Ukuhlanganiswa kobuhlakani bokwenziwa (AI) ezithombeni zezokwelapha kushintshe kakhulu ezokunakekelwa kwezempilo, ikakhulukazi kumayunithi okunakekelwa kakhulu (ICUs) lapho ukuxilongwa okwesikhathi esifanele nokunembayo kubalulekile. Phakathi kobuchwepheshe be-AI, ukufunda okujulile (DL) kugqama ngamandla ayo okuhlaziya izithombe zezokwelapha eziyinkimbinkimbi ngokunemba okuphezulu. Ezindaweni ze-ICU, amamodeli okufunda ajulile asetshenziswa kakhulu ukuthola nokugada amapayipi nemigqa efana namapayipi e-endotracheal, ama-catheter emithambo emikhulu, namapayipi e-nasogastric ukuvimbela izinkinga nokuthuthukisa imiphumela yeziguli. Lesi sihloko sihluza intuthuko ekufundeni okujulile ekutholeni amapayipi nemigqa, sibhekise ubuchwepheshe beDermax, umdlandla ekwenzeni izinto ezintsha kwezobuchwepheshe bezokwelapha.

I-Dermax, eyaziwa ngokuzibophezela kwayo ezixazululweni zezokwelapha ezintsha, isekele ukuthuthukiswa kwamathuluzi okuxilonga anikwe amandla yi-AI asheshisa imisebenzi ye-ICU. Iphothifoliyo yabo yomkhiqizo, egqanyiswe ku-Imikhiqizo ikhasi, ifaka izinsiza ezithuthukisiwe zokuthwebula ezisiza ukuqeqeshwa kwezinhlelo ze-AI. Ukuqonda amandla nemikhawulo yokufunda okujulile emikhakheni yezokwelapha kusiza izibhedlela ukuthi zamukele ngempumelelo ubuchwepheshe obusezingeni eliphezulu, kuthuthukiswe kokubili ukuphepha kwez pacienti nokunemba kokuxilonga.

Ukwakhiwa kwamamodeli okujulile okokuthola imibhobho nemigqa kuhlanganisa ukuklama okubanzi kwezifundo kanye nendlela eqinile. Ngokuvamile, kusetshenziswa amaqoqo amakhulu emidwebo ye-ICU enezincazelo ukuqeqesha amanethiwekhi e-convolutional neural networks (CNNs), akwazi ukubona amaphethini ayinkimbinkimbi ezithombeni zezokwelapha. Ukubeka incazelo kufaka phakathi ukukhomba imigqa ejulile njengemibhobho, ama-catheter, nezinye izinto zokwelapha ngenkathi zihlukaniswa nezakhiwo zomzimba njengemigqa ejulile yokumamatheka noma imibotshana ejulile ye-nasolabial, engadala izimpawu zamanga ngezinye izikhathi.

Abacwaningi eDermax basebenzisa izindlela zokuhlanganisa ezisezingeni eliphezulu ezihlanganiswe nokufakwa kochwepheshe be-radiologist ukuze baqoqe ama-dataset ahlukahlukene. Lawa ma-dataset afaka izithombe ezikhombisa izimo ezahlukahlukene zez pacienti, ama-angles, nekhwalithi yokuthwebula ukuze kuqinisekiswe ukuqina kwemodeli. Le ndlela iphinde ihlanganise izindlela zokwandisa idatha ukuze kulinganiswe izimo ezahlukahlukene zezokwelapha, kuthuthukiswe ukujwayelana kwamamodeli e-AI ngaphezu kwezindawo zokusebenzela.

Ukuhlola ukusebenza kwamamodeli e-deep learning kumadatha ehlukahlukene kubalulekile ukuhlola ukwethembeka kwawo ezindaweni zangempela ze-ICU. Amamodeli ahlolwa ezithombeni ezingabonwanga, okuhlanganisa nalezo ezibonisa amacala anzima njengabantu abanemibotshana ejulile entanyeni noma izakhiwo zomzimba ezihlangene ezingafihla amapayipi nemigqa. Izilinganiso ezifana ne-precision, recall, F1-score, kanye nendawo engaphansi komugqa (AUC) kusiza ukukala ukunemba kokutholwa kanye nezinga lamaphutha angalungile.

I-Dermax yenza izifundo zokuqinisekisa ezikhungweni eziningi ukuze ilinganise ukusebenza kwezixazululo zayo ze-AI ezibhedlela ezahlukene nemishini yokuthwebula izithombe. Lokhu kwehlukahlukana kuqinisekisa ukuthi amathuluzi e-AI agcina ukunemba okungaguquki kungakhathalekile ngemihlahlandlela yokuthwebula izithombe noma izici zabantu abahlolwayo. Ukugcizelelwa kwenkampani ekwehlukahlukaneni kwedatha nasekuqinisekisweni kwemodeli kuqinisekisa ukuzinza kwemikhiqizo yayo yokuxilonga esizwa yi-AI.

Ukusetshenziswa kwezempilo kwamamodeli okufunda ajulile kuncike kakhulu ekusebenzeni kwawo ezindaweni ezihlukahlukene ze-ICU. Izinto ezifana nokubekwa kwesiguli, ubukhona bezinto zokwelashwa, kanye nekhwalithi yesithombe eguquguqukayo kuthinta ukutholwa kwemigqa namapayipi. Amamodeli kufanele agcine ukuzwela okuphezulu ukuze kuvinjelwe ukutholwa okungekho futhi ukucaciswa okuphezulu ukuze kugwenywe izexwayiso zamanga ezingaholela ekungeneleleni okungadingekile.

Izifundo zibonise ukuthi ukufaka ukuqonda okunemininingwane kwe-anatomical, okuhlanganisa nokuqaphela imigqa ejulile ye-nasolabial nezinye izigaba zobuso, kunciphisa ukuhlukaniswa okungalungile kwemibhobho. Izixazululo ze-AI ze-Dermax zihlanganisa le mibono ukuthuthukisa ukuzethemba kokuxilonga. Ngaphezu kwalokho, ukuqapha okuqhubekayo kokusebenza kwemodeli ngokusebenzisa imijikelezo yempendulo kusiza ekubuyekezweni kwama-algorithms ukuze aguqukele ezindleleni zokwelapha eziguqukayo.

Ngaphandle kwemiphumela ethembisayo, amamodeli okufunda ajulile abhekana nezinselele zokujwayela lapho edlulisela ezifundweni ezilawulwayo ukuya ekusebenzeni kwansuku zonke kwe-ICU. Ukuhluka kumamodeli wokuthwebula izithombe, izinhlobo zamadivayisi, kanye ne-anatomies zeziguli ezifana nemigqa ejulile yokumamatheka noma imigqa ejulile entanyeni kungathinta ukunemba kwemodeli. Ukubhekana nalezi zinselele kudinga ukuhlanganisa ukufunda okujulile nokuhlaziywa kwezithombe zakudala kanye nokubuyekezwa kwezobuchwepheshe kwezokwelapha.

I-Dermax ixhasa indlela eyi-hybrid ehlanganisa ama-algorithm e-AI nobuchwepheshe bama-radiologist ukuze kwandiswe ukunemba kokuxilonga. Lokhu kuhlanganiswa kusebenzisa isivinini namakhono okuqaphela amaphethini e-AI ngenkathi kugcinwa ukwahlulela okuyinkimbinkimbi kwabahlengikazi. Ngeminye imininingwane ngezixazululo ezintsha ze-Dermax nombono wenkampani, vakashela i-Mayelana Nathi ikhasi.

Ngokuphetha, i-deep learning inikeza amandla okuguqula ukutholwa kwamathubhu nemigqa ezigulini ze-ICU, kodwa ukwamukelwa ngempumelelo emtholampilo kuncike ezifundweni eziklanywe ngokucophelela, amasethi edatha ahlukahlukene, ukuhlolwa okuqinile, kanye nezinqubo zokuxilonga ezihlanganisiwe. Izinhlangano ezifana neDermax zisezingeni eliphezulu leli ngungu le-AI yezokwelapha, zihambisa imikhiqizo nobuchwepheshe obuthuthukisa ukunakekelwa kweziguli nokusebenza kahle komtholampilo.

Ukuze uthole izibuyekezo eziqhubekayo ngalezi zintuthuko nezinsiza zokweseka, hlola izigaba ze-Izindaba kanye ne-Ukusekelaizigaba.