An AI System Performs Better On Key Heart Tests Than Humans

During the most rigorous trial ever conducted on the subject concerned with artificial intelligence and cardiology, it was discovered that artificial intelligence is better than humans at evaluating heart ultrasounds, the main test of overall cardiac health.

‍

This US study is the first blinded, randomized clinical trial for heart health that has been conducted, and it was conducted in order to evaluate the efficacy of AI models using medical scans, despite previous research that has demonstrated the potential power of AI models for reading medical scans.

‍

In an interview with AFP, the study's senior author David Ouyang told the news agency that AI is getting a lot of attention, but rigorous assessment remains critical, he also explained.

‍

The cardiologist at Cedars-Sinai Medical Center in Los Angeles believes that this successful trial helps demonstrate that we are now ready to go into prime time," he said.

‍

It is a procedure in which a patient is subjected to a heart ultrasound, also called an echocardiogram, during which a sonographer conducts the examination while a cardiologist assesses the scan.

‍

An AI model was compared with sonographers in a new study reported in the journal Nature to see which would give the most accurate estimate of the severity of the disease at the beginning.

‍

The two ultrasounds were used to assess what is known as left ventricular ejection fraction. This is a measurement of how well the heart can pump blood out to the body within the heartbeat, which enables blood to circulate.

‍

This test is used in order to determine whether a patient has suffered a heart attack or whether or not they will be able to undergo serious treatment like implanting a defibrillator as a result of such a heart attack. This test measures how well a heart is functioning.

‍

There were nearly 3,500 heart ultrasounds performed by sonographers and AI models randomised for this study and all were divided equally between the two.

‍

The cardiologists who evaluated these assessments didn't know the difference between the ones which had been made by humans and the ones that had been made by artificial intelligence.

‍

There is a significant difference between the cardiologists' assessment of a sonographer's performance and that of the AI model in nearly 17 percent of the cases in the study, using information gathered from the cardiologists' assessment.

‍

Ouyang noted that the AI was faster, more accurate, and indistinguishable to the cardiologists because of its speed.

‍

The United States and the rest of the world are currently facing a "tremendous shortage" of sonographers, and this would save them a great deal of time, he added.

‍

In order to process large amounts of data, the AI model, called EchoNet-Dynamic, uses a deep learning technique to learn from nearly 145,000 echocardiographies downloaded from a database.

‍

This method is currently being applied for approval by the Federal Drug Administration, and Ouyang, the lead scientist, hopes that they will have the opportunity to apply for approval elsewhere as well, including in the European Union.

‍

A cardiac specialist at the Mayo Clinic in the United States, Patricia Pellikka, told AFP that she saw the study as very promising and said that using AI tools together with traditional methods will help to improve efficiency and standardization as well as save time.

‍

Claude Zores, a French cardiologist who directs the institute for Cardiology and Ultrasound Research in Paris, said that the study was well conducted, but the technology might not be as useful in France, where cardiologists are responsible for assessing the health of the heart.