Los Abaphandi bobugcisa bobugcisa (I-WHO) Ukusuka kuGoogle esebenza neSibhedlele iYunivesithi yaseNyakatho-ntshona benze imodeli ye-AI enokufumana umhlaza wemiphunga. Ngokwedatha evela kuMbutho wezeMpilo weHlabathi, umhlaza wemiphunga (izicwili ezinobungozi emiphungeni) sesinye sezizathu zokufa kwabantu emhlabeni jikelele, sibulala abantu abangaphezulu kwezigidi ezibini ngonyaka kwaye sibulala abantu abaninzi. njengomhlaza wamabele.
Ukunceda abaqeqeshi bezempilo, ii-algorithms kunye neekhompyuter zinokunceda ekuphuhliseni iindlela eziphambili kwicandelo lokhathalelo lwempilo.
Nangona kunjalo, ukuze ezi zixhobo zibe luncedo, kufuneka zifikeleleke kwaye ziqondakale kuwo wonke umntu, oogqirha kunye nezigulana, ngaphandle kolwazi lobuchwephesha okanye lwekhompyuter.
Ngapha koko, kufanele ukuba kwaziwe ukuba ukusebenza kwazo zonke izixhobo zedijithali kusekwe kwiinkqubo zekhompyuter kunye nedatha.
Igama elithi "ubukrelekrele bokuzenzela" lithetha ukuba ezi zixhobo ziyakwazi ukuzicingela. Ukuba zilungiselelwe ngokuchanekileyo, izixhobo ezikrelekrele zinokuvavanya idatha ebonelelweyo kwaye zitshintshe iinkqubo okanye iiparameter "kubhabho." Banikwe ulwazi olwaneleyo, 'banokufunda' kwaye baguqule ikhowudi yabo ngokusekwe kwezi paramitha zintsha.
Kule minyaka mithathu idlulileyo, amaqela kuGoogle ebesebenzisa i-AI kwiingxaki kwezempilo, ukusuka ekuchongeni izifo zamehlo ukuya ekuqikeleleni iziphumo zezigulana kwiirekhodi zonyango.
Namhlanje sabelana ngophando olutsha olubonisa indlela i-AI ewuchaza ngayo umhlaza wemiphunga ngeendlela ezonyusa amathuba okusinda kubantu abaninzi abasemngciphekweni kwihlabathi liphela.
Ubukrelekrele bobugcisa bokuphucula umgangatho wobomi
Iinkcukacha kuphando olupapashwe ngoMeyi 20 kwindalo amayeza, imodeli yokufunda enzulu yayisetyenziselwa ukuqikelela ukuba isigulana sinomhlaza wemiphunga, Ukuvelisa amanqaku omngcipheko womhlaza wemiphunga kunye nokuchonga indawo yomhlaza wemiphunga.
"Ngokubonisa ukuba ukufunda okunzulu kunokonyusa ubuchwephesha ngaphandle kokuncama ubuntununtunu, sinethemba lokuvelisa uphando kunye nengxoxo malunga nendima ye-AI enokudlala ekutshintsheni indleko-yenqanaba lokufunyanwa komhlaza." , sinokufunda kwibhlog kaGoogle.
Inkqubo yobukrelekrele bokufakelwa isebenzisa ukufunda okunzulu kwe-3D yevolumetric ukuhlalutya yonke i-anatomy yesikena sesifuba, kunye neepatches ezisekwe kubuchwephesha bokuchonga into echonga imimandla enezilonda ezimbi, utshilo uShreeva Shetty. , umphathi wezobugcisa kuGoogle.
Ngokuhlalutya ukuskena kube kanye, imodeli yafumanisa umhlaza (ngokomndilili 5%) Rhoqo kunelo qela leengcali ezingabantu ezintandathu kwaye bekukho amathuba ali-11% okunciphisa izinto ezintle (into engeyonyani sisiphumo sesigqibo kunyulo olusinga kumacala amabini, lubhengezwe lulungile, apho lubi khona)
Iingcali ze-Radiologists zihlala zibona amakhulu emifanekiso ye-2D kwi-CT eyodwa, kwaye umhlaza unokuba mncinci kwaye kunzima ukuwubona. Senze imodeli engenakuvelisa kuphela ukubikezela okuqhelekileyo komhlaza wamaphaphu (ukubonwa ngevolumu ye-3D) kodwa ichonge nezicubu ezinobungozi emiphungeni (amaqhuqhuva emiphunga).
Imodeli inokuthi ithathele ingqalelo ulwazi olusuka kwizikena zangaphambili, eziluncedo ekuqikeleleni umngcipheko womhlaza wemiphunga kuba izinga lokukhula kwamaqhuqhuva emiphunga akrokrisayo anokubonisa ubundlobongela.
Ezi ziphumo zokuqala ziyakhuthaza, kodwa ezinye izifundo ziya kuvavanya ifuthe kunye nokusetyenziswa kwonyango.
Kuphando lwethu, sithathe iimeko ezingama-45,856 zesifuba ezingenanto esifubeni (ezinye apho umhlaza wafunyanwa khona) ukusuka kwidatha yophando ye-NIH ebekwe kuLingo lweSizwe lokuHlolwa kweMiphunga kunye neYunivesithi yaseNyakatho-ntshona. Siqinisekisile iziphumo ngokuseta kwedatha yesibini kwaye sikwaqhathanisa iziphumo zethu kunye neebhodi ze-radiologists eziqinisekisiweyo ze-6 zase-US.
UGoogle ubhengeza ukuba iyakwenza ukuba imodeli ifumaneke ngeGoogle Cloud Healthcare API njengoko iqhubeka nokuvavanya kunye nokuvavanya nemibutho yamaqabane.
Umthombo: https://www.blog.google/