologist, there was a “substantial change in diagnosis” for only 16.8% of the images.

Among the scans assessed by sonographers, the change in diagnosis was 27.2%.

Team was ‘pleasantly surprised’

Dr. Ouyang said the team did not expect AI would perform better than trained sonographers.

AI TESTING OF BRAIN TUMORS CAN DETECT GENETIC CANCER MARKERS IN LESS THAN 90 SECONDS, STUDY FINDS

“This trial was designed as a non-inferiority trial, and we initially only hoped to show that AI and sonographers were equivalent, but were pleasantly surprised to show that AI was superior,” he said.

“In some ways, this AI passed the ‘Turing test’ for reading echocardiogram videos.”

(A Turing test is a means of gauging whether a computer can respond intelligently enough to be indistinguishable from human responses.)

Many variables involved in reading echocardiograms

Dr. Neerav G. Sheth, an interventional cardiologist at the Cardiology Consultants of Philadelphia in Paolio, Pennsylvania, was not involved in the study but offered his analysis of the findings.

There are many variables that go into reading echocardiograms for the heart, he said.

“These include the experience of the sonographer, the quality of the study, the quality of the machine and others,” he told Fox News Digital. 

“In terms of basic information and standardized assessments, I think that AI could potentially perform better than sonographers,” he went on. 

“AI requires oversight — especially if there aren’t that many ‘normal’ data sets analyzed by the algorithm.”

“Having said that,” he added, “as with sonographers, AI requires oversight — especially if there aren’t that many ‘normal’ data sets analyzed by the algorithm.”

‘Be cautious’ and ‘ask questions’

As the role of AI in health care is still fairly new and in flux, Dr. Ouyang stressed the need to be cautious about implementing it in patient care.

“This is one of the few prospective trials of AI in health care, and most of them are not blinded or randomized,” he said. 

“Be cautious about which AI tools to use, and ask questions about how many examples the AI was trained on and what evidence of performance is there.”

“Nothing will replace the physician-patient relationship.”

Once they are validated, Dr. Ouyang said he believes AI systems can greatly streamline and improve patient care. 

“We think more prospective randomized trials are needed, but this study shows that AI of this nature is ready for prime time and deployment into the clinical workflow,” he added. 

Limitations exist and more research is needed

The study authors noted several limitations of the research.

Perhaps the biggest limitation is that this was a “single-center study,” meaning it was performed on a small scale at a single location.

“AI can enhance patient care but should not replace human oversight.”

Additionally, the study’s demographics were limited.

“Having a small normal data set and a homogenous population reduces the ability of the AI to generalize across all people,” said Dr. Sheth.

To strengthen these findings, the researchers plan to implement the AI scanning tool at additional hospitals and will continue to monitor its accuracy.

AI ‘would help enhance, not replace, clinical practice’

Overall, both doctors agree that while AI can enhance patient care, it should not replace human oversight.

CLICK HERE TO SIGN UP FOR OUR HEALTH NEWSLETTER

“We very much need clinicians to still be in charge — cardiologists still need to review and confirm findings, even though the AI can make it faster and get more precise measurements than sonographers,” said Dr. Ouyang. 

For Dr. Sheth’s part, he said he believes AI will be critical in maintaining clinical practice as patients’ needs become more complex. 

“As new medications, technologies and treatment options emerge, proper use of AI would help augment, not replace, clinical practice to improve the outcomes for our patients,” he said.

Dr. Sheth added, “Nothing will replace the physician-patient relationship — but when properly used, AI can certainly augment it.”