A new form of AI suggests the best diagnostic strategies

AI has, for some time, been deployed to find out about medical conditions in specific areas. It can build knowledge on specific topics to weave into details such as the shape of a tumor that indicates breast cancer or neural cells that indicate breast cancer.

While AI is excellent when trained on years of human data in specific areas, it has not been able to handle the large number of diagnostic tests (around 5000) and disorders (around 14,000) of use. modern day clinical. Now, a new algorithm developed by engineers at the USC Viterbi School of Engineering can think and learn not just as a doctor but with unlimited experience.

The work comes from the laboratory of Gerald Loeb, professor of biochemical engineering, pharmacy and neurology at the USC Viterbi School of Engineering and a trained physician. Loeb spent years applying AI algorithms to hackers and building robots to detect and recognize products and materials.

His previous research on this transcended modern conditions. While the state of AI for hackers to identify around 10 items with about 80 percent certainty, Loeb and Jeremy Fishel, his graduate student at the time, were able to identify 117 items with accuracy 95 per cent.

When they expanded it to 500 objects and 15 different possible tests, their algorithm got even faster and more accurate. That is, Loeb says, when he started thinking about changing it for medical diagnosis.

Loeb’s new form of AI proposes the best diagnostic strategies for mining electronic healthcare records in databases. This could lead to faster, better and more effective diagnoses and treatments. The work was published in the Journal of Biomedical Informatics.

The algorithm works just like a doctor- “thinking about what to do next at every stage of the medical work,” said Loeb, a pioneer in the field of neural prosthetics and one of the original developers of the cochlear implant, who -Now widely used for the treatment of hearing loss. “The difference is that it benefits from all the experiences in the general health care records.”

How it works

Conventional AI has long used a special algorithm to recommend to physicians the most likely studies provided by a set of observations. Called Bayesian Inference, it uses whatever information is currently available to determine which studies are most likely.

The Loeb algorithm negates this process and instead searches for those tests that would be likely to identify the correct illness or condition, no matter how vague. He names this Bayesian Exploration. The algorithm can also take into account the costs and delays associated with various diagnostic tests.

“This has never been done before,” he said. “This is new.”

Loeb said its new algorithm has several advantages.First, this algorithm could help doctors to make better diagnostic and testing decisions by suggesting a number of good options, including some a practitioner would not have considered otherwise. Next, the diagnostic software would automatically update and improve, as several physicians enter additional data into electronic medical records.

In addition, Loeb believes that doctors would be easier to generate complete and accurate medical records. Instead of hunting for codes or working their way through multiple menus, they could simply choose a specific illness or diagnostic method recommended by the AI, which would enter the correct information itself. -mobile in the electronic records.

Loeb emphasizes that physicians could, of course, go beyond the AI ​​and go with their own diagnosis.

The algorithm is not for making decisions for doctors or to replace them. He intends to support them and support them. “

Gerald Loeb, Train P.hysician and Professor, Biomedical Engineering, Pharmacy and Neurology, Viterbi School of Engineering, University of Southern California

Looking to the future

Loeb believes that this algorithm could reshape medical and test morphology. But USC professor Viterbi and the Keck School of Medicine acknowledge that the major financial and technological challenges in implementing AI in electronic health records.

The balkanized medical system of the United States and the spotty use of electronic medical records he believes makes it an unstable environment for his technology to be rooted.

Loeb says it would be easier to introduce its system in other countries, for example in Scandinavia or South Korea, places with national health care and widespread use of electronic medical records.

However, its implementation would be a major challenge even there, given the high cost and brain power required to develop and deploy the large database and user interface for acceptance and widespread integration of its algorithm.

Instead, Loeb puts his faith in tech. He believes that Amazon, Microsoft, and Google have the resources and expertise to block American healthcare as Uber and Lyft passed on the taxicab business.

“If the promise of success is big enough, then people are going to be motivated to do it,” Loeb said. “And that’s what we think this algorithm delivers: the potential , the promise of offering a solution to a major and resource – intensive problem, worth trillions of dollars. “