AI ef­forts are still early, ‘not at the so­lu­tion level’

Health Data Management - - WASHINGTON REPORT -

While there has been a lot of pub­lic­ity about the po­ten­tial of AI in ra­di­ol­ogy to rec­og­nize im­ages and, when paired with al­go­rithms, as­sist in the di­ag­no­sis of dis­eases, this work re­mains in the re­search realm.

“We are not at the so­lu­tion level yet,” says Paul Chang, MD, pro­fes­sor and vice chair­man of ra­di­ol­ogy in­for­mat­ics at the Univer­sity of Chicago School of Medicine. “It is a long way be­tween ini­tial promis­ing re­sults in the lab and true clin­i­cally val­i­dated so­lu­tions.”

And de­spite the im­ple­men­ta­tions of ma­chine-learn­ing as­sisted al­go­rithms for spe­cific use cases in clin­i­cal op­er­a­tions, there are many stum­bling blocks to wide­spread adop­tion of ma­chine learn­ing and other AI tech­nolo­gies. There are com­pet­ing pri­or­i­ties for spend­ing IT bud­gets and myr­iad tech­ni­cal is­sues, such as the lack of re­li­able large data sets, so­phis­ti­cated data man­age­ment or work­flows that can in­cor­po­rate new in­sights.

Nonethe­less, health sys­tem ex­ec­u­tives be­lieve the knowl­edge gen­er­ated from AI tools will help their or­ga­ni­za­tions re­main fi­nan­cially vi­able as they as­sume fi­nan­cial risk for clin­i­cal out­comes. Ma­chine learn­ing, for ex­am­ple, can help de­rive pre­dic­tive and pre­scrip­tive in­for­ma­tion from data, which then can be em­bed­ded into elec­tronic health records, so that the in­sights are read­ily avail­able to clin­i­cians as they in­ter­act with pa­tients.

“The real bot­tom-line ben­e­fit of this tech­nol­ogy is un­der­stand­ing the busi­ness of medicine, so I can ac­tu­ally be­come more ef­fi­cient, and re­duce er­ror and re­duce vari­abil­ity,” Chang ex­plains.

Us­ing AI to draw con­clu­sions from med­i­cal in­for­ma­tion, such as that con­tained in elec­tronic med­i­cal records— say, as a clin­i­cian would—is more com­plex and much far­ther out on the hori­zon, says Lu­ciano Prevedello, MD, di­vi­sion chief in med­i­cal imag­ing in­for­mat­ics at Ohio State Univer­sity Wexner Med­i­cal Cen­ter.

“I can de­velop an al­go­rithm that helps me de­tect whether there is blood in the brain or not—that is very doable,” he says. “But to get to the rea­son this pa­tient is de­vel­op­ing the hem­or­rhage? That is a dif­fer­ent ques­tion and a dif­fer­ent level of com­plex­ity.”

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.