izindaba

Kusukela i-IBM Watson yaqala ngo-2007, abantu bebelokhu bephishekela ukuthuthukiswa kobuhlakani bokwenziwa kwezokwelapha (AI). Uhlelo lwe-AI lwezokwelapha olusebenzisekayo nolunamandla lunamandla amakhulu okubumba kabusha zonke izici zokwelapha zesimanje, luvumela ukunakekelwa okuhlakaniphile, okunembe kakhudlwana, okuphumelelayo, nokubandakanya wonke umuntu, okuletha inhlalakahle kubasebenzi bezokwelapha neziguli, futhi ngalokho kuthuthukisa kakhulu impilo yabantu. Eminyakeni engu-16 edlule, nakuba abacwaningi be-AI bezokwelapha beqoqe emikhakheni emincane ehlukahlukene, kulesi sigaba, abakakwazi ukuletha inganekwane yesayensi ngokoqobo.

Kulo nyaka, ngokuthuthukiswa koguquko kobuchwepheshe be-AI obufana ne-ChatGPT, i-AI yezokwelapha yenze intuthuko enkulu ezicini eziningi. Impumelelo engakaze ibonwe ekhonweni le-AI yezokwelapha: Ijenali yezemvelo iqhubeke yethula ucwaningo lwemodeli yolimi olukhulu lwezokwelapha kanye nemodeli eyisisekelo yesithombe sezokwelapha; I-Google ikhipha i-Med-PaLM kanye nendlalifa yayo, ifinyelela ezingeni lochwepheshe emibuzweni yokuhlolwa kukadokotela wase-US. Amajenali amakhulu ezemfundo azogxila ku-AI yezokwelapha: Imvelo ikhipha umbono wemodeli eyisisekelo ye-AI yezokwelapha ejwayelekile; Ukulandela uchungechunge lokubuyekezwa kwe-AI ku-Medicine ekuqaleni konyaka, i-New England Journal of Medicine (NEJM) ishicilele ukubuyekezwa kwayo kokuqala kwezempilo yedijithali ngoNovemba 30, futhi yethula umagazini wokuqala we-NEJM sub-journal NEJM AI ngoDisemba 12. Inhlabathi ye-Medical AI yokufika iyaqhubeka nokuvuthwa: I-JAMA-sub-journal yashicilela isinyathelo sokwabelana ngedatha yesithombe sezokwelapha emhlabeni wonke; I-US Food and Drug Administration (FDA) yenza uhlaka lwezinkombandlela zokulawulwa kwe-AI yezokwelapha.

Ngezansi, sibuyekeza inqubekelaphambili ebalulekile eyenziwe abacwaningi emhlabeni wonke mayelana ne-AI yezokwelapha esebenzisekayo ngo-2023.

Imodeli Eyisisekelo ye-AI yezokwelapha

Ukwakhiwa kwemodeli eyisisekelo ye-AI yezokwelapha ngokungangabazeki kugxile ocwaningweni olushisa kakhulu kulo nyaka. Amajenali Ezemvelo ashicilele izindatshana zokubuyekeza mayelana nemodeli Eyisisekelo Eyisisekelo Yokunakekelwa Kwezempilo kanye nemodeli enkulu yolimi lokunakekelwa kwezempilo phakathi nonyaka. I-Medical Image Analysis, ijenali ephezulu embonini, ibuyekeze futhi ibheke phambili ezinseleleni namathuba ocwaningo lwemodeli eyisisekelo ekuhlaziyeni isithombe sezokwelapha, futhi yaphakamisa umqondo "wozalo lwemodeli eyisisekelo" ukuze ifingqise futhi iqondise ukuthuthukiswa kocwaningo lwemodeli eyisisekelo ye-AI yezokwelapha . Ikusasa lamamodeli e-AI ayisisekelo okunakekelwa kwezempilo liya licaca. Ithathela ezibonelweni eziyimpumelelo zamamodeli amakhulu olimi njenge-ChatGPT, kusetshenziswa izindlela ezithuthuke kakhulu zokuziqondisa kusengaphambili kanye nokuqoqwa okukhulu kwedatha yokuqeqeshwa, abacwaningi emkhakheni we-AI yezokwelapha bazama ukwakha 1) amamodeli ayisisekelo aqondene nezifo, 2) amamodeli ayisisekelo ajwayelekile, kanye 3) namamodeli amakhulu ama-multimodal ahlanganisa izinhlobo eziningi zepharamitha namakhono amakhulu.

I-Medical Data Acquisition AI Model

Ngaphezu kwamamodeli amakhulu e-AI adlala indima enkulu emisebenzini yokuhlaziya idatha yomtholampilo engezansi, ekutholweni kwedatha yomtholampilo ephezulu, ubuchwepheshe obumelwe amamodeli e-AI akhiqizayo nawo sebuvele. Inqubo, isivinini, kanye nekhwalithi yokutholwa kwedatha ingathuthukiswa kakhulu ngama-algorithms e-AI.

Ngasekuqaleni kwalo nyaka, i-Nature Biomedical Engineering ishicilele ucwaningo oluvela eNyuvesi yaseTurkey iStraits olwalugxile ekusebenziseni i-AI ekhiqizayo ukuxazulula inkinga yokuxilonga okusizwa yisithombe se-pathologic ekusetshenzisweni komtholampilo. Izinto zobuciko eziseziculweni zesigaba esiqandisiwe phakathi nokuhlinzwa ziyisithiyo ekuhlolweni okusheshayo kokuxilonga. Nakuba i-formalin ne-parafini embedded tissue (FFPE) inikeza isampula yekhwalithi ephezulu, inqubo yayo yokukhiqiza idla isikhathi futhi ngokuvamile ithatha amahora angu-12-48, okwenza ingafaneleki ukusetshenziswa ekuhlinzeni. Ngakho-ke ithimba locwaningo liphakamise i-algorithm ebizwa ngokuthi i-AI-FFPE, engenza ukubonakala kwethishu esigabeni esiqandisiwe kufane ne-FFPE. I-algorithm ilungise ngempumelelo ama-artifact ezigaba eziqandisiwe, yathuthukisa ikhwalithi yesithombe, futhi yagcina izici ezihambisana nempilo ngesikhathi esifanayo. Ekuqinisekisweni komtholampilo, i-algorithm ye-AI-FFPE ithuthukisa kakhulu ukunemba kokuxilonga kwama-pathologists kuma-tumor subtypes, kuyilapho ifinyeza kakhulu isikhathi sokuxilongwa komtholampilo.

I-Cell Reports Medicine ibika umsebenzi wocwaningo owenziwe yithimba elivela e-Third Clinical College yase-Jilin University, uMnyango Weze-Radiology, Isibhedlela sase-Zhongshan Esixhumene ne-Fudan University, kanye ne-Shanghai University of Science and Technology [25]. Lolu cwaningo luhlongoza uhlaka lokufunda okujulile lwenhloso evamile kanye nohlaka lokuhlanganisa lokwakha kabusha oluphindaphindayo (Hybrid DL-IR) olunokuguquguquka okukhulu nokuguquguquka, olubonisa ukusebenza okuhle kakhulu kokwakhiwa kabusha kwesithombe ku-MRI esheshayo, umthamo ophansi we-CT, kanye ne-PET esheshayo. I-algorithm ingakwazi ukufeza ukuskena okulandelanayo kwe-MR Single-organ multi-sequence ngemizuzwana eyi-100, yehlise umthamo wemisebe ibe ngu-10% kuphela wesithombe se-CT, futhi isuse umsindo, futhi ingakha kabusha izilonda ezincane ekutholeni i-PET ngokusheshisa izikhathi ezingu-2 kuya kwezingu-4, kuyilapho inciphisa umphumela wama-artifacts anyakazayo.

I-Medical AI ngokubambisana nabasebenzi bezokwelapha

Ukuthuthuka okusheshayo kwe-AI yezokwelapha kuphinde kwaholela ochwepheshe bezokwelapha ukuthi bacabangele futhi bahlole ukuthi bangasebenzisana kanjani ne-AI ukuthuthukisa izinqubo zomtholampilo. NgoJulayi kulo nyaka, i-DeepMind kanye nethimba labacwaningi bezikhungo eziningi ngokuhlanganyela bahlongoza uhlelo lwe-AI olubizwa nge-Complementary Driven Clinical Workflow Delay (CoDoC) . Inqubo yokuxilonga iqala ukutholwa ngohlelo lwe-AI olubikezelayo, bese ihlulelwa ngolunye uhlelo lwe-AI kumphumela wangaphambilini, futhi uma kunokungabaza, ukuxilongwa ekugcineni kwenziwa ngumtholampilo ukuze kuthuthukiswe ukunemba kokuxilonga nokusebenza kahle kokulinganisela. Uma kukhulunywa ngokuhlolelwa umdlavuza webele, i-CoDoC yehlise amazinga angama-25% ngenani elifanayo elibi elingalungile, kuyilapho yehlisa umsebenzi wodokotela ngo-66%, uma kuqhathaniswa nenqubo yamanje “yokulamula okufundwa kabili” e-UK. Ngokuya ngezigaba ze-TB, amazinga amahle angamanga ancishiswa ngamaphesenti angu-5 kuya kwangu-15 ngezinga elifanayo elingalungile elingelona iqiniso uma kuqhathaniswa ne-AI ezimele kanye nokugeleza kokusebenza komtholampilo.

Ngokufanayo, u-Annie Y. Ng et al., weKheiron Company eLondon, e-UK, wethule abafundi be-AI abengeziwe (ngokubambisana nabahloli abangabantu) ukuze baphinde bahlole imiphumela lapho ingekho imiphumela yokukhumbula enqubweni yokulamula efundwe kabili, ethuthukise inkinga yokungabonwa ekuhlolweni komdlavuza webele kusenesikhathi, futhi inqubo cishe yayingenazo izimpawu ezingamanga . Olunye ucwaningo, oluholwa ithimba eNyuvesi yaseTexas McGovern Medical School futhi lwaphothulwa ezikhungweni ezine zokushaywa unhlangothi, lwasebenzisa ubuchwepheshe be-AI obusekelwe ku-computed tomography (CTA) -based AI ukuze kutholakale ukutholakala kwe-vascular occlusive ischemic stroke (LVO). Odokotela kanye nama-radiologists bathola izexwayiso zesikhathi sangempela kumakhalekhukhwini wabo phakathi nemizuzu yokuqedwa kwesithombe se-CT, okubazisa ngokuba khona okungenzeka kwe-LVO. Le nqubo ye-AI ithuthukisa ukuhamba komsebenzi esibhedlela ngenxa ye-acute ischemic stroke, inciphisa isikhathi sokuya emnyango kusukela ekungenweni kuya ekwelashweni kanye nokuhlinzeka ngamathuba okuhlenga ngempumelelo. Okutholakele kushicilelwe ku-JAMA Neurology.

Imodeli Yokunakekelwa Kwezempilo ye-AI Yenzuzo Yomhlaba Wonke

U-2023 uzobona nomsebenzi omningi omuhle osebenzisa i-AI yezokwelapha ukuthola izici ezingabonakali ngeso lomuntu emininingwaneni etholakala kalula, evumela ukuxilonga kwendawo yonke kanye nokuhlolwa kusenesikhathi esikalini. Ekuqaleni konyaka, i-Nature Medicine yashicilela izifundo ezenziwe yi-Zhongshan Eye Centre yase-Sun Yat-sen University kanye ne-Second Affiliated Hospital of Fujian Medical University. Besebenzisa ama-smartphone njengamatheminali ohlelo lokusebenza, basebenzisa izithombe zevidiyo ezifana nekhathuni ukuze bahehe ukubuka kwezingane futhi barekhode indlela izingane ezibuka ngayo kanye nezici zobuso, baphinde bahlaziya amamodeli angavamile besebenzisa amamodeli okufunda ajulile ukuze bahlonze ngempumelelo izifo zamehlo eziyi-16, okuhlanganisa ulwelwesi oluzalwa nalo, i-congenital ptosis kanye ne-congenital glaucoma, enesilinganiso sokunemba esingaphezu kuka-85%. Lokhu kunikeza izindlela zobuchwepheshe eziphumelelayo nezilula zokwazisa ngokuhlolwa kwangaphambi kwesikhathi okukhulu kokukhubazeka kokubonakala kwengane kanye nezifo zamehlo ezihlobene.

Ekupheleni konyaka, i-Nature Medicine yabika umsebenzi owenziwe izikhungo zezokwelapha nokucwaninga ezingaphezu kwe-10 emhlabeni jikelele, kuhlanganise ne-Shanghai Institute of Pancreatic Disease kanye neSibhedlela Esihlanganisiwe Sokuqala saseZhejiang University. Umbhali usebenzise i-AI ekuhlolweni komdlavuza we-pancreatic wabantu abangenazimpawu ezikhungweni zokuhlola ngokomzimba, ezibhedlela, njll., ukuthola izici zezilonda ezithombeni ezicacile ze-CT okunzima ukuzibona ngeso lenyama kuphela, ukuze kuzuzwe ukutholwa okusebenzayo nokungaphazamisi umdlavuza we-pancreatic. Ekubuyekezeni idatha evela ezigulini ezingaphezu kuka-20,000, imodeli iphinde yaveza amacala angu-31 ezilonda ezigejiwe emtholampilo, ezithuthukise kakhulu imiphumela yomtholampilo.

Ukwabelana Ngedatha Yezokwelapha

Ngo-2023, izindlela eziningi ezinhle kakhulu zokwabelana ngedatha kanye namacala aphumelele avele emhlabeni wonke, aqinisekisa ukubambisana kwezikhungo eziningi nokuvuleka kwedatha ngaphansi kwesisekelo sokuvikela ubumfihlo nokuphepha kwedatha.

Okokuqala, ngosizo lobuchwepheshe be-AI ngokwayo, abacwaningi be-AI baye bafaka isandla ekwabelaneni ngedatha yezokwelapha. U-Qi Chang kanye nabanye abavela eNyuvesi yase-Rutgers e-United States bashicilele isihloko ku-Nature Communications, ehlongoza uhlaka lokufunda lukahulumeni lwe-DSL olusekelwe kumanethiwekhi okwenziwa okuphikisana nawo, esebenzisa i-AI ekhiqizayo ukuqeqesha idatha ethile ekhiqiziwe yezikhungo eziningi, bese ibuyisela idatha yangempela yezikhungo eziningi ngedatha ekhiqiziwe. Qinisekisa ukuqeqeshwa kwe-AI okusekelwe kudatha enkulu yezikhungo eziningi kuyilapho uvikela ubumfihlo bedatha. Ithimba elifanayo liphinde livule umthombo idathasethi yezithombe ezikhiqiziwe ze-pathological kanye nezichasiselo zazo ezihambisanayo. Imodeli yokuhlukanisa eqeqeshwe kusethi yedatha ekhiqiziwe ingafinyelela imiphumela efanayo kudatha yangempela.

Ithimba lika-Dai Qionghai waseNyuvesi yaseTsinghua lishicilele iphepha nge-npj Digital Health, ehlongoza i-Relay Learning, esebenzisa idatha enkulu yezindawo eziningi ukuqeqesha amamodeli e-AI ngaphansi kwesisekelo sobukhosi bedatha yendawo futhi akukho uxhumano lwenethiwekhi ye-cross-site. Ilinganisa ukuphepha kwedatha nokukhathazeka kobumfihlo nokuphishekela ukusebenza kwe-AI. Ithimba elifanayo kamuva lasungula ngokuhlanganyela futhi laqinisekisa i-CAIMEN, uhlelo lwesifuba lwe-CT pan-mediastinal tumor diagnosis olusekelwe ekufundeni kombuso, ngokubambisana neSibhedlela Esihlanganisiwe Sokuqala sase-Guangzhou Medical University kanye nezibhedlela ezingu-24 ezweni lonke. Lolu hlelo, olungase lusetshenziswe ezimila ezivamile ezingu-12 ze-mediastinal, lufinyelele ukunemba okungcono ngamaphesenti angu-44,9 lapho lusetshenziswa lulodwa kunalapho lusetshenziswa ochwepheshe abangabantu bebodwa, kanye namaphesenti angu-19 ukunemba okungcono kokuxilongwa lapho ochwepheshe babantu besizwa yilo.

Ngakolunye uhlangothi, kunezinhlelo ezimbalwa eziqhubekayo zokwakha amasethi edatha yezokwelapha evikelekile, yomhlaba wonke, nezinga elikhulu. NgoNovemba 2023, u-Agustina Saenz kanye nabanye boMnyango Wezokwaziswa Kwezinto Eziphilayo e-Harvard Medical School banyathelisa ku-inthanethi ku-Lancet Digital Health uhlaka lomhlaba wonke lokwabelana ngedatha yesithombe sezokwelapha olubizwa nge-Artificial Intelligence Data for All Healthcare (MAIDA). Basebenzisana nezinhlangano zezempilo emhlabeni wonke ukuze banikeze umhlahlandlela ophelele wokuqoqwa kwedatha kanye nokuyekiswa ukuhlonza, kusetshenziswa ithempulethi yase-US Federal Demonstration Partner (FDP) ukuze kumiswe ukwabelana kwedatha. Bahlela ukukhipha kancane kancane amasethi edatha aqoqwe ezifundeni ezahlukene kanye Nezilungiselelo zomtholampilo emhlabeni jikelele. Idathasethi yokuqala kulindeleke ukuthi ikhishwe ekuqaleni kuka-2024, nokunye okuzayo njengoba ubudlelwano bukhula. Le phrojekthi ingumzamo obalulekile wokwakha isethi yomhlaba wonke, enkulu futhi ehlukahlukene yedatha ye-AI etholakala esidlangalaleni.

Ngemuva kwalesi siphakamiso, i-UK Biobank yenze isibonelo. I-UK Biobank ikhiphe idatha entsha ngomhla zingama-30 kuLwezi kusukela kulo lonke uhlelo lofuzo lwabahlanganyeli bayo abangu-500,000. Isizindalwazi, esishicilela ukulandelana kofuzo okupheleleyo kwevolontiya ngalinye lama-500,000 aseBrithani, iyisizindalwazi esikhulu kunazo zonke esiphelele sofuzo somuntu emhlabeni. Abacwaningi emhlabeni wonke bangacela ukufinyelela kule datha engakhonjiwe futhi bayisebenzisele ukuhlola isisekelo sofuzo sezempilo nesifo. Idatha yezofuzo ibilokhu ibucayi kakhulu ekuqinisekisweni esikhathini esedlule, futhi le mpumelelo eyingqophamlando ye-UK Biobank ifakazela ukuthi kungenzeka ukwakha isizindalwazi somhlaba wonke esivulekile, esingenamfihlo. Ngalobu buchwepheshe kanye nedathabheyisi, i-AI yezokwelapha iboshelwe ukuletha igxathu elilandelayo.

Ukuqinisekiswa Nokuhlolwa Kwe-AI Yezokwelapha

Uma kuqhathaniswa nokuthuthuka okusheshayo kobuchwepheshe be-AI yezokwelapha ngokwayo, ukuthuthukiswa kokuqinisekisa nokuhlolwa kwe-AI yezokwelapha kuhamba kancane. Ukuqinisekisa nokuhlola emkhakheni we-AI ojwayelekile kuvame ukungazinaki izidingo zangempela zabahlengikazi neziguli ze-AI. Izivivinyo zendabuko ezilawulwa ngokungahleliwe zikhandla kakhulu ukufanisa ukuphindwaphindwa okusheshayo kwamathuluzi e-AI. Ukuthuthukisa uhlelo lokuqinisekisa nokuhlola olufanele amathuluzi e-AI yezokwelapha ngokushesha okukhulu kuyinto ebaluleke kakhulu ukukhuthaza i-AI yezokwelapha ocwaningweni lwe-leapfrog ngempela kanye nentuthuko ekufikeni komtholampilo.

Ephepheni lokucwaninga le-Google nge-Med-PaLM, elishicilelwe kwezeMvelo, ithimba liphinde lashicilela i-MultiMedQA yokuhlola ibhentshimakhi, esetshenziselwa ukuhlola ikhono lamamodeli ezilimi ezinkulu ukuze bathole ulwazi lomtholampilo. Ibhentshimakhi ihlanganisa amasethi edatha edatha ye-Q&A yezokwelapha ayisithupha, afaka ulwazi lwezokwelapha oluchwepheshile, ucwaningo nezinye izici, kanye nedathasethi yedatha yedatha yemibuzo yosesho lwezokwelapha ku-inthanethi, kucatshangelwa i-Q&A eku-inthanethi yesiguli nesiguli, ezama ukuqeqesha i-AI ibe udokotela oqeqeshiwe ovela ezicini eziningi. Ngaphezu kwalokho, ithimba lihlongoza uhlaka olusekelwe ekuhloleni komuntu olucabangela izingxenye eziningi zeqiniso, ukuqonda, ukucabanga, kanye nokuchema okungenzeka. Lona omunye wemizamo yocwaningo emele kakhulu yokuhlola i-AI ekunakekelweni kwezempilo eshicilelwe kulo nyaka.

Kodwa-ke, ingabe iqiniso lokuthi amamodeli amakhulu ezilimi abonisa izinga eliphezulu lombhalo wekhodi wolwazi lomtholampilo lisho ukuthi amamodeli amakhulu olimi ayakwazi ukwenza imisebenzi yomtholampilo yomhlaba wangempela? Njengoba nje umfundi wezokwelapha ophumelela ukuhlolwa kodokotela oqeqeshiwe ngamaphuzu aphelele kusekude nodokotela oyinhloko eyedwa, indlela yokuhlola ehlongozwe i-Google ingase ingabi impendulo ephelele esihlokweni sokuhlolwa kwe-AI yezokwelapha kumamodeli e-AI. Kusukela ngo-2021 nango-2022, abacwaningi baye bahlongoza imihlahlandlela yokubika efana ne-Decid-AI, SPIRIT-AI, ne-INTRPRT, ngethemba lokuqondisa ukuthuthukiswa kwangaphambi kwesikhathi nokuqinisekiswa kwe-AI yezokwelapha ngaphansi kwesimo sokucatshangelwa kwezinto ezifana nokusebenza komtholampilo, ukuphepha, izici zomuntu, kanye nokucaca/ukuchazwa. Muva nje, ijenali ye-Nature Medicine ishicilele ucwaningo olwenziwa abacwaningi abavela eNyuvesi yase-Oxford nase-Stanford University mayelana nokuthi kufanele kusetshenziswe "ukuqinisekisa kwangaphandle" noma "ukuqinisekisa kwasendaweni okuphindelelayo. "Ukuqinisekisa amathuluzi e-AI.

Imvelo engachemile yamathuluzi e-AI futhi iyisiqondiso sokuhlola esibalulekile esithole ukunakwa kulo nyaka kuzo zombili izindatshana zeSayensi ne-NEJM. I-AI ivame ukubonisa ukuchema ngoba ikhawulelwe kudatha yokuqeqesha. Lokhu kuchema kungase kubonise ukungalingani komphakathi, okuqhubeka kuguqukela ekubandlululweni kwe-algorithmic. I-National Institutes of Health isanda kwethula uhlelo lwe-Bridge2AI, okulinganiselwa ukuthi lubiza izigidi ezingu-130 zamaRandi, ukuze kwakhiwe amasethi edatha ahlukahlukene (ngokuhambisana nemigomo yesinyathelo se-MAIDA okukhulunywe ngenhla) esingasetshenziswa ukuze kuqinisekiswe ukungathathi hlangothi kwamathuluzi ezokwelapha e-AI. Lezi zici azibhekwa yi-MultiMedQA. Umbuzo wokuthi ulinganisa futhi uqinisekise kanjani amamodeli e-AI yezokwelapha usadinga ingxoxo ebanzi futhi ejulile.

NgoJanuwari, Imithi Yemvelo ishicilele ucezu lwemibono olubizwa ngokuthi “Isizukulwane Esilandelayo Somuthi Osekelwe Ebufakazini” evela ku-Vivek Subbiah wase-University of Texas MD Anderson Cancer Center, ebuyekeza imikhawulo yokuhlolwa kwemitholampilo edaluliwe kumqedazwe we-COVID-19 futhi yaveza ukungqubuzana phakathi kokusungula izinto ezintsha nokubambelela enqubweni yocwaningo lomtholampilo. Okokugcina, libonisa ikusasa lokuhlelwa kabusha kokuhlolwa kwemitholampilo - isizukulwane esilandelayo sokuhlolwa komtholampilo kusetshenziswa ubuhlakani bokwenziwa, okungukuthi, ukusetshenziswa kobuhlakani bokwenziwa kusuka kudatha enkulu yocwaningo lomlando, idatha yomhlaba wangempela, idatha yomtholampilo enezimo eziningi, idatha yedivayisi egqokekayo ukuze uthole ubufakazi obubalulekile. Ingabe lokhu kusho ukuthi ubuchwepheshe be-AI kanye nezinqubo zokuqinisekisa emtholampilo ze-AI zingase ziqinise futhi ziguquke ngokuzayo? Lona umbuzo ovulekile futhi oshukumisa ingqondo ka-2023.

Ukulawulwa kwe-AI yezokwelapha

Ukuthuthuka kobuchwepheshe be-AI nakho kubangela izinselelo ekulawuleni i-AI, futhi abenzi benqubomgomo emhlabeni wonke baphendula ngokucophelela nangokucophelela. Ngo-2019, i-FDA yaqala yashicilela Uhlaka Lokulawula Oluhlongozwayo Lwezinguquko Zesofthiwe Kumadivayisi Ezokwelapha Ezobuciko (Okusalungiswa Kwengxoxo), ichaza indlela engaba khona yokubuyekezwa kwemakethe ye-AI kanye nokuguqulwa kwesoftware eqhutshwa ukufunda ngomshini. Ngo-2021, i-FDA yenza isiphakamiso sokuthi “I-Artificial Intelligence/Isoftware-based Learning-based Software as a Medical Device Action Plan”, eyacacisa izinyathelo ezinhlanu eziqondile zokulawula ukwelashwa kwe-AI. Kulo nyaka, i-FDA iphinde yakhipha Ukuhanjiswa Kwemakethe Engaphambili Kwezici Zesofthiwe Yedivayisi ukuze inikeze ulwazi mayelana nezincomo zokuthunyelwa kwemakethe yangaphambi kwemakethe ukuze kuhlolwe i-FDA ukuphepha nokusebenza kahle kwezici zesofthiwe yedivayisi, okuhlanganisa izici ezithile zedivayisi yesofthiwe ezisebenzisa amamodeli okufunda omshini aqeqeshwe ngezindlela zokufunda zomshini. Inqubomgomo yokulawula ye-FDA iguqukile kusukela esiphakamisweni sokuqala kuya kumhlahlandlela osebenzayo.

Ngemva kokushicilelwa kwe-European Health Data Space ngoJulayi wonyaka odlule, i-EU isiphinde yashaya umthetho we-Artificial Intelligence Act. Owokuqala uhlose ukusebenzisa kahle idatha yezempilo ukuze unikeze ukunakekelwa kwezempilo okusezingeni eliphezulu, ukunciphisa ukungalingani, kanye nedatha yokusekela yokuvimbela, ukuxilongwa, ukwelashwa, ukusungulwa kwesayensi, ukuthathwa kwezinqumo kanye nomthetho, kuyilapho kuqinisekisa ukuthi izakhamuzi ze-EU zinokulawula okukhulu idatha yazo yezempilo yomuntu siqu. Lesi sakamuva sikwenza kucace ukuthi uhlelo lokuxilongwa kwezokwelapha luwuhlelo lwe-AI oluyingozi kakhulu, futhi ludinga ukwamukela ukugadwa okuqinile okuhlosiwe, ukugadwa komjikelezo wempilo yonke kanye nokugadwa kwangaphambili kokuhlola. I-European Medicines Agency (EMA) ishicilele i-Draft Reflection Paper mayelana nokusetshenziswa kwe-AI ukusekela ukuthuthukiswa kwezidakamizwa, ukulawulwa nokusetshenziswa, ngokugcizelela ekuthuthukiseni ukwethembeka kwe-AI ukuze kuqinisekiswe ukuphepha kwesiguli kanye nobuqotho bemiphumela yocwaningo lomtholampilo. Sekukonke, indlela yokulawula ye-EU iya ngokuya ikhula, futhi imininingwane yokugcina yokusetshenziswa ingase ibe nemininingwane eminingi futhi iqine. Ngokuphambene kakhulu nomthetho we-EU oqinile, ipulani yokulawula i-AI yase-UK ikwenza kucace ukuthi uhulumeni uhlela ukuthatha indlela ethambile futhi angashayi izikweletu ezintsha noma amise abalawuli abasha okwamanje.

E-China, i-Medical Device Technical Review Centre (NMPA) ye-National Medical Products Administration ike yakhipha ngaphambilini imibhalo efana nokuthi “Buyekeza Amaphuzu E-Deep Learning Assisted Decision Software”, “Izimiso Eziqondisayo Zokubuyekezwa Kokubhaliswa Kwamadivayisi Ezokwelapha Obuhlakani Okwenziwayo (Okusalungiswa Kokuphawula)” kanye “Nesekhula mayelana Nezimiso Eziqondisayo Zomkhiqizo Wokuklanywa Nobuhlakani Be-Artificial Intelligence Medical Devices (No. 47 ngo-2021)”. Kulo nyaka, "Isifinyezo semiphumela yokuhlukaniswa komkhiqizo wokuqala wensiza yezokwelapha ngo-2023" siphinde sakhululwa. Lolu chungechunge lwamadokhumenti lwenza incazelo, ukuhlelwa nokulawulwa kwemikhiqizo yesofthiwe yezokwelapha ezihlakaniphile kucace futhi kube lula ukusebenza, futhi inikeza isiqondiso esicacile sokumisa umkhiqizo kanye namasu okubhalisa amabhizinisi ahlukahlukene embonini. Le mibhalo ihlinzeka ngohlaka kanye nezinqumo zokuphatha ze-AI ezibheka isayensi yezokwelapha. I-China Medical Artificial Intelligence Conference ebibanjelwe e-Hangzhou kusukela mhla zingama-21 kuya zingama-23 kuZibandlela isungule inkundla ekhethekile yokuphatha kwezokwelapha ngedijithali kanye nokuthuthukiswa kwekhwalithi ephezulu kwezibhedlela zikahulumeni kanye nenkundla yokuhlolwa kwemishini yezokwelapha yokwenziwa kobuhlakani kanye nesithangami sokuthuthukiswa kobuchwepheshe bokulinganisa imboni Ngaleso sikhathi, izikhulu zeKhomishini Yentuthuko Kazwelonke Nezinguquko kanye ne-NMPA zizothamela umhlangano futhi zingase zikhiphe ulwazi olusha.

Isiphetho

Ngo-2023, i-AI yezokwelapha isiqalile ukuhlanganisa kuyo yonke inqubo yezokwelapha ekhuphukayo nephansi, ehlanganisa ukuqoqwa kwedatha yesibhedlela, ukuhlanganiswa, ukuhlaziya, ukuxilongwa nokwelashwa, nokuhlolwa komphakathi, kanye nokusebenzisana okuphilayo nabasebenzi bezokwelapha / bezifo, okubonisa amandla okuletha inhlalakahle empilweni yomuntu. Ucwaningo lwe-AI lwezokwelapha olusebenzisekayo seluqala ukuphuma. Esikhathini esizayo, inqubekelaphambili ye-AI yezokwelapha ayincikile kuphela ekuthuthukisweni kwezobuchwepheshe ngokwayo, kodwa futhi idinga ukubambisana okugcwele kwezimboni, inyuvesi kanye nocwaningo lwezokwelapha kanye nokusekelwa kwabenzi bezinqubomgomo nabalawuli. Lokhu kubambisana kwesizinda esihlukene kuyisihluthulelo sokuzuza izinsizakalo zezokwelapha ezididiyelwe ze-AI, futhi ngokuqinisekile kuzokhuthaza ukuthuthukiswa kwempilo yabantu.