Ny mpanoratra: Izany dia Ny vadiny dia Holger R. Roth. Aoxiao Zhong Ahmed Harouni Amilcare Gentili Anas Z. Abidin Andrew Liu Anthony Beardsworth Costa Bradford J. Wood Chien-Sung Tsai Chih-Hung Wang Chun-Nan Hsu C. K. Lee Peiying Ruan Daguang Xu Dufan Wu Eddie Huang Felipe Campos Kitamura Griffin Lacey Gustavo César de Antônio Corradi Gustavo Nino Hao-Hsin Shin Hirofumi Obinata Hui Ren Jason C. Crane Jesse Tetreault Jiahui Guan John W. Garrett Joshua D. Kaggie Jung Gil Park Keith Dreyer Krishna Juluru Kristopher Kersten Marcio Aloisio Bezerra Cavalcanti Rockenbach Marius George Linguraru Masoom A. Haider Meena AbdelMaseeh Nicola Rieke Pablo F. Damasceno Pedro Mario Cruz e Silva Pochuan Wang Sheng Xu Shuichi Kawano Sira Sriswasdi Soo Young Park Thomas M. Grist Varun Buch Watsamon Jantarabenjakul Weichung Wang Won Young Tak Xiang Li Xihong Lin Young Joon Kwon Abood Quraini Andrew Feng Andrew N. Priest Baris Turkbey Benjamin Glicksberg Bernardo Bizzo Byung Seok Kim Carlos Tor-Díez Chia-Cheng Lee Chia-Jung Hsu Chin Lin Chiu-Ling Lai Christopher P. Hess Colin Compas Deepeksha Bhatia Eric K. Oermann Evan Leibovitz Hisashi Sasaki Hitoshi Mori Isaac Yang Jae Ho Sohn Krishna Nand Keshava Murthy Li-Chen Fu Matheus Ribeiro Furtado de Mendonça Mike Fralick Min Kyu Kang Mohammad Adil Natalie Gangai Peerapon Vateekul Pierre Elnajjar Sarah Hickman Sharmila Majumdar Shelley L. McLeod Sheridan Reed Stefan Gräf Stephanie Harmon Tatsuya Kodama Thanyawee Puthanakit Tony Mazzulli Vitor Lima de Lavor Yothin Rakvongthai Yu Rim Lee Yuhong Wen Fiona J. Gilbert Mona G. Flores Quanzheng Li Ny mpanoratra: Izany dia Ny vadiny dia Holger R. Roth. Mifanohitra amin'i Zhong Amin'i Harouni Amilcare tsara tarehy Mpanoratra anarana anarana an'i Abidin Ny renivohiny Ny vadiny dia Anthony Beardsworth Costa. Ny vadiny dia Bradford J. Wood. Mifototra amin'ny tsirairay Ny Tompo dia Chih-Hung Wang Mifanohitra amin'ny Ny C. K. Lee Ny Rosiana Izany dia Ny Tompo dia Ny Tompo Mpanoratra anarana: Felipe Campos Kitamura Ao amin'ny Griffin Lacey Ny renivohiny dia Gustavo César de Antonio Corradi. Ny renivohiny Mifandraisa amin'ny Ny Hirofumi Obinata. Ny ren Sary avy amin'i Jason C. Crane. Ny renivohiny dia Ny Jiahui Guan Ny vadiny dia John W. Garrett. Toy ny hoe Joshua D. Kaggie. Ny tanànan'i Jung Gil Avy amin'i Dreyer Ny fitiavan'i Kristy Ny Kristy Kristy Marcio Aloisio Bezerra Cavalcanti ao amin'ny Rockenbach Marius George Linguraru amin'ny alalan'ny Mpanoratra anarana A. Haider Mifandraisa amin'ny Ny renivohiny Mpanoratra anarana: Pablo F. Damasceno Ny mpilalao dia Pedro Mario Cruz e Silva. Ny Tompo dia Wang Ny Tompo Toy ny Shuichi Kawano. Mifindra ao amin'ny Ny tanànan'i Soo Young Ny Tompo dia Grist Ny boky dia Ao anaty fitafiana vita amin'ny lamba Ny Tompo dia Wang Ny tanora dia tsy Ny renivohiny Ny renivohiny Ny tanora Joon Kwon Mifandraisa amin'ny Ny Tompo Feng Andrew N. Priest Ny tanànan'i Turkbey Ny vadiny dia Benjamin Glicksberg. Ny Tompo dia Bernardo Bizzo. Nitsidika an'i Kim Sary avy amin'i Carlos Tor-Díez. Ny renivohiny dia Chia-Cheng Lee. Ny Jiosy Jiosy Ny Chin Lin. Ny fahalalahana Sary avy amin'i Christopher P. Hess. Mifandraisa amin'ny Mifanohitra amin'i Bhatia Ny mpanoratra Eric K. Oermann. Ny renivohiny dia Leibovitz. Mifanohitra amin'i Sasaki Ny fahafatesana Izany dia Ny zanany dia Nand Keshava Murthy ary Kristy Nand Izany dia Matheus Ribeiro Furtado avy amin'ny Mendonça Avy amin'i Fralick Ny renivohiny dia Kang Ny fahamarinana Nandray anjara tamin'ny Mifototra amin'ny Mpanoratra anarana Ny vadiny dia Sarah Hickman. Mifanohitra amin'ny Sary avy amin'i Shelley L. McLeod. Ny renivohiny dia Ny Tompo Gräf Ny Harmony Tatsuya Kodama amin'izao fotoana izao Mifototra amin'ny Tany amin'ny Vitor Lima avy amin'ny asa Mifandray amin'ny Ny renivohiny dia Lee Rim. Ny Tompo Wen Sary avy amin'i Fiona J. Gilbert. Mona G. Ny voninkazo Ny fahamarinana Ny Abstract Ny federated learning setting (FL) dia fomba ampiasaina amin'ny fampiofanana ny tahirin-kevitra amin'ny tahirin-kevitra avy amin'ny loharanom-baovao maro ary mitazona ny anarana ny angon-drakitra, amin'izany ny fanesorana ny antony maro amin'ny famoahana angon-drakitra. Eto dia nampiasa angon-drakitra avy amin'ny oniversite 20 manerana izao tontolo izao izahay mba hampiofanana modely FL, antsoina hoe EXAM (Electronic Medical Record (EMR) chest X-ray AI model), izay manambara ny fangatahan'ny oksizenina ho avy amin'ireo marary marary marary miaraka amin'ny COVID-19 amin'ny fampiasana angon-drakitra momba ny zava-dehibe, angon-drakitra amin'ny laborato Main Ny fiaraha-monina ara-tsiansa, akademika, ara-pitsaboana ary siansa momba ny angon-drakitra dia niara-nifandimby teo anoloan'ny krizy COVID-19 mba hanombohana haingana ireo paradigma vaovao amin'ny fampahalalam-baovao ara-teknika (AI) izay haingana sy azo antoka, ary mety hamporisika ny fifanakalozan-kevitra sy ny fampiofanana sy ny fanandramana modely tsy misy ny fiarovana manokana sy ny fananana ny angon-drakitra amin'ny fiaraha-miasa mahazatra. , Ny mpamatsy ny fahasalamana, ny mpikaroka sy ny indostria dia mifantoka amin'ny fanatanterahana ny filàna tsy voatendry sy manan-danja nateraky ny krizy, amin'ny vokatra mahatalanjona. , , , , , , Ny fametrahana ny fitsapana ara-pitsaboana dia haingana sy mahomby amin'ny alalan'ny vatana ara-dalàna ao amin'ny firenena sy ny fanatanjahantena iraisam-pirenena. , , Ny fanadihadiana ny angon-drakitra sy ny AI dia mandrakariva nanatsara fomba fijerena misokatra sy miara-miasa, manangona hevitra toy ny rindrambaiko misokatra, fikarohana azo aseho, tahirin-kevitra ary ny fametrahana angon-drakitra anarana ho an'ny daholobe. , Ny aretina dia nanamafy ny ilaina hanatanterahana ny fiaraha-miasa amin'ny angon-drakitra izay ahafahan'ny fiaraha-monina ara-pitsaboana sy ara-tsiansa hamaly ny fanamby maneran-tany mihodina haingana sy be dia be. Ny fifanakalozana angon-drakitra dia manana fahasamihafana ara-tsiansa, ara-dalàna ary ara-dalàna izay voamarina, ary mety ho sarotra, amin'ny fidirana vao haingana amin'ny orinasa teknolojia lehibe ao amin'ny tontolo angon-drakitra ara-pahasalamana. , , . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Ity modely CDS ity dia novolavolaina tao amin'ny Mass General Brigham (MGB) ary voamarina tamin'ny angon-drakitra isan-karazany avy amin'ny rafitra ara-pahasalamana. Ny entana ho an'ny modely CDS dia sary X-ray (CXR) an-tendrombohitra, marika manan-danja, angon-drakitra demografika, ary valin'ny laboratoara izay hita tao amin'ny gazety taloha mba hanambara ny vokatra amin'ny marary COVID-19 , , , Ny CXR no voafidy ho toy ny fidirana amin'ny sary satria misy be dia be izany ary matetika no voalaza amin'ny fitsipika toy ny manome ny ACR. Ny fiarahamonina Fleischner Ny OMS dia Ny fiarahamonim-pirenena Thoracic , ministry ny fahasalamana nasionaly COVID handraisana sy ny radiology fiaraha-monina manerana izao tontolo izao . The output of the CDS model was a score, termed CORISK , izay mifanaraka amin'ny fepetra fanampiana oksizenina ary mety hanampy amin'ny fitsapana ny marary amin'ny mpitsabo voalohany , , Ny mpamatsy ny fahasalamana dia fantatra fa misafidy modely izay voamarina amin'ny angon-drakitra manokana. Hatramin'izao fotoana izao, ny ankamaroan'ny modely AI, anisan'izany ny modely CDS voalaza etsy ambony, dia efa nianatra sy nahazo fanamarinana amin'ny angon-drakitra "fehin-javatra" izay matetika tsy manana fahasamihafana. , , mety hiteraka overfitting sy ambany generalizability. Ity dia azo ampiharina amin'ny fampiofanana amin'ny angon-drakitra samihafa avy amin'ny tranonkala maromaro tsy misy centralization ny angon-drakitra mampiasa fomba fanao toy ny fampianarana , FL dia fomba ampiasain'ny fampiofanana ny modely AI amin'ny loharanom-baovao samihafa, raha tsy misy ny angon-drakitra alefa na voaporofo ivelan'ny toerana voalohany. Na dia azo ampiharina amin'ny indostria maro aza, ny FL dia novolavolaina vao haingana ho an'ny fikarohana ara-pahasalamana iraisam-pirenena. . 18 19 20 21 22 23 24 25 26 27 28 29 30 27 31 32 33 34 35 36 Federated Learning dia manohana ny fanombohana haingana amin'ny fanandramana miorina miaraka amin'ny fanatsarana ny famantarana ny angon-drakitra sy ny fanombanana ny fiovana algorithmic sy ny fiantraikany Ny fomba iray mifandraika amin'ny FL, antsoina hoe client-server, dia mandefa modely "tsy misy fampiofanana" amin'ny mpizara hafa ("nodes") izay mitondra asa fampiofanana ampahany, ary mitondra ny vokatra hiverina mba ho tafiditra ao amin'ny serivisy foibe ("federated"). . 37 36 Governance of data for FL is maintained locally, alleviating privacy concerns, with only model weights or gradients communicated between client sites and the federated server , FL efa naneho fanantenana tamin'ny fampiharana fitsaboana farany , , , Araka ny voalazan'ny COVID-19 , , Ohatra mahatalanjona dia modely famaritana ny fahafatesana amin'ny marary naratra amin'ny SARS-COV-2 izay mampiasa toetra ara-klinika, na dia voafetra aza amin'ny isan'ny fomba sy ny habeny. . 38 39 40 41 42 43 8 44 45 46 Ny tanjona dia ny mamorona modely mahery vaika, azo ampahibemaso izay afaka manampy amin'ny fanadihadiana ny marary. Nolazainay fa ny modely CDS dia azo mifandray amin'ny fahombiazana amin'ny fampiasana angon-drakitra izay mahazatra amin'ny fampiharana ara-klinika ary tsy miankina be amin'ny fanombanana miankina amin'ny mpampiasa ny toe-javatra ny marary (toy ny fahatsapana ara-klinika na ny aretina voalaza). Ankoatra izany, ny vokatra laboratoara, ny famantarana zava-dehibe, ny fanadihadiana momba ny sary sy ny fikarohana ankapobeny (tsy ny taona) dia nampiasaina. Noho izany dia nianatra ny modely CDS miaraka amin'ny Ny fanapahan-kevitra dia fa ny EXAM dia hanatanteraka tsara kokoa noho ny modely eo an-toerana ary hampivelatra tsara kokoa amin'ny sehatra ara-pahasalamana. Ny vokatra Ny modelin'ny fanadinana Ny modely EXAM dia mifototra amin'ny modely CDS voalaza etsy ambony Amin'ny ankapobeny, 20 endri-javatra (19 avy amin'ny EMR sy iray CXR) no ampiasaina ho toy ny fidirana amin'ny modely. Ny valin'ny vokatra (izany hoe 'ny marina') dia nomena mifototra amin'ny fitsaboana oxygen amin'ny marary taorian'ny vanim-potoana 24 sy 72 ora avy amin'ny fidirana voalohany tao amin'ny departemantan'ny fiarovana (ED). Ny lisitry ny endri-javatra sy ny vokatra nangatahana dia azo jerena ao amin'ny Tabilao . 27 1 Ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny valin'ny val Ho an'ny endri-javatra EMR, dia nampiasaina fotsiny ny lanjany voalohany voasambotra tao amin'ny ED ary ny fanomanana ny angon-drakitra dia ahitana ny fanamafisana, ny tsy fisian'ny lanjany, ary ny normalization amin'ny zero-mean sy ny singa variance. Noho izany, ny modely dia mampifandray vaovao avy amin'ny endri-javatra EMR sy CXR, mampiasa tambajotra neuronal convolutional 34-layer (ResNet34) mba hisambotra endri-javatra avy amin'ny CXR sy ny tambajotra Deep & Cross mba hampifandray ny endri-javatra miaraka amin'ny endri-javatra EMR (ho an'ny antsipirihany bebe kokoa, jereo Ny vokatry ny modely dia vokatry ny loza, antsoina hoe vokatry ny EXAM, izay lanjany tsy tapaka ao amin'ny sehatra 0-1 ho an'ny tsirairay amin'ireo fanambaran'ny 24 sy 72 ora mifanaraka amin'ireo marika voalaza etsy ambony. Ny fomba Ny federasiona ny modely Ny modely EXAM dia nianatra amin'ny fampiasana vondrona 16.148 tranga, izay mahatonga azy tsy ny iray amin'ireo modely FL voalohany ho an'ny COVID-19 fa koa tetikasa lehibe sy multi-kontinental fampandrosoana amin'ny zava-dehibe ara-klinika AI (Fig. Ny angon-drakitra eo amin'ny tranonkala dia tsy voamarina alohan'ny famokarana ary, amin'ny alalan'ny fiainana tena zava-misy momba ny informatika ara-pitsaboana toe-javatra, ny fanaraha-maso ny angon-drakitra fidirana dia tsy natao ny mpanoratra (Fig. Ny 1A ary B. 1C ary D , World map izay mampiseho ny 20 samihafa mpanjifa toerana mandray anjara amin'ny fanadihadiana EXAM. , Ny isan'ny tranga nanampy avy amin'ny orinasa na tranonkala tsirairay (client 1 dia maneho ny tranonkala manampy ny isan'ny tranga indrindra). , X-ray toradroa fampidirana hery amin'ny toerana tsirairay ny mpanjifa. , Age of patients at each client site, showing minimum and maximum ages (asterisks), mean age (triangles) and standard deviation (horizontal bars). The number of samples of each client site is shown in Supplementary Table . a b c d 1 Nandray anjara tamin'ny fandaharam-potoan'ny fanadihadiana momba ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana amin'ny alalan'ny fandaharam-potoana ( « 1 × 10–3, Wilcoxon sign-rank-test) amin'ny 16% ( araka ny voafaritra amin'ny AUC eo ho eo rehefa mihazakazaka ny modely amin'ny toeram-pitsaboana eo an-toerana: avy amin'ny 0,795 ka hatramin'ny 0,920, na 12.5 isan-jato) (Fig. Nisy ihany koa ny vokatra amin'ny 38% fanatsarana amin'ny famoaham-bahiny (izay voafaritra amin'ny AUC eo ho eo rehefa mihazakazaka ny modely amin'ny fanandramana rehetra: avy amin'ny 0.667 ka hatramin'ny 0.920, na 25.3 isan-jato) ny tsara indrindra ankapobeny modely ho an'ny famaritana ny fitsaboana oksizenina 24 ora raha oharina amin'ny modely nianatra fotsiny amin'ny angon-drakitra manokana (Fig. Ho an'ny vokatra faminaniana amin'ny fitsaboana oksizenina 72 ora, ny fampiofanana ny modely tsara indrindra eran-tany dia nahatonga ny fampitomboana eo ho eo amin'ny 18% raha oharina amin'ny modely nianatra any an-toerana, raha oharina amin'ny 34% ny fampitomboana ny modely maneran-tany (Extended Data Fig. Ny fahamarinan'ny vokatra dia voamarina amin'ny alalan'ny famerenana telo dingana amin'ny fampiofanana eo an-toerana sy FL amin'ny tahirin-kevitra samihafa. P Ny 2A Ny B 1 , Ny fahombiazana amin'ny fanandramana tsirairay amin'ny mpanjifa dia voatendry ny fitsaboana oksizenina 24 ora ho an'ny modely nianatra amin'ny angon-drakitra ao an-toerana ihany (Local) raha oharina amin'ny modely tsara indrindra manerantany azo jerena ao amin'ny mpizara (FL). , Generalizability (fanatanterahana eo ho eo amin'ny angon-drakitra fanandramana avy amin'ny tranonkala hafa, araka ny AUC eo ho eo) toy ny fiantraikany amin'ny habetsahan'ny angon-drakitra momba ny mpanjifa (tsy misy raharaha). Ny tsipika mainty dia maneho ny fanatanterahana ny fanatanterahana ny modely tsara indrindra manerantany. Ny vokatra ho an'ny 18 amin'ny 20 mpanjifa dia aseho, satria ny mpanjifa 12 dia nanana vokatra ho an'ny 72 ora oksizenina (Extended Data Fig. ) ary ny mpanjifa 14 dia nanana tranga amin'ny fitsaboana RA ihany, ka ny metrik'ny fanombanana (av. AUC) dia tsy azo ampiharina amin'ireo tranga ireo ( Ny angon-drakitra ho an'ny mpanjifa 14 ihany koa dia voafetra avy amin'ny fanombanana ny fanamafisana eo ho eo amin'ny modely eo an-toerana. a b 1 Methods Ny modely ao an-toerana izay voaomana amin'ny fampiasana cohorts tsy miovaova (ohatra, ny ankamaroan'ny tranga tsotra ny COVID-19) dia nahazo tombontsoa lehibe avy amin'ny FL fomba, amin'ny fanatsarana lehibe eo amin'ny famaritana eo amin'ny AUC fampisehoana ho an'ny sokajy amin'ny tranga vitsivitsy ihany. Izany dia hita ao amin'ny tranokalan'ny mpanjifa 16 (nitranga amin'ny dataset tsy miova), amin'ny ankamaroan'ny marary manana aretina matetika sy amin'ny tranga matetika vitsivitsy ihany. Ny modely FL dia nahatratra ambaratonga avo lenta-po ho an'ny tranga roa tsara (matetika) sy ny tahan'ny lainga-po ambany Ny angon-drakitra dia ampahany amin'ny Fig. Ny tena zava-dehibe dia ny fanitarana ny modely FL dia niakatra be noho ny modely voafehin'ny toerana. Ny 3A 2 , ROC ao amin'ny toerana mpanjifa 16, miaraka amin'ny angon-drakitra tsy miovaova sy ny ankamaroan'ny tranga madio. , ROC of the local model at client site 12 (a small dataset), mean ROC of models trained on larger datasets corresponding to the five client sites in the Boston area (1, 4, 5, 6, 8) and ROC of the best global model in prediction of 72-h oxygen treatment for different thresholds of EXAM score (left, middle, right). The mean ROC is calculated based on five locally trained models while the gray area denotes the ROC standard deviation. ROCs for three different cutoff values ( ) of the EXAM risk score are shown. Pos and neg denote the number of positive and negative cases, respectively, as defined by this range of EXAM score. a b t In the case of client sites with relatively small datasets, the best FL model markedly outperformed not only the local model but also those trained on larger datasets from five client sites in the Boston area of the USA (Fig. Ny Ny B Ny modely maneran-tany dia nahavita tsara amin'ny fanambaran'ny fangatahana oksizenina amin'ny 24 / 72 ora amin'ny marary COVID tsara sy ratsy (Extended Data Fig. ). 3 Validation at independent sites Taorian'ny fampiofanana voalohany, ny EXAM dia novolavolaina tao amin'ny toerana telo tsy miankina: Cooley Dickinson Hospital (CDH), Martha's Vineyard Hospital (MVH) ary Nantucket Cottage Hospital (NCH), rehetra any Massachusetts, Etazonia. Ny modely dia tsy voaomana indray tao amin'ireo toerana ireo ary ampiasaina fotsiny ho an'ny tanjona fanamarinana. , ary ny ROC curves sy confusion matrices ho an'ny lehibe indrindra dataset (tsy avy amin'ny CDH) dia hita ao amin'ny Fig. . The operating point was set to discriminate between nonmechanical ventilation and mechanical ventilation (MV) treatment (or death). The FL global trained model, EXAM, achieved an average AUC of 0.944 and 0.924 for 24- and 72-h prediction tasks, respectively (Table Ho an'ny fanambaran'ny fitsaboana MV (na fahafatesana) amin'ny 24 ora, ny EXAM dia nahatratra ny fahatsapana 0.950 sy ny fahatsapana 0.882 amin'ny CDH, ary ny fahatsapana 1.000 amin'ny fahatsapana 0.934 amin'ny MVH. Ny NCH dia tsy manana raharaha amin'ny MV / fahafatesana amin'ny 24 ora. Momba ny fahatsapana 72-ora, ny EXAM dia nahatratra fahatsapana 0.929 sy fahatsapana 0.880 amin'ny CDH, fahatsapana 1.000 sy fahatsapana 0.976 amin'ny MVH ary fahatsapana 1.000 sy fahatsapana 0.929 amin'ny NCH. 2 4 2 , , Performance (ROC) (top) sy confusion matrices (bottom) ny EXAM FL modely amin'ny CDH dataset ho an'ny fangatahana ny fangatahana oksizenina amin'ny 24 h ( ) and 72 h ( Ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ny fametrahana ( ) of the EXAM risk score are shown. a b a b t For MV at CDH at 72 h, EXAM had a low false-negative rate of 7.1%. Representative failure cases are presented in Extended Data Fig. , showing two false-negative cases from CDH where one case had many missing EMR data features and the other had a CXR with a motion artifact and some missing EMR features. 4 Use of differential privacy A primary motivation for healthcare institutes to use FL is to preserve the security and privacy of their data, as well as adherence to data compliance measures. For FL, there remains the potential risk of model ‘inversion’ na na dia ny famerenana ny fampiofanana sary avy amin'ny modely gradients . To counter these risks, security-enhancing measures were used to mitigate risk in the event of data ‘interception’ during site-server communication Nanandrana teknika mba hisorohana ny angon-drakitra FL izahay, ary nanampy fiarovana iray izay heverintsika fa mety hamporisika ny fikambanana maro kokoa hampiasa FL. Noho izany dia nanamarika ny vokatra teo aloha izay mampiseho fa ny fizarana ny habetsaky ny lanjany, ary ny teknika hafa amin'ny fiarovana ny fiainana manokana, dia azo ampiharina amin'ny FL. Amin'ny alalan'ny fanadihadiana ny fandaharam-potoana ampahany , , Naneho fa ny modely dia afaka mahatratra fampisehoana mitovy amin'izany na dia amin'ny 25% ihany aza ny fanavaozana ny lanjany (Extended Data Fig. Ny 47 48 49 50 50 51 52 5 Discussion Ity fianarana ity dia ahitana fianarana FL lehibe amin'ny sehatra ara-pahasalamana ara-toekarena eo amin'ny habetsaky ny toerana sy ny isan'ny daty toerana ampiasaintsika. Mino isika fa manome fanamafisana mahery vaika amin'ny fampiasana FL ho an'ny fampandrosoana haingana sy miara-miasa amin'ny modely AI ilaina amin'ny fahasalamana. Ny fianarana dia ahitana tranonkala maromaro manerana ny tany efatra sy eo ambany fanaraha-maso ny vatana samihafa, ary noho izany dia mitazona ny fampanantenana ny hanome ny tsena samihafa amin'ny fomba haingana. Ny modely FL maneran-tany, EXAM, dia nahita fa matanjaka kokoa ary nahatratra vokatra tsara amin'ny toerana Ho an'ny tranokalan'ny mpanjifa miaraka amin'ny angon-drakitra kely dia mety hampiasa fomba roa mahazatra ho an'ny modely mahasoa: ny iray dia ny fampiofanana eo an-toerana amin'ny angon-drakitra manokana, ny iray hafa dia ny fampiharana modely nampiofanana amin'ny daty lehibe kokoa. Ho an'ny tranokalan'ny angon-drakitra kely, dia mety tsy ho azo atao ny mamorona modely fampiofanana goavana mahomby amin'ny fampiasana ny angon-drakitra eo an-toerana ihany. Ny zava-misy fa ireo fomba roa ireo dia mihoatra amin'ny asa famaritana telo amin'ny modely FL maneran-tany, dia mampiseho fa ny tombontsoa ho an'ny tranokalan'ny mpanjifa amin' . 46 Ny vokatra fanamarinana dia nanamafy fa ny modely iraisam-pirenena dia matanjaka, manohana ny fanapahan-kevitra fa ny modely FL-trained dia azo ampiharina amin'ny sehatra ara-pahasalamana. Manome tranga mahatalanjona ho an'ny fampiasana algorithms predictive amin'ny fitsaboana ny marary COVID-19, ary ny fampiasana ny FL amin'ny famoronana modely sy ny fanandramana. Amin'ny fandraisana anjara amin'ity fianarana ity ny tranonkala mpanjifa nahazo fidirana amin'ny EXAM, ho fanamarinana bebe kokoa alohan'ny handray ny fankatoavana ara-dalàna na ny fidirana amin'ny fitsaboana ara-pahasalamana amin'ny hoavy. , as well as at different sites that were not a part of the EXAM training. 53 Over 200 prediction models to support decision-making in patients with COVID-19 have been published . Unlike the majority of publications focused on diagnosis of COVID-19 or prediction of mortality, we predicted oxygen requirements that have implications for patient management. We also used cases with unknown SARS-COV-2 status, and so the model could provide input to the physician ahead of receiving a result for PCR with reverse transcription (RT–PCR), making it useful for a real-life clinical setting. The model’s imaging input is used in common practice, in contrast with models that use chest computed tomography, a nonconsensual diagnostic modality. The model’s design was constrained to objective predictors, unlike many published studies that leveraged subjective clinical impressions. The data collected reflect varied incidence rates, and thus the ‘population momentum’ we encountered is more diverse. This implies that the algorithm can be useful in populations with different incidence rates. 19 Ny famantarana ny fitsaboana sy ny fanamafisana ny angon-drakitra dia tsy olana vaovao ao amin'ny fikarohana sy ny siansa momba ny angon-drakitra. Ny fanatsarana ny rafitra fampahalalana ara-pitsaboana dia ilaina mba hanatsarana ny fanomanana ny angon-drakitra, izay mitarika amin'ny fanatsarana tsara kokoa ny tambajotran'ny tranonkala mandray anjara amin'ny FL. Izany, miaraka amin'ny injeniera hyperparameter, dia ahafahan'ny algorithms 'mahalala' mahomby kokoa avy amin'ny angon-drakitra lehibe kokoa ary hampifanaraka ny parametra modely amin'ny tranonkala iray manokana ho an'ny fanamafisana bebe kokoa - ohatra, amin'ny alalan'ny fanamafisana bebe kokoa amin'izany tranonkala izany. Ny rafitra izay ahafahan'ny tsy manam-pahamehana, akaiky ny fotoana ara-toekarena modeling sy ny vokatra fanodinana dia ho tombony ihany koa ary 'mitsahatra ny lalana' avy amin'ny fampiofanana ny modely fametrahana. 54 39 Because data were not centralized they are not readily accessible. Given that, any future analysis of the results, beyond what was derived and collected, is limited. Tahaka ny modely hafa amin'ny fianarana milina, ny EXAM dia voafetra amin'ny kalitaon'ny angon-drakitra fampiofanana. Ny fikambanana tian'ny fampiharana io algorithm io ho an'ny fitsaboana ara-pitsaboana dia mila mahatakatra ny fahasamihafana mety amin'ny fampiofanana. Ohatra, ny marika ampiasaina ho toy ny fahamarinana fototra amin'ny fampiofanana ny modely EXAM dia avy amin'ny fampiasana oksizenina 24 sy 72 ora ao amin'ny marary; Mihevitra fa ny oksizenina atolotra amin'ny marary dia mitovy amin'ny zavatra ilaina amin'ny oksizenina. Na izany aza, tamin'ny vanim-potoana voalohany ny COVID-19 pandemia, marobe ireo marary Satria voafetra ny fidirana amin'ny angon-drakitra, tsy ampy ny fampahalalana azo jerena ho an'ny famoronana antontan-taratasy momba ny antony tsy fahombiazana, post-hoc, amin'ny ankamaroan'ny toerana. Na izany aza, dia nianatra ny tranga tsy fahombiazana avy amin'ny toerana tsy miankina lehibe indrindra, CDH, ary afaka mamorona fanapahan-kevitra izay azontsika atao amin'ny hoavy. Ho an'ny toerana mahomby, toa fa ny ankamaroan'ny tranga tsy fahombiazana dia ahitana iray amin'ireo sokajy roa: (1) ny kalitaon'ny angon-drakitra - ohatra, ny angon-drakitra tsy misy na ny fanatanjahan-tena ao amin'ny CXR; Amin'ny hoavy, dia mikasa ihany koa isika hanadihady ny mety hitranga amin'ny "fandosahan'ny mponina" noho ny dingana samihafa amin'ny fivoaran'ny aretina. Mino isika fa, noho ny fahasamihafana eo amin'ny toerana 20, mety ho voavonjy io loza io. Ny endri-javatra iray izay hanatsarana ireo karazana fiaraha-miasa amin'ny ankapobeny dia ny fahafahana mandinika ny fandraisana anjara avy amin'ny tranokalan'ny mpanjifa tsirairay amin'ny fanatsarana ny modely FL maneran-tany. Izany dia hanampy amin'ny fisafidianana ny tranokalan'ny mpanjifa, ary amin'ny fanapahan-kevitra ny fanangonana tahirin-kevitra sy ny fanamarihana ezaka. Ny fomba fiasa amin'ny hoavy dia mety hampiditra ny fikarohana ara-toekarena hyperparameter Ny fikarohana ny neural architecture and other automated machine learning fomba hahita ny fampiofanana tsara indrindra ho an'ny toerana tsirairay mpanjifa amin'ny fomba mahomby kokoa. 55 56 57 Known issues of batch normalization (BN) in FL motivated us to fix our base model for image feature extraction to reduce the divergence between unbalanced client sites. Future work might explore different types of normalization techniques to allow the training of AI models in FL more effectively when client data are nonindependent and identically distributed. 58 49 Recent works on privacy attacks within the FL setting have raised concerns on data leakage during model training . Meanwhile, protection algorithms remain underexplored and constrained by multiple factors. While differential privacy algorithms , , show good protection, they may weaken the model’s performance. Encryption algorithms, such as homomorphic encryption , maintain performance but may substantially increase message size and training time. A quantifiable way to measure privacy would allow better choices for deciding the minimal privacy parameters necessary while maintaining clinically acceptable performance , , . 59 36 48 49 60 36 48 49 Following further validation, we envision deployment of the EXAM model in the ED setting as a way to evaluate risk at both the per-patient and population level, and to provide clinicians with an additional reference point when making the frequently difficult task of triaging patients. We also envision using the model as a more sensitive population-level metric to help balance resources between regions, hospitals and departments. Our hope is that similar FL efforts can break the data silos and allow for faster development of much-needed AI models in the near future. Ny fomba Ethics approval Ny fandaharam-pahalalana rehetra ao amin'ny Institut Canyland Health Maronsin dia novolavolaina araka ny fitsipika momba ny fanandramana olombelona araka ny voalaza ao amin'ny Deklaration of Helsinki and International Conference on Harmonization Good Clinical Health Practice Guidelines, ary nankatoavin'ny Beth Newungner Hospital amin'ny fitsaboana momba ny fitsaboana ao amin'ny toerana manaraka ireto: CDH, MVH, NCH ary ao amin'ny tranonkala fampiofanana manaraka ireto: MGB, Mass General Hospital (MGH), Brigham and Women's Hospital, Newton-Wellesley Hospital, San Shore Public Medical Center, New Faulkner Hospital (tsy an-tapitrisany ny fitsaboana momba ny fitsaboana ao amin'ny fitsaboana ao amin'ny fitsabo Ny fitsipika mifanaraka amin'ny MI-CLAIM momba ny fampahafantarana ny modely ara-pitsaboana amin'ny AI (Supplementary Note) ) 2 Ny fikarohana The study included data from 20 institutions (Fig. ): MGB, MGH, Brigham and Women's Hospital, Newton-Wellesley Hospital, North Shore Medical Center sy Faulkner Hospital; Children's National Hospital in Washington, DC; NIHR Cambridge Biomedical Research Centre; The Self-Defense Forces Central Hospital in Tokyo; National Taiwan University MeDA Lab sy MAHC ary Taiwan National Health Insurance Administration; Tri-Service General Hospital in Taiwan; Kyungpook National University Hospital in Korea; Faculty of Medicine, Chulalongkorn University in Thailand; Diagnosticos da America SA in Brazil; University of California, San Francisco; VA San Diego; University of Toronto; National Institutes of Health in Bethesda, Maryland; University of Wisconsin-Madison School of Medicine and Public Health; Memorial Sloan Kettering Cancer Center in New York; ary Mount Sinai Health System , , . Data from three independent sites were used for independent validation: CDH, MVH and NCH, all in Massachusetts, USA. These three hospitals had patient population characteristics different from the training sites. The data used for the algorithm validation consisted of patients admitted to the ED at these sites between March 2020 and February 2021, and that satisfied the same inclusion criteria of the data used to train the FL model. Ny 1A 61 62 63 Ny fanangonana Ny tranonkala 20 mpanjifa dia nanomana total 16148 tranga (tsy misy fiantraikany na tsy misy fiantraikany) ho an'ny tanjona ny fampiofanana, ny fanamarinana sy ny fitsapana ny modely (Fig. ). Medical data were accessed in relation to patients who satisfied the study inclusion criteria. Client sites strived to include all COVID-positive cases from the beginning of the pandemic in December 2019 and up to the time they started local training for the EXAM study. All local training had started by 30 September 2020. The sites also included other patients in the same period with negative RT–PCR test results. Since most of the sites had more SARS-COV-2-negative than -positive patients, we limited the number of negative patients included to, at most, 95% of the total cases at each client site. 1b A ‘case’ included a CXR and the requisite data inputs taken from the patient’s medical record. A breakdown of the cohort size of the dataset for each client site is shown in Fig. Ny fampielezan-kevitra sy ny endrika amin'ny endrika CXR (pixel values) dia niova be eo amin'ny toerana noho ny isan-karazany amin'ny mpitsabo sy ny toerana manokana, toy ny mpamokatra fitaovana samihafa sy ny famolavolana protocols, araka ny hita ao amin'ny sary. . Patient age and EMR feature distribution varied greatly among sites, as expected owing to the differing demographics between globally distributed hospitals (Extended Data Fig. Ny 1b 1C ary D 6 Ny fitsipika momba ny fidirana amin'ny marary Ny fitsipika momba ny fidirana amin'ny marary dia: (1) ny marary nomena ny ED ao amin'ny hopitaly na mifanaraka amin'izany; (2) ny marary dia nanao fanadinana RT-PCR na oviana na oviana eo anelanelan'ny fidirana amin'ny ED sy ny fidirana avy ao amin'ny hopitaly; (3) ny marary dia manana CXR ao amin'ny ED; ary (4) ny rakitra marary dia manana farafaharatsiny dimy amin'ireo valin'ny EMR voalaza ao amin'ny Tabilao. , all obtained in the ED, and the relevant outcomes captured during hospitalization. Of note, The CXR, laboratory results and vitals used were the first available for capture during the visit to the ED. The model did not incorporate any CXR, laboratory results or vitals acquired after leaving the ED. 1 Model input Amin'ny ankapobeny, 21 endri-javatra EMR no ampiasaina ho toy ny fidirana amin'ny modely. Ny valin'ny vokatra (izany hoe, marina ny tany) dia nomena mifototra amin'ny fepetra ilain'ny marary taorian'ny 24 sy 72 ora vanim-potoana avy amin'ny fidirana voalohany amin'ny ED. Ny lisitry ny endri-javatra EMR nangatahana sy ny vokatra azo jerena ao amin'ny Tabilao . 1 The distribution of oxygen treatment using different devices at different client sites is shown in Extended Data Fig. , izay mamaritra ny fampiasana ny fitaovana amin'ny fidirana amin'ny ED sy aorian'ny vanim-potoana 24 sy 72 ora. Ny fahasamihafana eo amin'ny fizarana dataset eo amin'ny lehibe indrindra sy ny kely indrindra mpanjifa toerana dia azo jerena ao amin'ny Extended Data Fig. . 7 8 Ny isan'ny tranga COVID-19 tsara, araka ny fanamafisana amin'ny fitsapana RT-PCR tsirairay azo avy amin'ny fotoana rehetra eo anelanelan'ny fandefasana ny ED sy ny famotsorana avy amin'ny hopitaly, dia voasoratra ao amin'ny Tabilao fanampiny. Ny tranokalan'ny mpanjifa tsirairay dia nangataka ny hametraka ny angon-drakitra ao amin'ny ampahany telo: 70% ho an'ny fampiofanana, 10% ho an'ny fanamarinana ary 20% ho an'ny fitsapana. 1 Fanandramana ny modely fanadinana Misy fahasamihafana goavana amin'ny fitsaboana ara-pitsaboana amin'ny marary izay mitsidika ny hopitaly miaraka amin'ny aretina COVID-19, miaraka amin'ny sasany amin'ny fiantraikany haingana amin'ny asa fanadiovana izay mitaky fanafody samihafa mba hisorohana na hampihenana ny hypoxemia. , Ny fanapahan-kevitra manan-danja natao nandritra ny fanombanana ny marary ao amin'ny toerana voalohany amin'ny fitsaboana, na ao amin'ny ED, dia raha mety ho ilaina ny marary amin'ny fanafody na ny fanafody (ohatra MV na monoklonal antibodies), ary noho izany dia tokony handray fitsaboana kely fa mahomby, fitsaboana miaraka amin'ny fiantraikany ambany noho ny fiantraikany ara-pahasalamana noho ny fiantraikany, na ny ambaratonga avo lenta amin'ny fitsaboana, toy ny fidirana ao amin'ny toeram-pitsaboana. Raha oharina amin'izany, ny marary izay manana risika ambany kokoa amin'ny fangatahana fitsaboana oksizenina dia mety ho voasambotra ao amin'ny toeram-pitsaboana tsy dia matanjaka toy ny toeram-pitsaboana mahazatra, na dia voavonjy avy amin'ny ED ho an'ny fanaraha-maso hatrany ao an-trano. Ny EXAM dia novolavolaina mba hanampy amin'ny fanodinana ireo marary ireo. 62 63 64 65 Noho izany, ny modely dia tsy nahazo fankatoavana avy amin'ny orinasam-pandraharahana amin'izao fotoana izao ary tokony hampiasaina fotsiny ho an'ny tanjona fikarohana. Ny fanadinana dia Ny EXAM dia nianatra amin'ny fampiasana FL; dia mitarika vokatry ny loza (amin'ny teny hoe vokatry ny EXAM) mitovy amin'ny CORISK (Ny angon-drakitra azo avy amin'ny angon-drakitra. Izany dia mifanaraka amin'ny zavatra ilain'ny marary amin'ny fanampiana oksizenina ao anatin'ny varavarankely roa - 24 sy 72 ora - aorian'ny fampidirana voalohany amin'ny ED. mampiseho ny fomba ampiasain'i CORISK sy ny valin'ny EXAM amin'ny fanodinana ny marary. 27 Ny faha-9A 9b Ny sary X-ray boribory dia novolavolaina alohan'ny hisafidy ny sary eo anoloan'ny toerana ary tsy ahitana ny sary fijerena ankavanana, ary avy eo nohamasinina ho amin'ny fanapahan-kevitra 224 × 224. , ny modely dia mampifandray ny fampahalalana avy amin'ny endri-javatra EMR sy CXR (miorina amin'ny ResNet34 navoakan'ny toerana pretrained on the CheXpert dataset) Ary ny tambajotra Deep & Cross Mba hampifandray ireo karazana angon-drakitra samihafa ireo, ny vektor 512-dimensiona dia navoaka avy amin'ny sary CXR tsirairay amin'ny fampiasana ResNet34, miaraka amin'ny fifantohana ara-piarahamonina, ary avy eo mifandray amin'ny endri-javatra EMR ho toy ny fidirana ho an'ny tambajotra Deep & Cross. Ny fidirana farany dia isa tsy tapaka ao amin'ny sehatra 0-1 ho an'ny fanambaran'ny 24 sy 72 ora, mifanaraka amin'ny marika voalaza etsy ambony, araka ny hita ao amin'ny Data Extended Fig. Nampiasa ny cross-entropy ho toy ny fahaverezan'ny asa sy ny "Adam" ho toy ny optimizer. Amin'ny fampiasana ny NVIDIA Clara Train SDK Ny AUC ankapobeny ho an'ny asa famaritana (≥LFO, ≥HFO/NIV na ≥MV) dia novolavolaina ary ampiasaina ho toy ny famaritana farany, miaraka amin'ny normalization amin'ny zero-mean-javatra sy ny singa variance. Ny 9a 66 67 68 Ny B 69 70 27 Ny fametrahana sy ny normalization Ny algorithm dia Nampiasaina mba hanamarinana ny endri-javatra EMR, mifototra amin'ny angon-drakitra fampiofanana ao an-toerana. Raha tsy misy endri-javatra EMR amin'ny angon-drakitra ao amin'ny tranokalan'ny mpanjifa, dia nampiasaina ny vidin'izany endri-javatra izany, voafantina avy amin'ny angon-drakitra avy amin'ny tranokalan'ny mpanjifa MGB. 71 Details of EMR–CXR data fusion using the Deep & Cross network To model the interactions of features from EMR and CXR data at the case level, a deep-feature scheme was used based on a Deep & Cross network architecture Ny endri-javatra binary sy categorical ho an'ny fampidirana EMR, ary koa ny endri-javatra 512-dimensional ao amin'ny CXR, dia niova ho vektoro matanjaka amin'ny lanjany tena amin'ny alàlan'ny fametrahana sy fametrahana ny loharano. Ny endri-javatra matanjaka transformed dia nanompo ho toy ny fidirana amin'ny rafitra fametrahana, izay mampiasa tambajotra fidirana manokana mba hampiharana ny fidirana eo amin'ny fidirana avy amin'ny loharano samihafa. Ny tambajotra fidirana dia nanao fidirana manokana ao anatin'ny loharano, amin'ny alàlan'ny vokatra anatiny eo amin'ny endri-javatra fidirana 68 Ny antsipirihany Azo lazaina fa ny endrika tsara indrindra amin'ny FL dia ny fampiharana ny algorithm federated averaging araka ny voalazan'i McMahan et al. ary Salmona niteraka an'i Boaza tamin-dRahaba ; ary Boaza niteraka an'i Obeda tamin-dRota ; ary Obeda niteraka an'i Jese ; ). 72 9c A pseudoalgorithm of FL is shown in Supplementary Note Ao amin'ny fanandramana, dia nametraka ny isan'ny federated rounds amin'ny = 200, miaraka amin'ny fotoana fampiofanana iray eo an-toerana isaky ny dingana ny isan'ny mpanjifa. ny isan'ny mpanjifa, , was up to 20 depending on the network connectivity of clients or available data for a specific targeted outcome period (24 or 72 h). The number of local training iterations, , depends on the dataset size at each client and is used to weigh each client’s contributions when aggregating the model weights in federated averaging. During the FL training task, each client site selects its best local model by tracking the model’s performance on its local validation set. At the same time, the server determines the best global model based on the average validation scores sent from each client site to the server after each FL round. After FL training finishes, the best local models and the best global model are automatically shared with all client sites and evaluated on their local test data. 1 T t K Ny k When training on local data only (the baseline), we set the epoch number to 200. The Adam optimizer was used for both local training and FL with an initial learning rate of 5 × 10–5 and a stepwise learning rate decay with a factor 0.5 after every 40 epochs, which is important for the convergence of federated averaging Ny famolavolana ara-tsosialy, anisan'izany ny rotation, ny translation, ny fanapahana, ny fanapahana ary ny feo sy ny fifindran'ny hery ara-tsosialy, dia ampiharina amin'ny sary mba hampitomboana ny angon-drakitra nandritra ny fampiofanana. 73 Noho ny fahatsapana ny BN layers Rehefa mandray anjara amin'ny mpanjifa samihafa amin'ny sehatra tsy miankina sy mitazona mitovy, dia hitantsika fa ny vokatra tsara indrindra amin'ny modely dia nitranga rehefa mitazona ny ResNet34 amin'ny fijerena ny toerana. Ny tambajotra Deep & Cross izay mampifandray ny endri-javatra miaraka amin'ny endri-javatra EMR dia tsy ahitana tabataba BN ary noho izany dia tsy voahelingelina amin'ny olana amin'ny tsy fandriampahalemana BN. 58 47 Ao amin'ity fianarana ity dia nanadihady rafitra fiarovana ny fiainana manokana izay mizara fanavaozana modely ampahany ihany eo amin'ny mpizara sy ny tranonkala mpanjifa. Ny fanavaozana ny lanjany dia voasoratra ao anatin'ny herinandro tsirairay araka ny habetsahan'ny fandraisana anjara, ary ny isan-jaton'ny fanavaozana ny lanjany lehibe indrindra ihany no hizara amin'ny mpizara. (t) (Extended Data Fig. ), which was computed from all non-zero gradients, Δ , ary mety ho hafa ho an'ny mpanjifa tsirairay Ao anaty fitafiana vita amin'ny lamba. Ny fahasamihafana amin'ity rafitra ity dia mety ahitana fanapahan-kevitra fanampiny amin'ny gradients lehibe na rafitra fiarovana fahasamihafana. that add random noise to the gradients, or even to the raw data, before feeding into the network . k 5 Ny tsirairay (t) k t 49 51 Ny fanadihadiana statistika Nanao fitsapana voasoratra anarana Wilcoxon izahay mba hanamafisana ny fiantraikany amin'ny fanatsarana hita eo amin'ny fampisehoana eo amin'ny modely voaomana amin'ny toerana sy ny modely FL amin'ny fotoana 24 sy 72 ora (Fig. Ny angon-drakitra dia ampahany amin'ny Fig. ). The null hypothesis was rejected with one-sided « 1 × 10–3 in both cases. 2 1 P Nandray anjara tamin'ny fandaharam-potoan'ny fanadihadiana momba ny fahasamihafana eo amin'ny sehatry ny fampahalalam-baovao amin'ny sehatry ny fampahalalam-baovao momba ny fahasamihafana eo amin'ny sehatry ny fampahalalam-baovao amin'ny sehatry ny fampahalalam-baovao ( Ary ny 0.43 = 0.035, degrees of freedom (df) = 17 for the 24-h model and Ary ny 62. = 0.003, df = 16 ho an'ny modely 72-h). Izany dia mampiseho fa ny habetsaky ny dataset dia tsy ny singa iray izay mamaritra ny fitovian'ny modely amin'ny angon-drakitra tsy hita. r P r P Mba hanara-maso ny fivoaran'ny ROC avy amin'ny modely FL maneran-tany sy modely eo an-toerana nianatra tao amin'ny toerana samihafa (Extended Data Fig. ), we bootstrapped 1,000 samples from the data and computed the resulting AUCs. We then calculated the difference between the two series and standardized using the formula = (AUC1 – AUC2)/ , where is the standardized difference, is the standard deviation of the bootstrap differences and AUC1 and AUC2 are the corresponding bootstrapped AUC series. By comparing Amin'ny fizarana mahazatra, dia nahazo ny values illustrated in Supplementary Table . The results show that the null hypothesis was rejected with very low values, indicating the statistical significance of the superiority of FL outcomes. The computation of Ny valiny dia nandeha tamin'ny R miaraka amin'ny pROC library. . 3 D s D s D P 2 P P 74 Since the model predicts a discrete outcome, a continuous score from 0 to 1, a straightforward calibration evaluation such as a qqplot is not possible. Hence, for a quantified estimate of calibration we quantified discrimination (Extended Data Fig. Nanao fanadihadiana amin'ny endrika iray (ANOVA) izahay mba hampitaha ny valin'ny modely ao an-toerana sy FL eo amin'ny sokajy efatra (RA, LFO, HFO, MV). -statistic, calculated as the variation between the sample means divided by variation within the samples and representing the degree of dispersion among different groups, was used to quantify the models. Our results show that the -Ny lanjany amin'ny toerana dimy samihafa dia 245.7, 253.4, 342.3, 389.8 ary 634.8, raha ny amin'ny modely FL dia 843.5. -values mean that groups are more separable, the scores from our FL model clearly show a greater dispersion among the four ground truth categories. Furthermore, the value of the ANOVA test on the FL model is <2 × 10–16, indicating that the FL prediction scores are statistically significantly different among the different prediction classes. 10 F F F P Ny famoahana ny tatitra Ny fampahalalana fanampiny momba ny fikarohana dia azo jerena ao amin'ny linked to this article. Fampahalalana momba ny fikarohana momba ny natiora Ny fidirana amin'ny data Ireo angon-drakitra ireo dia nampiasaina ho an'ny fampiofanana tao amin'ny tsirairay amin'ireo tranonkala eo an-toerana ary tsy nandray anjara tamin'ny fikambanana hafa na amin'ny mpizara federated, ary tsy azo jerena ho an'ny daholobe. Ny angon-drakitra avy amin'ireo tranonkala fanamarinana tsy miankina dia voatahiry amin'ny CAMCA, ary ny fidirana dia azo mangataka amin'ny mifandray amin'ny Q.L. Miorina amin'ny fanapahan-kevitra amin'ny CAMCA, ny fanaraha-maso sy ny fanitsiana ny IRB ho an'ny tanjona fikarohana dia azo atao amin'ny fitantanana ny fikarohana MGB ary mifanaraka amin'ny MGB IRB sy ny politika. Fidirana amin'ny code Ny code sy ny rindrambaiko rehetra ampiasaina amin'ity fikarohana ity dia azo aseho amin'ny NGC. Mba hahazoana ny fidirana, hidirana amin'ny mpitsidika na mamorona mombamomba dia alefa ao amin'ny iray amin'ireo URL etsy ambany. Ny modely nianatra, ny fitsipika momba ny famolavolana angona, ny code ho an'ny fampiofanana, ny fanamarinana ny modely, ny rakitra readme, ny fitsipika fametrahana ary ny rakitra fahazoan-dalana dia azo jerena amin'ny NVIDIA NGC. Ny : Ny rindrambaiko federated fampianarana dia azo jerena ao anatin'ny Clara Train SDK: Na oviana na oviana, mampiasa ity baiko ity mba hamerina ny modely "wget --content-disposition" -O clara_train_covid19_exam_ehr_xray_1.zip”. 61 https://ngc.nvidia.com/catalog/models/nvidia:med:clara_train_covid19_exam_ehr_xray https://ngc.nvidia.com/catalog/containers/nvidia:clara-train-sdk https://api.ngc.nvidia.com/v2/models/nvidia/med/clara_train_covid19_exam_ehr_xray/versions/1/zip Ny References Budd, J. et al. Digital technologies in the public-health response to COVID-19. , 1183–1192 (2020). Nat. Med. 26 Moorthy, V., Henao Restrepo, A. M., Preziosi, M.-P. & Swaminathan, S. Data sharing for novel coronavirus (COVID-19). , 150 (2020). Bull. World Health Organ. 98 Chen, Q., Allot, A. & Lu, Z. Keep up with the latest coronavirus research. , 193 (2020). Nature 579 Fabbri, F., Bhatia, A., Mayer, A., Schlotter, B. & Kaiser, J. BCG IT spend pulse: how COVID-19 is shifting tech priorities. (2020). https://www.bcg.com/publications/2020/how-covid-19-is-shifting-big-it-spend Candelon, F., Reichert, T., Duranton, S., di Carlo, R. C. & De Bondt, M. The rise of the AI-powered company in the postcrisis world. (2020). https://www.bcg.com/en-gb/publications/2020/business-applications-artificial-intelligence-post-covid Chao, H. et al. Integrative analysis for COVID-19 patient outcome prediction. , 101844 (2021). Med. Image Anal. 67 Zhu, X. et al. Joint prediction and time estimation of COVID-19 developing severe symptoms using chest CT scan. , 101824 (2021). Med. Image Anal. 67 Yang, D. et al. Federated semi-supervised learning for Covid region segmentation in chest ct using multi-national data from China, Italy, Japan. , 101992 (2021). Med. Image Anal. 70 Minaee, S., Kafieh, R., Sonka, M., Yazdani, S. & Jamalipour Soufi, G. Deep-COVID: predicting COVID-19 from chest X-ray images using deep transfer learning. , 101794 (2020). Med. Image Anal. 65 COVID-19 Studies from the World Health Organization Database. (2020). https://clinicaltrials.gov/ct2/who_table ACTIV. (2020). https://www.nih.gov/research-training/medical-research-initiatives/activ Coronavirus Treatment Acceleration Program (CTAP). US Food and Drug Administration (2020). https://www.fda.gov/drugs/coronavirus-covid-19-drugs/coronavirus-treatment-acceleration-program-ctap Gleeson, P., Davison, A. P., Silver, R. A. & Ascoli, G. A. A commitment to open source in neuroscience. , 964–965 (2017). Neuron 96 Piwowar, H. et al. The state of OA: a large-scale analysis of the prevalence and impact of open access articles. , e4375 (2018). PeerJ. 6 European Society of Radiology (ESR). What the radiologist should know about artificial intelligence – an ESR white paper. , 44 (2019). Insights Imaging 10 Pesapane, F., Codari, M. & Sardanelli, F. Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine. , 35 (2018). Eur. Radiol. Exp. 2 Price, W. N. 2nd & Cohen, I. G. Privacy in the age of medical big data. , 37–43 (2019). Nat. Med. 25 Liang, W. et al. Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19. , 1081–1089 (2020). JAMA Intern. Med. 180 Wynants, L. et al. Prediction models for diagnosis and prognosis of covid-19 infection: systematic review and critical appraisal. , m1328 (2020). Brit. Med. J. 369 Zhang, L. et al. D-dimer levels on admission to predict in-hospital mortality in patients with Covid-19. , 1324–1329 (2020). J. Thromb. Haemost. 18 Sands, K. E. et al. Patient characteristics and admitting vital signs associated with coronavirus disease 2019 (COVID-19)-related mortality among patients admitted with noncritical illness. (2020). https://doi.org/10.1017/ice.2020.461 American College of Radiology. CR recommendations for the use of chest radiography and computed tomography (CT) for suspected COVID-19 infection. (2020). https://www.acr.org/Advocacy-and-Economics/ACR-Position-Statements/Recommendations-for-Chest-Radiography-and-CT-for-Suspected-COVID19-Infection Rubin, G. D. et al. The role of chest imaging in patient management during the COVID-19 pandemic: a multinational consensus statement from the Fleischner Society. , 172–180 (2020). Radiology 296 World Health Organization. Use of chest imaging in COVID-19. (2020). https://www.who.int/publications/i/item/use-of-chest-imaging-in-covid-19 Jamil, S. et al. Diagnosis and management of COVID-19 disease. , 10 (2020). Am. J. Respir. Crit. Care Med. 201 Redmond, C. E., Nicolaou, S., Berger, F. H., Sheikh, A. M. & Patlas, M. N. Emergency radiology during the COVID-19 pandemic: The Canadian Association of Radiologists Recommendations for Practice. , 425–430 (2020). Can. Assoc. Radiologists J. 71 Buch, V. et al. Development and validation of a deep learning model for prediction of severe outcomes in suspected COVID-19 Infection. Preprint at (2021). https://arxiv.org/abs/2103.11269 Lyons, C. & Callaghan, M. The use of high-flow nasal oxygen in COVID-19. , 843–847 (2020). Anaesthesia 75 Whittle, J. S., Pavlov, I., Sacchetti, A. D., Atwood, C. & Rosenberg, M. S. Respiratory support for adult patients with COVID-19. , 95–101 (2020). J. Am. Coll. Emerg. Physicians Open 1 Ai, J., Li, Y., Zhou, X. & Zhang, W. COVID-19: treating and managing severe cases. , 370–371 (2020). Cell Res. 30 Esteva, A. et al. A guide to deep learning in healthcare. , 24–29 (2019). Nat. Med. 25 Cahan, E. M., Hernandez-Boussard, T., Thadaney-Israni, S. & Rubin, D. L. Putting the data before the algorithm in big data addressing personalized healthcare. , 78 (2019). NPJ Digit. Med. 2 Thrall, J. H. et al. Artificial intelligence and machine learning in radiology: opportunities, challenges, pitfalls, and criteria for success. , 504–508 (2018). J. Am. Coll. Radiol. 15 Shilo, S., Rossman, H. & Segal, E. Axes of a revolution: challenges and promises of big data in healthcare. , 29–38 (2020). Nat. Med. 26 Gao, Y. & Cui, Y. Deep transfer learning for reducing health care disparities arising from biomedical data inequality. , 5131 (2020). Nat. Commun. 11 Rieke, N. et al. The future of digital health with federated learning. , 119 (2020). NPJ Dig. Med. 3 Yang, Q., Liu, Y., Chen, T. & Tong, Y. Federated machine learning: concept and applications. , 12 (2019). ACM Trans. Intell. Syst. Technol. 10 Ma, C. et al. On safeguarding privacy and security in the framework of federated learning. , 242–248 (2020). IEEE Netw. 34 Brisimi, T. S. et al. Federated learning of predictive models from federated Electronic Health Records. , 59–67 (2018). Int. J. Med. Inform. 112 Roth, H. R. et al. Federated learning for breast density classification: a real-world implementation. In , (eds. Albarqouni, S. et al.) Vol. 12,444, 181–191 (Springer International Publishing, 2020). Proc. Second MICCAI Workshop, DART 2020 and First MICCAI Workshop, DCL 2020 Domain Adaptation and Representation Transfer, and Distributed and Collaborative Learning Sheller, M. J. et al. Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data. , 12598 (2020). Sci. Rep. 10 Remedios, S. W., Butman, J. A., Landman, B. A. & Pham, D. L. in (eds Remedios, S. W. et al.) (Springer, 2020). Federated Gradient Averaging for Multi-Site Training with Momentum-Based Optimizers Xu, Y. et al. A collaborative online AI engine for CT-based COVID-19 diagnosis. Preprint at (2020). https://www.medrxiv.org/content/10.1101/2020.05.10.20096073v2 Raisaro, J. L. et al. SCOR: A secure international informatics infrastructure to investigate COVID-19. , 1721–1726 (2020). J. Am. Med. Inform. Assoc. 27 Vaid, A. et al. Federated learning of electronic health records to improve mortality prediction in hospitalized patients with COVID-19: machine learning approach. , e24207 (2021). JMIR Med. Inform. 9 Nino, G. et al. Pediatric lung imaging features of COVID-19: a systematic review and meta-analysis. , 252–263 (2021). Pediatr. Pulmonol. 56 Fredrikson, M., Jha, S. & Ristenpart, T. Model inversion attacks that exploit confidence information and basic countermeasures. In 1322–1333, (2015). Proc. 22nd ACM SIGSAC Conference on Computer and Communications Security https://doi.org/10.1145/2810103.2813677 Zhu, L., Liu, Z. & Han, S. in (eds Wallach, H. et al.) 14774–14784 (Curran Associates, Inc., 2019). Advances in Neural Information Processing Systems 32 Kaissis, G. A., Makowski, M. R., Rückert, D. & Braren, R. F. Secure, privacy-preserving and federated machine learning in medical imaging. , 305–311 (2020). Nat. Mach. Intell. 2 Li, W. et al. in 133–141 (Springer, 2019). Privacy-Preserving Federated Brain Tumour Segmentation Shokri, R. & Shmatikov, V. Privacy-preserving deep learning. In (2015). Proc. 53rd Annual Allerton Conference on Communication, Control, and Computing (Allerton) https://doi.org/10.1109/allerton.2015.7447103 Li, X. et al. Multi-site fMRI analysis using privacy-preserving federated learning and domain adaptation: ABIDE results. , 101765 (2020). Med. Image Anal. 65 Estiri, H. et al. Predicting COVID-19 mortality with electronic medical records. , 15 (2021). NPJ Dig. Med. 4 Jiang, G. et al. Harmonization of detailed clinical models with clinical study data standards. , 65–74 (2015). Methods Inf. Med. 54 Yang, D. et al. in . (2019). Searching Learning Strategy with Reinforcement Learning for 3D Medical Image Segmentation https://doi.org/10.1007/978-3-030-32245-8_1 Elsken, T., Metzen, J. H. & Hutter, F. Neural architecture search: a survey. , 1–21 (2019). J. Mach. Learning Res. 20 Yao, Q. et al. Taking human out of learning applications: a survey on automated machine learning. Preprint at (2019). https://arxiv.org/abs/1810.13306 Ioffe, S. & Szegedy, C. Batch normalization: accelerating deep network training by reducing internal covariate shift. In , PMLR , 448–456 (2015). Proc. 32nd International Conf. Machine Learning 37 Kaufman, S., Rosset, S. & Perlich, C. Leakage in data mining: formulation, detection, and avoidance. In , 556–563 (2011). Proc. 17th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining Zhang, C. et al. BatchCrypt: efficient homomorphic encryption for cross-silo federated learning. In , 493–506 (2020). Proc. 2020 USENIX Annual Technical Conference, ATC 2020 . (2020). Nvidia NGC Catalog: COVID-19 Related Models https://ngc.nvidia.com/catalog/models?orderBy=scoreDESC&pageNumber=0&query=covid&quickFilter=models&filters Marini, J. J. & Gattinoni, L. Management of COVID-19 respiratory distress. , 2329–2330 (2020). JAMA 323 Cook, T. M. et al. Consensus guidelines for managing the airway in patients with COVID-19: Guidelines from the Difficult Airway Society, the Association of Anaesthetists the Intensive Care Society, the Faculty of Intensive Care Medicine and the Royal College of Anaesthetist. , 785–799 (2020). Anaesthesia 75 Galloway, J. B. et al. A clinical risk score to identify patients with COVID-19 at high risk of critical care admission or death: an observational cohort study. , 282–288 (2020). J. Infect. 81 Kilaru, A. S. et al. Return hospital admissions among 1419 COVID-19 patients discharged from five U.S. emergency departments. , 1039–1042 (2020). Acad. Emerg. Med. 27 He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. In (2016). Proc. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) https://doi.org/10.1109/cvpr.2016.90 Irvin, J. et al. CheXpert: a large chest radiograph dataset with uncertainty labels and expert comparison. , 590–597 (2019). Proc. AAAI Conf. Artif. Intell. 33 Wang, R., Fu, B., Fu, G. & Wang, M. Deep & Cross network for Ad Click predictions. In Article no. 12 (2017). Proc. ADKDD’17 Abadi, M. et al. TensorFlow: asystem for large-scale machine learning. In , USENIX Association 265–283 (2016). 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16) . (2020). NVIDIA Clara Imaging https://developer.nvidia.com/clara-medical-imaging Stekhoven, D. J. & Bühlmann, P. MissForest–non-parametric missing value imputation for mixed-type data. , 112–118 (2012). Bioinformatics 28 McMahan, H., Moore, E., Ramage, D., Hampson, S. & y Arcas, B. A. Communication-efficient learning of deep networks from decentralized data. (2017). http://proceedings.mlr.press/v54/mcmahan17a.html Hsieh, K., Phanishayee, A., Mutlu, O. & Gibbons, P. B. The non-IID data quagmire of decentralized machine learning. In PMLR 119 (2020). Proc. 37th International Conf. Machine Learning Robin, X. et al. pROC: an open-source package for R and S+ to analyze and compare ROC curves. , 77 (2011). BMC Bioinformatics 12 Acknowledgements MGB misaotra ireo olona manaraka ireto noho ny fanohanany: J. Brink, Department of Radiology, Massachusetts General Hospital, Boston, MA; N. Guo, Center for Advanced Medical Computing and Analysis, Massachusetts General Hospital, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Harvard Medical School, Boston, MA; MA; J. K. Cramer, Center for Clinical Data Science, Massachusetts General Brigham, Boston, MA; T. Schultz, Department of Radiology, Massachusetts General Hospital, Boston, MA; N. Guo, Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA; MA; J. K. Cramer, Director, QTIM Research, ity lab Harvard A. Martinos Center for Biomical Imaging amin'ny MGH; S. Pom Amin'ny alalan'ny Faculty of Medicine, Chulalongkorn University dia misaotra ny Ratchadapisek Sompoch Endowment Fund RA (PO) (no. 001/63) noho ny fanangonana sy ny fitantanana ny angon-drakitra ara-pitsaboana mifandraika amin'ny COVID-19 sy ny sampana ara-biolojika ho an'ny Research Task Force, Faculty of Medicine, Chulalongkorn University. NIHR Cambridge Biomedical Research Center misaotra A. Priest, izay manohana ny NIHR (Cambridge Biomedical Research Centre at Cambridge University Hospitals NHS Foundation Trust). National Taiwan University MeDA Lab sy ny MAHC sy Taiwan National Health Insurance Administration misaotra ny MOST Joint Research Center for AI technology, ny All Vista Healthcare National Health Insurance Administration, https://data.ucsf.edu/covid19 Ity lahatsoratra ity dia azo jerena amin'ny ankapobeny amin'ny lisansa CC by 4.0 Deed (Attribution 4.0 International). Ity lahatsoratra ity dia azo jerena amin'ny ankapobeny amin'ny lisansa CC by 4.0 Deed (Attribution 4.0 International).