I-Endoscope yezoNyango iTekhnoloji yoMnyama (3) I-AI yexesha elililo loNcedo oluNcedayo

Ixesha lokwenyani i-AI incedise ukuxilonga i-endoscopes yezonyango yenye yezona teknoloji ziguqukayo kwicandelo lobukrelekrele bokwenziwa kwezonyango kwiminyaka yakutshanje. Ngokudityaniswa okunzulu kwe-algorithms yokufunda okunzulu kunye nemifanekiso ye-endoscopic, iphumelele uphuhliso lwe-leapfrog ukusuka "kweyeza elinobungqina" ukuya "kunyango lobukrelekrele oluchanekileyo". Oku kulandelayo kubonelela ngohlalutyo olubanzi ukusuka kwimilinganiselo yesi-8:

1. Imigaqo yobugcisa kunye noyilo lwenkqubo

Amalungu angundoqo:

Umaleko wokufunyanwa komfanekiso: 4K/8K ikhamera yenkcazo ephezulu+uphuculo lokukhanya (NBI/FECE)

Umaleko wokusetyenzwa kwedatha: itshiphu yokunyusa i-AI ezinikeleyo (efana ne-NVIDIA IGX)

Umaleko wemodeli ye-algorithm:

Convolutional Neural Networks (CNN): ResNet50, EfficientNet, njl

Imodeli yokuhlalutya uthotho lwexesha: I-LSTM yokusetyenzwa komjelo wevidiyo

I-Multimodal fusion: ukudibanisa ukukhanya okumhlophe / i-NBI / imifanekiso ye-fluorescence

Ujongano olusebenzayo: isichasiselo sexesha lokwenyani+umboniso womgangatho womngcipheko

Ukuhamba komsebenzi:

Ukufunyanwa komfanekiso → ukusetyenzwa kwangaphambili (i-denoising/uphuculo) → uhlalutyo lwe-AI (ukubona isilonda/ukuhlelwa) → umboniso wexesha lokwenyani (ukumakishwa komda/ukwenziwa komgangatho) → ukukhangela ngotyando

2. Iimpumelelo eziphambili zeteknoloji

I-algorithm entsha:

Isampula encinci yokufunda: ukusombulula ingxaki yedatha enenkcazo enganelanga

Itekhnoloji yokulungelelanisa isizinda: Ukulungelelanisa kwimifanekiso yezixhobo ezivela kubavelisi abahlukeneyo

Ukwakhiwa kwakhona kwesilonda se-3D: uqikelelo lwevolumu olusekwe kwimifanekiso yesakhelo ezininzi

Ukufunda kwemisebenzi emininzi: ukuphunyezwa okuhambelanayo kokufunyanwa / ukuhlelwa / ukwahlula

Ukukhawulezisa izixhobo:

Umda wezixhobo zekhompyutha (ukuqiqa ukulibaziseka<50ms)

Iprosesa ye-AI ye-endoscope ekhethekileyo (enjenge-chip ye-Olympus ENDO-AID)

3. Iimeko eziphambili zesicelo seklinikhi

Imeko yoxilongo:

Uvavanyo lomhlaza wesisu sokuqala (ubuntununtunu 96.3%)

Ucalucalulo lwexesha lokwenyani lweepropathi zepolyp (ukwanda kwenqanaba lokubona i-adenoma ngama-28%).

Uvavanyo olubukhali lwesifo samathumbu esivuthayo (ukubala okuzenzekelayo kwendawo yesilonda)

Imeko yonyango:

ESD/EMR navigation yotyando (ukuchaneka kwenqanawa 99.1%)

Uqikelelo lwengozi yokopha (isilumkiso sangexesha lokwenyani)

Ukucwangciswa okukrelekrele koluhlu lwe-resection

4. Ukuthelekiswa kweemveliso eziqhelekileyo kunye neeparitha zobugcisa

5. Ukuqinisekiswa kwexabiso leklinikhi

Idatha yophando lwamaziko amaninzi:

Iziko leSizwe loMhlaza laseJapan: I-AI iNcedisa Ukunyuka kwiNqanaba lokuFumana uMhlaza weGastric ekuqaleni ukusuka kwi-72% ukuya kwi-89%

Uphononongo lweKlinikhi yaseMayo: Inkqubo yeColonoscopy AI yehlisa izinga lokuxilongwa kwe-adenoma ngama-45%

Uphononongo lwaseTshayina lwe-REAL: Ukonyuka kobuntununtunu bokuchongwa komhlaza we-esophageal ngama-32%

Iinzuzo zoqoqosho lwezempilo:

I-27% yokunciphisa iindleko zokuhlola (ukunciphisa i-biopsies engadingekile)

Umjikelo woqeqesho lukaGqirha ufinyeziwe nge-40%

Umthamo wokuhlolwa kwemihla ngemihla unyuke nge-35%

6. Iibhotile kuphuhliso lobuchwepheshe

Imingeni yangoku:

Umba wesilo sedatha (imigangatho yokucinga engangqinelaniyo phakathi kwezibhedlele)

Ukwenziwa kwezigqibo kwibhokisi emnyama (ukungatoliki ngokwaneleyo kwesiseko sokugweba se-AI)

Ukuhambelana kwezixhobo (kunzima ukuziqhelanisa neempawu ezahlukeneyo ze-endoscopes)

Iimfuno zexesha langempela (ulawulo lokulibaziseka kwevidiyo ye-4K)

Isisombululo:

Ukufunda ngokudibeneyo kuphelisa imiqobo yedatha

Imephu yobushushu ebonwayo ichaza ukwenziwa kwezigqibo kwe-AI

Ujongano olusemgangathweni lwe-DICOM-MEIS

Ukuphuculwa kwe-AI ezinikeleyo ye-inference chip

7. Inkqubela phambili yeteknoloji yamva nje

Umda womda:

Amawele edijithali otyando: ukulinganisa kwangaphambili + uthelekiso lwexesha lokwenyani ngexesha lotyando

I-Multimodal fusion: ukudibanisa idatha ye-endoscopic ultrasound / OCT

Ukufunda ngokuzigadayo: ukunciphisa ukuxhomekeka kwiinkcazo

Intsebenziswano yamafu: 5G + edge computing architecture

Ukuphumelela okuphunyeziweyo:

I-EndoGPT, i "Endoscopic Vision Model" echazwe kwiNdalo ye-BME ngo-2023

Ixesha lokwenyani le-3D yokujonga inkqubo yotyando ye-AI eyenziwe yiYunivesithi yaseStanford

IRobhothi yeShurui yasekhaya eDityanisiweyo ye-AI yokuLawula umbono

8. IiNdlela zoPhuhliso lwexesha elizayo

Ukuvela kobuchwepheshe:

Indaleko ukusuka kuxilongo oluncedisayo ukuya kuqhaqho oluzimeleyo

Inkqubo yoBoniso ye-AI ye-Multidisciplinary (Endoscopy+Pathology+Imaging)

I-AI echazwayo (XAI) yongeza ukuthenjwa kwezonyango

I-Quantum computing ikhawulezisa uqeqesho lwemodeli

I-ikholoji yoshishino:

I-Endoscopy AI njengeNkonzo (EaaS) imodeli

Izinto ezisetyenziswayo ezidibeneyo ezikrelekrele (ezifana neenaliti ze-AI ze-biopsy)

Ukuxilongwa ngokuzenzekelayo kunye nenkqubo yonyango (ukususela ekuhlolweni ukuya ekulandeleni)

Umboniso wemeko yeklinikhi

Iimeko zosetyenziso oluqhelekileyo:

(1) Uvavanyo lomhlaza wesisu:

I-AI yexesha lokwenene ileyibhile yezilonda ezikrokrelekayo (imida / ii-microvessels / izakhiwo ezingaphezulu)

Veza ngokuzenzekela ingxelo yohlelo lwe-LABC

Ingcebiso ekrelekrele kwindawo ye-biopsy

(2) Uqhaqho lwe-ESD enemibala:

Ukuqikelelwa kobunzulu bokungena kwethumba

Ukwakhiwa kwakhona kwe-vascular course emithathu

Ukhuseleko lomda oguqukayo ukukhawuleza

Isishwankathelo kunye nembonakalo

I-medical endoscope AI iqhuba inguqu ukusuka "kwindawo enye yokuphumelela" ukuya "kwinkqubo yobukrelekrele":

Ixesha elifutshane (i-1-3 iminyaka): I-AI iba yinkqubo esemgangathweni ye-endoscopy, enezinga lokungena elingaphezu kwe-60%

Ixesha eliphakathi (iminyaka emi-3-5): Ukufezekisa ukuzenzekelayo kwayo yonke inkqubo yokuxilongwa kunye nonyango

Ixesha elide (iminyaka emi-5 ukuya kwe-10): Ukupapashwa kweerobhothi zotyando ezizimeleyo

Le teknoloji iya kuhlengahlengisa i-paradigm yokuxilongwa kwe-endoscopic kunye nonyango, ekugqibeleni iqonde umbono wokhathalelo lwezempilo olubandakanyayo apho sonke isigulane sinokonwabela ukuxilongwa kwenqanaba leengcali kunye neenkonzo zonyango.