סופרים : איטאי דאיאן 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 ג'ון W. גרט 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 פבלו F. Damasceno Pedro Mario Cruz e Silva Pochuan Wang Sheng Xu שואיצ'י קוואנו 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 טורקיה Turkbey Benjamin Glicksberg Bernardo Bizzo Byung Seok Kim Carlos Tor-Díez Chia-Cheng Lee צ'יא-ג'ונג Hsu Chin Lin Chiu-Ling Lai Christopher P. Hess Colin Compas Deepeksha Bhatia אריק K. אורמן 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 פטריוטים פטריוטים פיטר אלנייאר Sarah Hickman Sharmila Majumdar Shelley L. McLeod שרידאן ריד Stefan Gräf סטפני הרמון Tatsuya Kodama Thanyawee Puthanakit Tony Mazzulli Vitor Lima de Lavor יוסי רקובטאי Yu Rim Lee Yuhong Wen Fiona J. Gilbert Mona G. Flores Quanzheng Li Authors: איטאי דאיאן Holger R. Roth Aoxiao זונג Ahmed Harouni Amilcare Gentili אנה ז'אבידיין אנדרו ליו Anthony Beardsworth Costa Bradford J. Wood Chien-Sung Tsai Chih-Hung Wang צ'ון נאן Hsu C. K. Lee רויאל רויאל Daguang Xu Dufan Wu אדי הונג Felipe Campos Kitamura Griffin Lacey Gustavo César de Antônio Corradi גוסטב נינו Hao-Hsin Shin Hirofumi Obinata תערוכות Hui Ren Jason C. Crane Jesse Tetreault ג'ואי גואן ג'ון W. גרט ג'ושאו ד. קג'י פארק ג'ונג גיל Keith Dreyer Krishna Juluru Kristopher Kersten Marcio Aloisio Bezerra Cavalcanti רוקנבאך Marius George Linguraru Masoom A. Haider Meena AbdelMaseeh ניקולא ריצ'ה פבלו F. Damasceno Pedro Mario Cruz e Silva Pochuan Wang Sheng Xu שואיצ'י קוואנו Sira Sriswasdi Soo Young Park Thomas M. Grist ספר Varun ארוחת בוקר חגיגית Weichung Wang נוער לא זכה Xiang Li Xihong Lin Young Joon Kwon Abood Quraini Andrew Feng Andrew N. Priest טורקיה Turkbey בנג'מין גליקסברג Bernardo Bizzo Byung Seok Kim Carlos Tor-Díez Chia-Cheng Lee צ'יא-ג'ונג Hsu סין לין צ'י-לינג Lai כריסטופר פ. האס קולין קומפס Deepeksha בטיה אריק K. אורמן אוואן לייבוביץ Hisashi Sasaki Hitoshi Mori Isaac Yang Jae Ho Sohn Krishna Nand Keshava Murthy לי-צ'ן פו Matheus Ribeiro Furtado de Mendonça Mike Fralick Min Kyu Kang Mohammad Adil Natalie Gangai פטריוטים פטריוטים פיטר אלנייאר Sarah Hickman Sharmila Majumdar Shelley L. McLeod שרידאן ריד Stefan Gräf סטפני הרמון Tatsuya Kodama Thanyawee Puthanakit Tony Mazzulli Vitor Lima de Lavor יוסי רקובטאי Yu Rim Lee יואנג ון Fiona J. Gilbert Mona G. Flores Quanzheng Li Abstract Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site’s data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare. ראשי 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 , , , , , , גיוס ניסויים קליניים נעשה במהירות וקל על ידי גופים רגולטוריים לאומיים ורוח שיתוף פעולה בינלאומית. , , דיסציפלינות האנליטיקה של הנתונים והאינטליגנציה האינטלקטואלית תמיד קידמו גישות פתוחות ושיתוף פעולה, הכוללות מושגים כגון תוכנה קוד פתוחה, מחקר חוזר, מחסני נתונים והפצה ציבורית של קבוצות נתונים אנונימיות. , . The pandemic has emphasized the need to expeditiously conduct data collaborations that empower the clinical and scientific communities when responding to rapidly evolving and widespread global challenges. Data sharing has ethical, regulatory and legal complexities that are underscored, and perhaps somewhat complicated, by the recent entrance of large technology companies into the healthcare data world , , . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 דוגמה ספציפית לסוגים אלה של שיתוף פעולה היא העבודה הקודמת שלנו על מודל תמיכה בהחלטות קליניות (CDS) SARS-COV-2 מבוסס AI. מודל CDS זה פותח ב Mass General Brigham (MGB) והוכח על ידי נתוני מערכות בריאות מרובות. , , , . CXR was selected as the imaging input because it is widely available and commonly indicated by guidelines such as those provided by ACR , the Fleischner Society , the WHO , national thoracic societies , national health ministry COVID handbooks and radiology societies across the world . 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 שימוש בשיטות כגון לימוד העברה , 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 הלמידה הפדרלית תומכת בהשקת מהירה של ניסויים מאורגן מרכזי עם מעקב משופץ של נתונים והערכה של שינויים אלגוריתמיים והשפעה גישה אחת ל-FL, הנקראת לקוח-שרת, שולחת מודל 'לא מאומץ' לשרתים אחרים ('נודים') המביאים משימות אימון חלקיות, בתורו שולחת את התוצאות בחזרה כדי להתמזג בשרת המרכזי ('פדרציה'). . 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 כבר הראה הבטחה ביישומים רפואיים לאחרונה , , , בדיקת COVID-19 , , . 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 Our objective was to develop a robust, generalizable model that could assist in triaging patients. We theorized that the CDS model can be federated successfully, given its use of data inputs that are relatively common in clinical practice and that do not rely heavily on operator-dependent assessments of patient condition (such as clinical impressions or reported symptoms). Rather, laboratory results, vital signs, an imaging study and a commonly captured demographic (that is, age), were used. We therefore retrained the CDS model with diverse data using a client-server FL approach to develop a new global FL model, which was named EXAM, using CXR and EMR features as input. By leveraging FL, the participating institutions would not have to transfer data to a central repository, but rather leverage a distributed data framework. Our hypothesis was that EXAM would perform better than local models and would generalize better across healthcare systems. Results The EXAM model architecture The EXAM model is based on the CDS model mentioned above בסך הכל, 20 תכונות (19 מ EMR ו CXR אחד) שימשו ככניסה למודל. התוצאות (כלומר, 'אמת כדור הארץ') תוויות הוענקו בהתבסס על הטיפול החמצן של המטופל לאחר 24 ו 72 שעות של תקופות מההקלטה הראשונית למחלקה חירום (ED). רשימה מפורטת של התכונות והתוצאות הנדרשות ניתן לראות בטבלה . 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). For EMR features, only the first values captured in the ED were used and data preprocessing included deidentification, missing value imputation and normalization to zero-mean and unit variance. For CXR images, only the first obtained in the ED was used. המודל משלב מידע משני תכונות EMR ו- CXR, תוך שימוש ברשת נוירלית קונפוציונלית של 34 שכבות (ResNet34) כדי לחלץ תכונות מרשת CXR ו- Deep & Cross כדי להתחבר את התכונות יחד עם תכונות EMR. תוצאת המודל היא ציון סיכון, הנקרא ציון EXAM, שהוא ערך מתמשך בטווח 0–1 עבור כל אחת מהתחזיות של 24 ו-72 שעות המתאימות לתוויות המתוארות לעיל. Methods Federating the model 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. נתונים בין האתרים לא היו מאוזנים לפני החילוץ, לאור נסיבות מחשב קליניות אמיתיות, הרמוניה מעמיקה של כניסת הנתונים לא נערכה על ידי המחברים (איור. ). 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 על פי נתוני המבחן של כל לקוח, ההשוואה בין הדגמים המומלצים באופן מקומי למודל ה- FL הביאה לשיפור משמעותי ביצועים ( « 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. ). The stability of our results was validated by repeating three runs of local and FL training on different randomized data splits. P 2a 2b 1 , ביצועים במבחן של כל לקוח על ידי תחזית של טיפול חמצן 24 שעות עבור מודלים מאומנים על נתונים מקומיים בלבד (Local) לעומת המודל הגלובלי הטוב ביותר הזמין על השרת (FL (ראה הטוב ביותר). , Generalizability (average performance on other sites’ test data, as represented by average AUC) as a function of a client’s dataset size (no. of cases). The green horizontal line denotes the generalizability performance of the best global model. The performance for 18 of 20 clients is shown, because client 12 had outcomes only for 72-h oxygen (Extended Data Fig. ) ולקוח 14 היו מקרים רק עם טיפול ב- RA, כך שהמטריקת ההערכה (AUC) לא הייתה חלה בשום מקרה ( ). Data for client 14 were also excluded from computation of average generalizability in local models. a b 1 שיטות 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 באתר הלקוח 16, עם נתונים לא מאוזנים וברוב המקרים קלים. , 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 במקרה של אתרי לקוח עם קבוצות נתונים קטנות יחסית, המודל FL הטוב ביותר עבר באופן משמעותי לא רק את המודל המקומי, אלא גם את אלה מאומנים על קבוצות נתונים גדולות יותר מחמישה אתרי לקוח באזור בוסטון של ארה"ב (איור. ). 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 Following initial training, EXAM was subsequently tested at three independent validation sites: Cooley Dickinson Hospital (CDH), Martha’s Vineyard Hospital (MVH) and Nantucket Cottage Hospital (NCH), all in Massachusetts, USA. The model was not retrained at these sites and it was used only for validation purposes. The cohort size and model inference results are summarized in Table , and the ROC curves and confusion matrices for the largest dataset (from CDH) are shown in 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 ), 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 , תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: תוצאות הבדיקה: ) and 72 h ( ). ROCs for three different cutoff values ( ) של ציון הסיכון של המבחן מוצגים. 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’ 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 הדיון 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 שיפורים במערכות מידע קליניות נדרשים כדי להקל על הכנת הנתונים, מה שמוביל לגיבוי טוב יותר של רשת של אתרים המשתתפים ב- FL. זה, בשילוב עם הנדסת היפרפרפרמטר, יכול לאפשר לאלגוריתם 'ללמוד' ביעילות רבה יותר ממספקי נתונים גדולים יותר ולהתאים פרמטרים מודל לאתר מסוים כדי להתאים אישית יותר - למשל, באמצעות התאמה מעמיקה נוספת באתר זה. מערכת שמאפשרת ייסורים מודלים קרובים בזמן אמת ועיבוד תוצאות יהיה גם מועיל ו 'סגור את המעגל' מן ההכשרה לשימוש מודל. 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. A feature that would enhance these kinds of large-scale collaboration is the ability to predict the contribution of each client site towards improving the global FL model. This will help in client site selection, and in prioritization of data acquisition and annotation efforts. The latter is especially important given the high costs and difficult logistics of these large-consortia endeavors, and it will enable these endeavors to capture diversity rather than the sheer quantity of data samples. Future approaches may incorporate automated hyperparameter searching , neural architecture search and other automated machine learning approaches to find the optimal training parameters for each client site more efficiently. 55 56 57 בעיות ידועות של נורמליזציה במלאי (BN) ב- 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. Methods אישור אתיקה כל התהליכים המודיעים למרכז הבריאות של טורונטו הועברו בהתאם לעקרונות הניסוי האנושי, כפי שנקבעו בהצהרה של הלסינקי ובתוכנית הבינלאומית על הרמוניה של מדריכים לטיפול קליני טוב, והיו מאושרים על ידי המועצות המוסדיות הרלוונטיות במקומות ההסכמה הבאים: CDH, MVH, NCH ובמקומות ההסכמה הבאים: MGB, בית החולים הכללי (MGH), בית החולים המוסדי Brigham and Women's Hospital, בית החולים Newton-Wellesley, מרכז הרפואי הציבורי סן סן סן-שורי ובית החולים New Faulkner (כל שמונה מהחולים האלה היו מכוסים על פי הסכמת המועצה עקבו אחרי MI-CLAIM guidelines for reporting of clinical AI models ) 2 Study setting The study included data from 20 institutions (Fig. ): 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 , , נתונים משלושה אתרים עצמאיים שימשו לאימות עצמאי: CDH, MVH ו- NCH, כולם במצ'וסטס, ארה"ב.שלושת בתי החולים הללו היו מאפיינים של אוכלוסיית המטופלים שונה מאשר באתרי האימון. 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. הגיל של המטופל וההפצה של תכונות EMR השתנו מאוד בין האתרים, כפי שציפינו בשל ההבדלים הדמוגרפיים בין בתי החולים המפוזרים ברחבי העולם (איור נתונים מורחבים. ). 1b 1c,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 Model input 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 There is wide variation in the clinical course of patients who present to hospital with symptoms of COVID-19, with some experiencing rapid deterioration in respiratory function requiring different interventions to prevent or mitigate hypoxemia , החלטה קריטית שנעשתה במהלך ההערכה של מטופל בנקודת הטיפול הראשונית, או ב- ED, היא האם המטופל צפוי לדרוש אמצעי נגד או התערבויות פולשניים יותר או מוגבלים במשאבים (כגון MV או נוגדנים מונוקלוניים), ולכן צריך לקבל טיפול נדיר אך יעיל, טיפול עם יחס סיכון-תועלת צמצום עקב תופעות לוואי או רמה גבוהה יותר של טיפול, כגון קבלת יחידת טיפול אינטנסיבי. . 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 was developed to help triage such patients. 62 63 64 65 יש לציין כי המודל אינו מאושר על ידי שום סוכנות רגולטורית בשלב זה, והוא צריך לשמש רק למטרות מחקר. EXAM score 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. illustrates how CORISK and the EXAM score can be used for patient triage. 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 pretrained on the CheXpert dataset) and the Deep & Cross network . 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. ). 9a 66 67 68 9b 69 70 27 Feature imputation and normalization A MissForest algorithm was used to impute EMR features, based on the local training dataset. If an EMR feature was completely missing from a client site dataset, the mean value of that feature, calculated exclusively on data from MGB client sites, was used. Then, EMR features were rescaled to zero-mean and unit variance based on statistics calculated on data from the MGB client sites. 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 details כנראה הצורה המוכרת ביותר של FL היא יישום של אלגוריתם הממוצע הפדרלי כפי שהוצע על ידי McMahan et al. ניתן לחשוב על FL כשיטה שמטרתה למזער פונקציה אובדן גלובלית על ידי הפחתת קבוצה של פונקציות אובדן מקומיות, המוערכות בכל אתר. על ידי מינימום האובדן המקומי של כל אתר הלקוח תוך סינכרון משקל האתר הלקוח המומלץ על שרת אוסף מרכזי, ניתן למזער את האובדן הגלובלי מבלי צורך לגשת לכל קבוצת הנתונים במיקום מרכזי. כל אתר הלקוח לומד באופן מקומי, ומשתף עדכוני משקל מודל עם שרת מרכזי שמאחד תרומות באמצעות פרוטוקולי הצפנה של שכבות סופיות מאובטחות ותקשורת. השרת שולח לאחר מכן קבוצה מעודכנת של משקלים לכל אתר הלקוח לאחר אוסף, וממשיך את האימון ). 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, , 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 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 . Random affine transformations, including rotation, translations, shear, scaling and random intensity noise and shifts, were applied to the images for data augmentation during training. 73 Owing to the sensitivity of BN layers when dealing with different clients in a nonindependent and identically distributed setting, we found the best model performance occurred when keeping the pretrained ResNet34 with spatial attention 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 In this study we investigated a privacy-preserving scheme that shares only partial model updates between server and client sites. The weight updates were ranked during each iteration by magnitude of contribution, and only a certain percentage of the largest weight updates was shared with the server. To be exact, weight updates (also known as gradients) were shared only if their absolute value was above a certain percentile threshold, (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 We conducted a Wilcoxon signed-rank test to confirm the significance of the observed improvement in performance between the locally trained model and the FL model for the 24- and 72-h time points (Fig. and Extended Data Fig. ). The null hypothesis was rejected with one-sided « 1 × 10–3 in both cases. 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 עבור המודל 72-h).זה מצביע על כך שהגודל של קבוצת הנתונים לבדו אינו הגורם היחיד לקבוע את עמידותו של המודל לנתונים בלתי נראים. r P r P כדי להשוות את קווי ROC מן המודל FL הגלובלי ומודלים מקומיים מאומנים באתרים שונים (איור נתונים מורחבים. ), 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 – AUC2 – AUC2 , where is the standardized difference, הוא ההפרש הסטנדרטי של ההבדלים ב-bootstrap ו-AUC1 ו-AUC2 הם סדרת ה-AUC המתאימה ב-bootstrapped. עם ההפצה הרגילה, קיבלנו את values illustrated in Supplementary Table . The results show that the null hypothesis was rejected with very low ערכים, המעידים על המשמעות הסטטיסטית של העליונות של תוצאות FL. values was conducted in R with the pROC library . 3 D s D s D P 2 P P 74 מאחר שהמודל מצפה לתוצאה דיסקרטית, ציון מתמשך מ- 0 עד 1, לא ניתן לבצע הערכה פשוטה של מיקרוסופט, כגון qqplot. ביצענו ניתוח חד-משמעי של שינויים (ANOVA) כדי להשוות את ציוני המודל המקומי וה-FL בין ארבעת קטגוריות האמת על הקרקע (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 -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 linked to this article. Nature Research Reporting Summary נתונים זמינים 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 כל הקוד והתוכנה המשמשים במחקר זה זמינים לציבור ב-NGC. כדי לגשת אליהם, להיכנס אליהם כאורח או ליצור פרופיל, הקלד אחד מהכתובות הבאות.המודלים המומלצים, הנחיות הכנת הנתונים, קוד הכשרה, בדיקות אימות של המודל, קובץ readme, הוראות התקנה וקבצי רישיון זמינים לציבור ב-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 התייחסות 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 The views expressed in this study are those of the authors and not necessarily those of the NHS, the NIHR, the Department of Health and Social Care or any of the organizations associated with the authors. MGB thank the following individuals for their support: J. Brink, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA; M. Kalra, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA; N. Neumark, 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; J. K. Cramer, Director, QTIM lab at the Athinoula A. Martinos Center for Biomedical Imaging at MGH; S. Pomerantz, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA; G. Boland, Department of Radiology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA; W. Mayo-Smith, Department of Radiology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA. UCSF thank P. B. Storey, J. Chan and J. Block for implementing the UCSF FL client infrastructure, and W. Tellis for providing the source imaging repository for this work. The UCSF EMR and clinical notes for this study were accessed via the COVID-19 Research Data Mart, . The Faculty of Medicine, Chulalongkorn University thank the Ratchadapisek Sompoch Endowment Fund RA (PO) (no. 001/63) for the collection and management of COVID‐19-related clinical data and biological specimens for the Research Task Force, Faculty of Medicine, Chulalongkorn University. NIHR Cambridge Biomedical Research Centre thank A. Priest, who is supported by the NIHR (Cambridge Biomedical Research Centre at the Cambridge University Hospitals NHS Foundation Trust). National Taiwan University MeDA Lab and the MAHC and Taiwan National Health Insurance Administration thank the MOST Joint Research Center for AI technology, the All Vista Healthcare National Health Insurance Administration, Taiwan, the Ministry of Science and Technology, Taiwan and the National Center for Theoretical Sciences Mathematics Division. National Institutes of Health (NIH) acknowledge that the NIH Medical Research Scholars Program is a public–private partnership supported jointly by the NIH and by generous contributions to the Foundation for the NIH from the Doris Duke Charitable Foundation, the American Association for Dental Research, the Colgate-Palmolive Company, Genentech, alumni of student research programs and other individual supporters via contributions to the Foundation for the NIH. https://data.ucsf.edu/covid19 This paper is under CC by 4.0 Deed (Attribution 4.0 International) license. available on nature This paper is under CC by 4.0 Deed (Attribution 4.0 International) license. זמין בטבע
