AI beats ex­pert doc­tors at find­ing cer­vi­cal pre­can­cers

Daily Observer (Jamaica) - - NEWS -

TAMPA, United States (AFP) — Ar­ti­fi­cial in­tel­li­gence may be poised to wipe out cer­vi­cal cancer, af­ter a study showed on Thurs­day com­puter al­go­rithms can de­tect pre­can­cer­ous le­sions far bet­ter than trained ex­perts or con­ven­tional screen­ing tests.

Ac­cord­ing to the World Health Or­ga­ni­za­tion, cer­vi­cal cancer is the fourth-most com­mon cancer in women, with an es­ti­mated 570,000 new cases glob­ally in 2018.

De­spite ma­jor ad­vances in screen­ing and vac­ci­na­tion, which can pre­vent the spread of hu­man pa­pil­loma virus that causes most cases of cer­vi­cal cancer, those gains have mainly ben­e­fited women in rich na­tions.

Some 266,000 women died of cer­vi­cal cancer glob­ally in 2012, 90 per cent of them in low- and mid­dle-in­come na­tions, ac­cord­ing to the WHO.

“Cer­vi­cal cancer is now a dis­ease of poverty, of low re­sources,” said se­nior au­thor Mark Schiff­man, a doc­tor at the Na­tional Cancer In­sti­tute’s Divi­sion of Cancer Epi­demi­ol­ogy and Ge­net­ics near Washington, who has been search­ing for a cure to cer­vi­cal cancer for 35 years.

“We are try­ing to find ways that are ex­tremely cheap, ex­tremely easy but very ac­cu­rate, so that we can at­tack cer­vi­cal cancer by vac­cine and also a bit later through a sim­ple tech­nique that is cell­phone based or some­thing like it,” he told AFP.

Schiff­man was part of a team that built an al­go­rithm from an ar­chive of more than 60,000 cer­vi­cal im­ages col­lected from Costa Rica.

The pic­tures were taken us­ing just a specu­lum, small light and cam­era — no ad­vanced imag­ing re­quired.

The study be­gan in the 1990s, in­volv­ing more than 9,400 women who were fol­lowed for up to 18 years.

The AI tech­nique, called au­to­mated vis­ual eval­u­a­tion, found pre­can­cer­ous cells with 91 per cent ac­cu­racy, ac­cord­ing to a re­port pub­lished in the Jour­nal of the Na­tional Cancer In­sti­tute.

In com­par­i­son, a hu­man ex­pert re­view found 69 per cent of pre­can­cers, while con­ven­tional lab tests like Pap smears found 71 per cent.

Among women aged 25-49, who face the high­est risk of cer­vi­cal cancer, the AI al­go­rithm was even more ac­cu­rate, find­ing 97.7 per cent of pre­can­cer­ous cells.

“It per­formed much bet­ter than hu­mans look­ing at those same pic­tures. It cer­tainly per­formed a lot bet­ter than me,” Schiff­man said.

The goal is to roll out the tech­nol­ogy in the next three to five years, en­rolling more pa­tients in clin­i­cal tri­als world­wide and even­tu­ally mak­ing it eas­ily ac­ces­si­ble ev­ery­where.

Schiff­man said a deal has just been struck with a ma­jor phil­an­thropic group to as­sist in the process.

The tech­nol­ogy has not been patented on pur­pose, Schiff­man said. The aim is to keep costs very low so that women most in need can ben­e­fit.

“I think now we have a pos­si­ble tool that can go any­where — and not sac­ri­fice sci­en­tific qual­ity but ac­tu­ally of­fer a med­i­cally valid screen,” he said.

Jen­nifer Wu, an ob­ste­tri­cian-gy­nae­col­o­gist at Lenox Hill Hos­pi­tal in New York who was not in­volved in the re­search, called the find­ings “very ex­cit­ing”.

“This could re­ally cut down on a lot of missed cases of cer­vi­cal cancer, and al­low more pa­tients ac­cess to di­ag­no­sis and treat­ment,” she told AFP.

(Photo: AFP)

Ac­cord­ing to the World Health Or­ga­ni­za­tion, cer­vi­cal cancer is the fourth-most com­mon cancer in women, with an es­ti­mated 570,000 new cases glob­ally in 2018.

Newspapers in English

Newspapers from Jamaica

© PressReader. All rights reserved.