AI beats doc­tors in spot­ting breast can­cer

What takes a pathol­o­gist hours, ma­chines do in a wink.

Cosmos - - Digest -

It’s a re­sult that may fur­ther jan­gle the nerves of doc­tors al­ready skit­tish in the face of ma­chine medi­cos. Breast can­cer is the lat­est dis­ease that ar­ti­fi­cial in­tel­li­gence (AI) can di­ag­nose bet­ter than hu­mans, ac­cord­ing to a re­cent study in the Jour­nal of the Amer­i­can Med­i­cal Association.

Led by Babak Eht­e­shami Be­jnordi at Rad­boud Univer­sity Med­i­cal Cen­tre in the Nether­lands, the study pit­ted dif­fer­ent ma­chine-learn­ing al­go­rithms against 11 pathol­o­gists in analysing 129 biop­sies.

While the pathol­o­gists had years of ex­pe­ri­ence, the al­go­rithms were trained with just 270 dig­i­tal scans of lymph node sec­tions, 110 with ma­lig­nant cells metic­u­lously la­belled by pathol­o­gists to show the can­cers’ lo­ca­tions.

The hu­man pathol­o­gists were given two hours to ex­am­ine the slides, mim­ick­ing real-life work­load in the Nether­lands. On av­er­age they spot­ted just 31 of 49 can­cers. One fur­ther pathol­o­gist, given no time limit and tak­ing 30 hours, found 46. The top-per­form­ing al­go­rithm, from the Har­vard Med­i­cal School and Mas­sachusetts In­sti­tute of Tech­nol­ogy, sig­nif­i­cantly out­per­formed the time­poor doc­tors and per­formed on par with the pathol­o­gist given 30 hours – a time, the au­thors note, “in­fea­si­ble in clin­i­cal prac­tice”.


Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.