Kirjoittajat : Ittai Dayan 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 Jääkiekko Ruan Jääkiekko Xu Dufan Wu Eddie Huang Felipe Campos Kitamura Griffin Lacey Gustavo César de Antônio Corradi Gustavo Nino Hao-Hsin Shin Hirofumi Obinata Hui Ren Pääosat Jason C. Crane Jesse Tetreault Jiahui Guan John W. Garrett Joshua D. Kaggie Jung Gil Park Keith Dreyer Krishna Juluru Kristopher Kersten Pääosat Marcio Aloisio Bezerra Cavalcanti Rockenbach Marius George Linguraru Masoom A. Haider Meena AbdelMaseeh Nicola Rieke Pablo F. Damasceno Pääosat Pedro Mario Cruz e Silva Jääkiekko Wang Sheng Xu Shuichi Kawano Sira Särkänen Soo Young Park Thomas M. Grist Varun Buch Watsamon Jantarabenjakul Weichung Wang Won Young Tak Tiina 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 Kolin Kompas Deepeksha Bhatia Eric K. Oermann Pääosat 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 Kirjoittajat : Ittai Dayan Pääosat Holger R. Roth Aoxiao Zhong Ahmed Harouni Amilcare ystävällinen Anas Z. Abidin Andrew Liu Pääosat Anthony Beardsworth Costa Pääosat Bradford J. Wood Mäntsäläinen Tsai Pääosat Chih-Hung Wang Chun-Nan Hsu Pääosat C. K. Lee Jääkiekko Ruan Jääkiekko Xu Dufan Wu Eddie Huang Pääosat Felipe Campos Kitamura Pääosat Griffin Lacey Gustavo César de Antônio Corradi Gustavo Nino Hao-Hsin Shin Hirofumi ObinataMuokkaa Hui Ren Pääosat Jason C. Crane Pääosat Jesse Tetreault Jiahui Guan Pääosat John W. Garrett Joshua D. Kaggie Jung Gil -puisto Pääosat Keith Dreyer Krishna Juluru Kristinusko keräsi Pääosat Marcio Aloisio Bezerra Cavalcanti Rockenbach Pääosat Marius George Linguraru Masoom A. Haider Meena AbdelMaseeh Niko Riekko Pääosat Pablo F. Damasceno Pääosat Pedro Mario Cruz e Silva Jääkiekko Wang Sheng Xu Pääosat Shuichi Kawano Sira Särkänen Soo Young Park Thomas M. Grist Varun kirja Watsamon Jantarabenjakul Säätiö Wang Won Young Tak Tiina Li Jääkiekko Lin Nuori Joon Kwon Abood Quraini Andrew Feng Andrew N. Priest Baris Turkbey Benjamin Glicksberg Pääosat Bernardo Bizzo Byung Seok Kim Carlos Tor-Díez Chia-Cheng Lee Chia-Jung Hsu Kiinalainen Lin Chiu-Ling Lai Pääosat Christopher P. Hess Kolin Kompas Deepeksha BhatiaMuokkaa Pääosat Eric K. Oermann Pääosat Evan Leibovitz Hisashi Sasaki Hitoshi Mori Isaac Yang Jae Ho Sohn Krishna Nand Keshava Murthy Lähde: Li-Chen Fu Matheus Ribeiro Furtado de Mendonça Pääosat Mike Fralick Min Kyu Kang Mohammad Adil Natalie Gangai Pyöräilijä Vateekul Pierre Elnajjar Sarah Hickman Sharmila Majumdar Pääosat Shelley L. McLeod Sheridan Reed Stefan Gräf Stephanie Harmon Tatsuya Kodama Hämäläinen Puthanakit Tony Mazzulli Vitor Lima de Lavor Yothin Rakvongthai Yu Rim Lee Yuhong Wen Pääosat Fiona J. Gilbert Mona G. Kukkia Quanzheng Li Abstract Federated learning setting (FL) on menetelmä, jota käytetään kouluttamaan keinotekoisen älykkyyden malleja useista lähteistä saatavilla tiedoilla säilyttäen samalla tietojen nimettömyyden, mikä poistaa monia esteitä tietojen jakamiselle. Täällä käytimme tietoja 20 instituutista ympäri maailmaa kouluttamaan FL-mallia, jota kutsutaan EXAMiksi (electronic medical record (EMR) chest X-ray AI-malliksi), joka ennustaa COVID-19-oireisten potilaiden tulevia hapen tarpeita käyttämällä elintärkeiden merkkien syöttöjä, laboratoriotietoja ja rintakehän röntgensäteitä. EXAM saavutti keskimääräisen alueen kaaren alla (AUC) > 0,92 ennustettaessa tuloksia 24 tunnissa ja 72 tunnissa ensimmäisestä Main The scientific, academic, medical and data science communities have come together in the face of the COVID-19 pandemic crisis to rapidly assess novel paradigms in artificial intelligence (AI) that are rapid and secure, and potentially incentivize data sharing and model training and testing without the usual privacy and data ownership hurdles of conventional collaborations , . Healthcare providers, researchers and industry have pivoted their focus to address unmet and critical clinical needs created by the crisis, with remarkable results , , , , , , . Clinical trial recruitment has been expedited and facilitated by national regulatory bodies and an international cooperative spirit , , Tietojen analysointi ja tekoälyn tieteenalat ovat aina edistäneet avoimia ja yhteistoiminnallisia lähestymistapoja, jotka käsittelevät sellaisia käsitteitä kuin avoimen lähdekoodin ohjelmistot, toistettavissa oleva tutkimus, tietovarastot ja anonymisoidut tietokokonaisuudet julkisesti saatavilla. , Pandemia on korostanut tarvetta nopeuttaa datayhteistyötä, joka antaa kliinisille ja tieteellisille yhteisöille mahdollisuuden vastata nopeasti kehittyviin ja laajalle levinneisiin globaaleihin haasteisiin. , , . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Tämä CDS-malli kehitettiin Mass General Brighamissa (MGB) ja se validoitiin useiden terveydenhuoltojärjestelmien tietojen kautta. CDS-mallin syötteet olivat rintakehän röntgenkuvia (CXR), elintärkeitä merkkejä, väestötietoja ja laboratorioarvoja, jotka osoitettiin aiemmissa julkaisuissa ennustavan COVID-19-potilaiden tuloksia , , , . CXR was selected as the imaging input because it is widely available and commonly indicated by guidelines such as those provided by ACR Fleischnerin yhteiskunta ja WHO , national thoracic societies , kansallinen terveysministeriö COVID käsikirjat ja radiologian yhteisöt ympäri maailmaa . The output of the CDS model was a score, termed CORISK , that corresponds to oxygen support requirements and that could aid in triaging patients by frontline clinicians , , . Healthcare providers have been known to prefer models that were validated on their own data . To date most AI models, including the aforementioned CDS model, have been trained and validated on ‘narrow’ data that often lack diversity , , potentially resulting in overfitting and lower generalizability. This can be mitigated by training with diverse data from multiple sites without centralization of data using methods such as transfer learning , or FL. FL is a method used to train AI models on disparate data sources, without the data being transported or exposed outside their original location. While applicable to many industries, FL has recently been proposed for cross-institutional healthcare research . 18 19 20 21 22 23 24 25 26 27 28 29 30 27 31 32 33 34 35 36 Federated learning supports the rapid launch of centrally orchestrated experiments with improved traceability of data and assessment of algorithmic changes and impact . One approach to FL, called client-server, sends an ‘untrained’ model to other servers (‘nodes’) that conduct partial training tasks, in turn sending the results back to be merged in the central (‘federated’) server. This is conducted as an iterative process until training is complete . 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 has already shown promise in recent medical imaging applications , , , , including in COVID-19 analysis , , . A notable example is a mortality prediction model in patients infected with SARS-COV-2 that uses clinical features, albeit limited in terms of number of modalities and scale . 38 39 40 41 42 43 8 44 45 46 Tavoitteenamme oli kehittää vankka, yleistettävissä oleva malli, joka voisi auttaa potilaiden seulonnassa. Teorioimme, että CDS-mallia voidaan yhdistää menestyksekkäästi, kun otetaan huomioon sen kliinisessä käytännössä suhteellisen yleisten tietojen syöttötietojen käyttö ja jotka eivät riipu suuresti potilaan tilan operaattorista riippuvaisista arvioinneista (kuten kliinisistä vaikutelmista tai raportoiduista oireista). Sen sijaan käytettiin laboratoriotuloksia, elintärkeitä merkkejä, kuvantamistutkimusta ja yleisesti kerättyä väestörakennetta (eli ikää). Siksi koulutettiin CDS-mallia uudelleen erilaisilla tiedoilla käyttäen asiakaspalvelimen FL-lähestymistapaa uuden globaalin FL-mallin kehittämiseksi, joka nimettiin EXAM Our hypothesis was that EXAM would perform better than local models and would generalize better across healthcare systems. tulokset The EXAM model architecture EXAM-malli perustuu edellä mainittuun CDS-malliin . In total, 20 features (19 from the EMR and one CXR) were used as input to the model. The outcome (that is, ‘ground truth’) labels were assigned based on patient oxygen therapy after 24- and 72-hour periods from initial admission to the emergency department (ED). A detailed list of the requested features and outcomes can be seen in Table . 27 1 The outcome labels of patients were set to 0, 0.25, 0.50 and 0.75 depending on the most intensive oxygen therapy the patient received in the prediction window. The oxygen therapy categories were, respectively, room air (RA), low-flow oxygen (LFO), high-flow oxygen (HFO)/noninvasive ventilation (NIV) or mechanical ventilation (MV). If the patient died within the prediction window, the outcome label was set to 1. This resulted in each case being assigned two labels in the range 0–1, corresponding to each of the prediction windows (that is, 24 and 72 h). EMR-ominaisuuksien osalta käytettiin vain ED:ssä tallennettuja ensimmäisiä arvoja ja tietojen esikäsittelyyn sisältyivät deidentifiointi, puuttuvan arvon imputaatio ja normalisointi nollan keskiarvoon ja yksikön vaihteluun. Näin ollen malli yhdistää tietoja sekä EMR- että CXR-ominaisuuksista käyttämällä 34-kerroksista konvolutionaalista hermoverkkoa (ResNet34) poimimaan ominaisuuksia CXR- ja Deep & Cross-verkosta ominaisuuksien yhdistämiseksi EMR-ominaisuuksien kanssa (lisätietoja on ) Mallin tulos on riskipiste, jota kutsutaan EXAM-pisteeksi, joka on jatkuva arvo alueella 0–1 kullekin 24 ja 72 tunnin ennusteelle, joka vastaa edellä kuvattuja etikettejä. Methods Mallin liittoutuminen The EXAM model was trained using a cohort of 16,148 cases, making it not only among the first FL models for COVID-19 but also a very large and multicontinent development project in clinically relevant AI (Fig. ). Data between sites were not harmonized before extraction and, in light of real-life clinical informatics circumstances, a meticulous harmonization of the data input was not conducted by the authors (Fig. ). 1a,b 1c,d , World map indicating the 20 different client sites contributing to the EXAM study. , Number of cases contributed by each institution or site (client 1 represents the site contributing the largest number of cases). , Chest X-ray intensity distribution at each client site. , 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 Olemme vertailleet paikallisesti koulutettuja malleja globaaliin FL-malliin kunkin asiakkaan testaustietojen perusteella. « 1 × 10–3, Wilcoxon signed-rank test) of 16% (as defined by average AUC when running the model on respective local test sets: from 0.795 to 0.920, or 12.5 percentage points) (Fig. ). It also resulted in 38% generalizability improvement (as defined by average AUC when running the model on all test sets: from 0.667 to 0.920, or 25.3 percentage points) of the best global model for prediction of 24-h oxygen treatment compared with models trained only on a site’s own data (Fig. ). For the prediction results of 72-h oxygen treatment, the best global model training resulted in an average performance improvement of 18% compared to locally trained models, while generalizability of the global model improved on average by 34% (Extended Data Fig. Tulosten vakaus vahvistettiin toistamalla kolme paikallista ja FL-koulutusta eri satunnaistetuilla datasplitteillä. P 2a 2b 1 , Performance on each client’s test set in prediction of 24-h oxygen treatment for models trained on local data only (Local) versus that of the best global model available on the server (FL (gl. best)). Av., average test performance across all sites. , Yleistettävyys (toisten sivustojen testitietojen keskimääräinen suorituskyky, kuten keskimääräinen AUC) asiakkaan tietokokonaisuuden koon (ei tapauksia). Vihreä vaakasuora viiva osoittaa parhaan globaalin mallin yleistettävyyden suorituskykyä. ) and client 14 had cases only with RA treatment, such that the evaluation metric (av. AUC) was not applicable in either of these cases ( ). Data for client 14 were also excluded from computation of average generalizability in local models. a b 1 Methods Local models that were trained using unbalanced cohorts (for example, mostly mild cases of COVID-19) markedly benefited from the FL approach, with a substantial improvement in prediction average AUC performance for categories with only a few cases. This was evident at client site 16 (an unbalanced dataset), with most patients experiencing mild disease severity and with only a few severe cases. The FL model achieved a higher true-positive rate for the two positive (severe) cases and a markedly lower false-positive rate compared to the local model, both shown in the receiver operating characteristic (ROC) plots and confusion matrices (Fig. and Extended Data Fig. ). More important, the generalizability of the FL model was considerably increased over the locally trained model. 3a 2 , ROC at client site 16, with unbalanced data and mostly mild cases. , 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 ( Pos ja neg osoittavat positiivisten ja negatiivisten tapausten määrän vastaavasti, sellaisena kuin se on määritelty tällä EXAM-pistemäärällä. 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. ) on 3b The global model performed well in predicting oxygen needs at 24/72 h in patients both COVID positive and negative (Extended Data Fig. ). 3 Validation at independent sites Ensimmäisen koulutuksen jälkeen EXAM testattiin myöhemmin kolmella riippumattomalla validointipaikalla: Cooley Dickinson Hospital (CDH), Martha’s Vineyard Hospital (MVH) ja Nantucket Cottage Hospital (NCH), kaikki Massachusettsissa, Yhdysvalloissa. , and the ROC curves and confusion matrices for the largest dataset (from CDH) are shown in Fig. Toimintapiste asetettiin erottamaan mekaanisen ilmanvaihdon ja mekaanisen ilmanvaihdon (MV) hoidon (tai kuoleman) välillä. FL:n maailmanlaajuinen koulutettu malli, EXAM, saavutti keskimääräisen AUC-arvon 0,944 ja 0,924 24 ja 72 tunnin ennakointitehtävissä (taulukko). ), which exceeded the average performance among sites used in training EXAM. For prediction of MV treatment (or death) at 24 h, EXAM achieved a sensitivity of 0.950 and specificity of 0.882 at CDH, and a sensitivity of 1.000 specificity of 0.934 at MVH. NCH did not have any cases with MV/death at 24 h. In regard to 72-h MV prediction, EXAM achieved a sensitivity of 0.929 and specificity of 0.880 at CDH, sensitivity of 1.000 and specificity of 0.976 at MVH and sensitivity of 1.000 and specificity of 0.929 at NCH. 2 4 2 , , Suorituskyky (ROC) (ylhäältä) ja sekavuusmatriisit (alla) EXAM FL -mallin CDH-tietokannassa hapen kysynnän ennustamiseksi 24 h ( ) and 72 h ( ). ROCs for three different cutoff values ( Tutkimuksen riskiarvioinnin tulokset on esitetty. 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 Yksityisyyden eriyttäminen 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’ or even the reconstruction of training images from the model gradients themselves . To counter these risks, security-enhancing measures were used to mitigate risk in the event of data ‘interception’ during site-server communication . We experimented with techniques to avoid interception of FL data, and added a security feature that we believe could encourage more institutions to use FL. We thus validated previous findings showing that partial weight sharing, and other differential privacy techniques, can successfully be applied in FL . Through investigation of a partial weight-sharing scheme , , , we showed that models can reach a comparable performance even when only 25% of weight updates are shared (Extended Data Fig. ). 47 48 49 50 50 51 52 5 Discussion This study features a large, real-world healthcare FL study in terms of number of sites and number of data points used. We believe that it provides a powerful proof-of-concept of the feasibility of using FL for fast and collaborative development of needed AI models in healthcare. Our study involved multiple sites across four continents and under the oversight of different regulatory bodies, and thus holds the promise of being provided to different regulated markets in an expedited way. The global FL model, EXAM, proved to be more robust and achieved better results at individual sites than any model trained on only local data. We believe that consistent improvement was achieved owing to a larger, but also a more diverse, dataset, the use of data inputs that can be standardized and avoidance of clinical impressions/reported symptoms. These factors played an important part in increasing the benefits from this FL approach and its impact on performance, generalizability and, ultimately, the model’s usability. For a client site with a relatively small dataset, two typical approaches could be used for fitting a useful model: one is to train locally with its own data, the other is to apply a model trained on a larger dataset. For sites with small datasets, it would have been virtually impossible to build a performant deep learning model using only their local data. The finding, that these two approaches were outperformed on all three prediction tasks by the global FL model, indicates that the benefit for client sites with small datasets arising from participation in FL collaborations is substantial. This is probaby a reflection of FL’s ability to capture more diversity than local training, and to mitigate the bias present in models trained on a homogenous population. An under-represented population or age group in one hospital/region might be highly represented in another region—such as children who might be differentially affected by COVID-19, including disease manifestations in lung imaging . 46 The validation results confirmed that the global model is robust, supporting our hypothesis that FL-trained models are generalizable across healthcare systems. They provide a compelling case for the use of predictive algorithms in COVID-19 patient care, and the use of FL in model creation and testing. By participating in this study the client sites received access to EXAM, to be further validated ahead of pursuing any regulatory approval or future introduction into clinical care. Plans are under way to validate EXAM prospectively in ‘production’ settings at MGB leveraging COVID-19 targeted resources , 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 Patient cohort identification and data harmonization are not novel issues in research and data science , but are further complicated, when using FL, given the lack of visibility on other sites’ datasets. Improvements to clinical information systems are needed to streamline data preparation, leading to better leverage of a network of sites participating in FL. This, in conjunction with hyperparameter engineering, can allow algorithms to ‘learn’ more effectively from larger data batches and adapt model parameters to a particular site for further personalization—for example, through further fine-tuning on that site . A system that would allow seamless, close-to real-time model inference and results processing would also be of benefit and would ‘close the loop’ from training to model deployment. 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. Similar to other machine learning models, EXAM is limited by the quality of the training data. Institutions interested in deploying this algorithm for clinical care need to understand potential biases in the training. For example, the labels used as ground truth in the training of the EXAM model were derived from 24- and 72-h oxygen consumption in the patient; it is assumed that oxygen delivered to the patient equates the oxygen need. However, in the early phase of the COVID-19 pandemic, many patients were provided high-flow oxygen prophylactically regardless of their oxygen need. Such clinical practice could skew the predictions made by this model. Since our data access was limited, we did not have sufficient available information for the generation of detailed statistics regarding failure causes, post hoc, at most sites. However, we did study failure cases from the largest independent test site, CDH, and were able to generate hypotheses that we can test in the future. For high-performing sites, it seems that most failure cases fall into one of two categories: (1) low quality of input data—for example, missing data or motion artifact in CXR; or (2) out-of-distribution data—for example a very young patient. In future, we also intend to investigate the potential for a ‘population drift’ due to different phases of disease progression. We believe that, owing to the diversity across the 20 sites, this risk may have been mitigated. Yksi ominaisuus, joka parantaisi tällaista laajamittaista yhteistyötä, on kyky ennustaa kunkin asiakassivuston panos globaalin FL-mallin parantamiseen.Tämä auttaa asiakassivuston valinnassa ja tietojen hankkimisen ja merkintöjen priorisoinnissa. Future approaches may incorporate automated hyperparameter searching , neural architecture search and other automated machine learning lähestymistapoja löytää optimaaliset koulutusparametrit kullekin asiakassivustolle tehokkaammin. 55 56 57 FL: n erän normalisoinnin (BN) tunnetut ongelmat motivoi meitä korjaamaan perusmallimme kuvan ominaisuuksien uuttamiseksi 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 Samaan aikaan suojausalgoritmeja ei ole tutkittu ja niitä rajoittavat useat tekijät. , , 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. Methods Eettinen hyväksyntä All procedures were conducted in accordance with the principles for human experimentation as defined in the Declaration of Helsinki and International Conference on Harmonization Good Clinical Practice guidelines, and were approved by the relevant institutional review boards at the following validation sites: CDH, MVH, NCH and at the following training sites: MGB, Mass General Hospital (MGH), Brigham and Women’s Hospital, Newton-Wellesley Hospital, North Shore Medical Center and Faulkner Hospital (all eight of these hospitals were covered under MGB’s ethics board reference, no. 2020P002673, and informed consent was waived by the instititional review board (IRB). Similarly, participation of the remaining sites was approved by their respective relevant institutional review processes: Children’s National Hospital in Washington, DC (no. 00014310, IRB certified exempt); NIHR Cambridge Biomedical Research Centre (no. 20/SW/0140, informed consent waived); The Self-Defense Forces Central Hospital in Tokyo (no. 02-014, informed consent waived); National Taiwan University MeDA Lab and MAHC and Taiwan National Health Insurance Administration (no. 202108026 W, informed consent waived); Tri-Service General Hospital in Taiwan (no. B202105136, informed consent waived); Kyungpook National University Hospital in South Korea (no. KNUH 2020-05-022, informed consent waived); Faculty of Medicine, Chulalongkorn University in Thailand (nos. 490/63, 291/63, informed consent waived); Diagnosticos da America SA in Brazil (no. 26118819.3.0000.5505, informed consent waived); University of California, San Francisco (no. 20-30447, informed consent waived); VA San Diego (no. H200086, IRB certified exempt); University of Toronto (no. 20-0162-C, informed consent waived); National Institutes of Health in Bethesda, Maryland (no. 12-CC-0075, informed consent waived); University of Wisconsin-Madison School of Medicine and Public Health (no. 2016-0418, informed consent waived); Memorial Sloan Kettering Cancer Center in New York (no. 20-194, informed consent waived); and Mount Sinai Health System in New York (no. IRB-20-03271, informed consent waived). MI-CLAIM guidelines for reporting of clinical AI models were followed (Supplementary Note ) 2 Study setting Tutkimuksessa oli mukana 20 tutkimuslaitosta (kuva 4). ): MGB, MGH, Brigham and Women’s Hospital, Newton-Wellesley Hospital, North Shore Medical Center and 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 and MAHC and Taiwan National Health Insurance Administration; Tri-Service General Hospital in Taiwan; Kyungpook National University Hospital in South 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; and Mount Sinai Health System in New York. Institutions were recruited between March and May 2020. Dataset curation started in June 2020 and the final data cohort was added in September 2020. Between August and October 2020, 140 independent FL runs were conducted to develop the EXAM model and, by the end of October 2020, EXAM was made public on NVIDIA NGC , , Kolmen riippumattoman sivuston tietoja käytettiin riippumattomaan validointiin: CDH, MVH ja NCH, kaikki Massachusettsin osavaltiossa Yhdysvalloissa. Näillä kolmella sairaalalla oli erilaiset potilasryhmän ominaisuudet kuin koulutuspaikoilla. Algoritmin validointiin käytetyt tiedot koostuivat potilaista, jotka hyväksyttiin ED:ään näillä sivustoilla maaliskuun 2020 ja helmikuun 2021 välillä, ja jotka täyttävät samat sisällyttämiskriteerit kuin FL-mallin kouluttamiseen käytetyt tiedot. 1a 61 62 63 Data collection The 20 client sites prepared a total of 16,148 cases (both positive and negative) for the purposes of training, validation and testing of the model (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. . The distribution and patterns of CXR image intensity (pixel values) varied greatly among sites owing to a multitude of patient- and site-specific factors, such as different device manufacturers and imaging protocols, as shown in Fig. . Patient age and EMR feature distribution varied greatly among sites, as expected owing to the differing demographics between globally distributed hospitals (Extended Data Fig. ) on 1b 1 C ja D 6 Patient inclusion criteria Patient inclusion criteria were: (1) patient presented to the hospital’s ED or equivalent; (2) patient had a RT–PCR test performed at any time between presentation to the ED and discharge from the hospital; (3) patient had a CXR in the ED; and (4) patient’s record had at least five of the EMR values detailed in Table , 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 Sisääntulon malli In total, 21 EMR features were used as input to the model. The outcome (that is, ground truth) labels were assigned based on patient requirements after 24- and 72-h periods from initial admission to the ED. A detailed list of the requested EMR features and outcomes can be seen in Table . 1 The distribution of oxygen treatment using different devices at different client sites is shown in Extended Data Fig. , which details the device usage at admission to the ED and after 24- and 72-h periods. The difference in dataset distribution between the largest and smallest client sites can be seen in Extended Data Fig. . 7 8 The number of positive COVID-19 cases, as confirmed by a single RT–PCR test obtained at any time between presentation to the ED and discharge from the hospital, is listed in Supplementary Table . Each client site was asked to randomly split its dataset into three parts: 70% for training, 10% for validation and 20% for testing. For both 24- and 72-h outcome prediction models, random splits for each of the three repeated local and FL training and evaluation experiments were independently generated. 1 EXAM model development Potilaiden kliinisessä kurssissa, joilla on COVID-19-oireita, on laaja vaihtelu, ja jotkut kokevat nopean hengityselinten toiminnan heikkenemisen, joka vaatii erilaisia toimenpiteitä hypoksemian ehkäisemiseksi tai lieventämiseksi. , . A critical decision made during the evaluation of a patient at the initial point of care, or in the ED, is whether the patient is likely to require more invasive or resource-limited countermeasures or interventions (such as MV or monoclonal antibodies), and should therefore receive a scarce but effective therapy, a therapy with a narrow risk–benefit ratio due to side effects or a higher level of care, such as admittance to the intensive care unit . In contrast, a patient who is at lower risk of requiring invasive oxygen therapy may be placed in a less intensive care setting such as a regular ward, or even released from the ED for continuing self-monitoring at home EXAM kehitettiin auttamaan tällaisten potilaiden lajittelua. 62 63 64 65 Of note, the model is not approved by any regulatory agency at this time and it should be used only for research purposes. Testien tulos EXAM was trained using FL; it outputs a risk score (termed EXAM score) similar to CORISK (Extended Data Fig. ) and can be used in the same way to triage patients. It corresponds to a patient’s oxygen support requirements within two windows—24 and 72 h—after initial presentation to the ED. Extended Data Fig. Näytetään, miten CORISK- ja EXAM-pisteitä voidaan käyttää potilaiden lajittelussa. 27 9a 9b Chest X-ray images were preprocessed to select the anterior position image and exclude lateral view images, and then scaled to a resolution of 224 × 224. As shown in Extended Data Fig. , the model fuses information from both EMR and CXR features (based on a modified ResNet34 with spatial attention esikoulutettu CheXpert-tietokannassa) Deep & Cross -verkosto . To converge these different data types, a 512-dimensional feature vector was extracted from each CXR image using a pretrained ResNet34, with spatial attention, then concatenated with the EMR features as the input for the Deep & Cross network. The final output was a continuous value in the range 0–1 for both 24- and 72-h predictions, corresponding to the labels described above, as shown in Extended Data Fig. . We used cross-entropy as the loss function and ‘Adam’ as the optimizer. The model was implemented in Tensorflow using the NVIDIA Clara Train SDK . The average AUC for the classification tasks (≥LFO, ≥HFO/NIV or ≥MV) was calculated and used as the final evaluation metric, with normalization to zero-mean and unit variance. CXR images were preprocessed to select the correct series and exclude lateral view images, then scaled to a resolution of 224 × 224 (ref. ) on 9a 66 67 68 9b 69 70 27 Feature imputation and normalization A MissForest algorithm Jos EMR-ominaisuus puuttui kokonaan asiakassivuston tietokannasta, käytettiin kyseisen ominaisuuden keskiarvoa, joka laskettiin yksinomaan MGB-asiakassivustojen tietoihin. 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 . Binary and categorical features for the EMR inputs, as well as 512-dimensional image features in the CXR, were transformed into fused dense vectors of real values by embedding and stacking layers. The transformed dense vectors served as input to the fusion framework, which specifically employed a crossing network to enforce fusion among input from different sources. The crossing network performed explicit feature crossing within its layers, by conducting inner products between the original input feature and output from the previous layer, thus increasing the degree of interaction across features. At the same time, two individual classic deep neural networks with several stacked, fully connected feed-forward layers were trained. The final output of our framework was then derived from the concatenation of both classic and crossing networks. 68 FL yksityiskohdat Arguably the most established form of FL is implemention of the federated averaging algorithm as proposed by McMahan et al. , tai sen muunnelmia. Tämä algoritmi voidaan toteuttaa käyttämällä asiakas-palvelin-asennusta, jossa jokainen osallistuva sivusto toimii asiakkaana. Voidaan ajatella FL: tä menetelmänä, jolla pyritään minimoimaan globaali menetysfunktio vähentämällä kussakin sivustossa arvioitua joukkoa paikallisia menetysfunktioita. Pienentämällä kunkin asiakassivuston paikallista menetystä sekä synkronoimalla oppitunutta asiakassivuston painoa keskitettyyn aggregointipalvelimeen, voidaan minimoida maailmanlaajuinen menetys ilman, että tarvitaan pääsyä koko tietokokonaisuuteen keskitettyyn sijaintiin. Jokainen asiakassivusto oppii paikallisesti ja jakaa mallipainon päivitykset keskuspalvelimeen, joka aggregoi panokset käyttäen ). 72 9c A pseudoalgorithm of FL is shown in Supplementary Note . In our experiments, we set the number of federated rounds at = 200, with one local training epoch per round at each client. The number of clients, , 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, , riippuu kunkin asiakkaan tietokokonaisuuden koosta 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 nk 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 Satunnaisia affine-muunnoksia, mukaan lukien pyöriminen, käännökset, leikkaaminen, skaalaaminen ja satunnaisen voimakkuuden melu ja muutokset, sovellettiin kuviin tietojen lisäämiseksi koulutuksen aikana. 73 Owing to the sensitivity of BN layers Kun käsittelimme eri asiakkaita ei-riippumattomassa ja identtisesti jakautuneessa ympäristössä, havaitsimme, että paras mallin suorituskyky tapahtui pitämällä esivalmistettua ResNet34:ää paikkatietoisena. parameters fixed during FL training (that is, using a learning rate of zero for those layers). The Deep & Cross network that combines image features with EMR features does not contain BN layers and hence was not affected by BN instability issues. 58 47 Tässä tutkimuksessa tutkittiin yksityisyyden säilyttämisen järjestelmää, joka jakaa vain osittaiset mallin päivitykset palvelimen ja asiakassivustojen välillä. Painon päivitykset sijoitettiin jokaisen iteraation aikana panoksen suuruuden mukaan, ja vain tietty prosenttiosuus suurimmista painon päivityksistä jaettiin palvelimen kanssa. Tarkemmin sanottuna painon päivitykset (tunnetaan myös gradienteina) jaettiin vain, jos niiden absoluuttinen arvo oli tietyn prosenttiyksikköarvon yläpuolella, (t) (Extended Data Fig. ), which was computed from all non-zero gradients, Δ , and could be different for each client in each FL round . Variations of this scheme could include additional clipping of large gradients or differential privacy schemes that add random noise to the gradients, or even to the raw data, before feeding into the network . k 5 Wk(t) k t 49 51 Statistical analysis Teimme Wilcoxonin allekirjoittaman testin vahvistaaksemme paikallisesti koulutettujen mallien ja FL-mallin välillä havaitun suorituskyvyn parannuksen merkityksen 24 ja 72 tunnin aikavyöhykkeillä (kuva. and Extended Data Fig. ). The null hypothesis was rejected with one-sided 1 × 10–3 molemmissa tapauksissa. 2 1 P Pearson’s correlation was used to assess the generalizability (robustness of the average AUC value to other client sites’ test data) of locally trained models in relation to respective local dataset size. Only a moderate correlation was observed ( = 0.43, = 0.035, degrees of freedom (df) = 17 for the 24-h model and = 0.62, = 0.003, df = 16 for the 72-h model). This indicates that dataset size alone is not the only factor determining a model’s robustness to unseen data. r P r P To compare ROC curves from the global FL model and local models trained at different sites (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)/ Missä is the standardized difference, is the standard deviation of the bootstrap differences and AUC1 and AUC2 are the corresponding bootstrapped AUC series. By comparing with normal distribution, we obtained the 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 values was conducted in R with the 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. ). We conducted one-way analysis of variation (ANOVA) tests to compare local and FL model scores among four ground truth categories (RA, LFO, HFO, MV). The -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 -values of five different local sites are 245.7, 253.4, 342.3, 389.8 and 634.8, while that of the FL model is 843.5. Given that larger -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 Reporting Summary Further information on research design is available in the linkitetty tähän artikkeliin. Nature Research Reporting Summary Tietojen saatavuus The dataset from the 20 institutes that participated in this study remains under their custody. These data were used for training at each of the local sites and were not shared with any of the other participating institutions or with the federated server, and they are not publicly available. Data from the independent validation sites are maintained by CAMCA, and access can be requested by contacting Q.L. Based on determination by CAMCA, a data-sharing review and amendment of IRB for research purposes can be conducted by MGB research administration and in accordance with MGB IRB and policy. Code availability All code and software used in this study are publicly available at NGC. To access, log in as a guest or create a profile then enter one of the URLs below. The trained models, data preparation guidelines, code for training, validating testing of the model, readme file, installation guideline and license files are publicly available at NVIDIA NGC : The federated learning software is available as part of the Clara Train SDK: . Alternatively, use this command to download the model “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 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 Tunnustukset MGB kiittää seuraavia yksilöitä tuesta: J. Brink, Department of Radiology, Massachusetts General Hospital, Boston, MA; N. Guo, Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital, Department of Radiology, Massachusetts General 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, this QTIM lab at the Athina Harvard A. Martinos Center for Biomical Imaging at MGH; S. Pomantzer, Department Lääketieteellinen tiedekunta, Chulalongkornin yliopisto kiittää Ratchadapisek Sompoch Endowment Fund RA (PO) (nro 001/63) keräämisestä ja hallinnoinnista COVID-19-tietoihin liittyvistä kliinisistä tiedoista ja biologisista näytteistä tutkimustyöryhmälle, lääketieteelliselle tiedekunnalle, Chulalongkornin yliopistolle. NIHR Cambridge Biomedical Research Center kiittää A. Priestia, jota tukee NIHR (Cambridge Biomedical Research Centre at Cambridge University Hospitals NHS Foundation Trust). National Taiwan University MeDA Lab ja MAHC ja Taiwan National Health Insurance Administration kiittävät MOST Joint Research Center for AI technology, All Vista Healthcare National Health Insurance Administration, Taiwan, Ministry of Science and Technology, Taiwan https://data.ucsf.edu/covid19 Tämä artikkeli on saatavilla luonnossa lisenssillä CC by 4.0 Deed (Attribution 4.0 International). Tämä artikkeli on saatavilla luonnossa lisenssillä CC by 4.0 Deed (Attribution 4.0 International).
