Authors: 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 Peiying Ruan Daguang Xu Dufan Wu Eddie Huang Felipe Campos Kitamura Griffin Lacey Gustavo César de Antônio Corradi Gustavo Nino Hao-Hsin Shin Hirofumi Obinata Hui Ren Jason C. Crane Jesse Tetreault Jiahui Guan John W. Garrett Joshua D. Kaggie Jung Gil Park Keith Dreyer Krishna Juluru Kristopher Kersten Marcio Aloisio Bezerra Cavalcanti Rockenbach Marius George Linguraru Masoom A. Haider Meena AbdelMaseeh Nicola Rieke Pablo F. Damasceno Pedro Mario Cruz e Silva Pochuan Wang Sheng Xu Shuichi Kawano Sira Sriswasdi Soo Young Park Thomas M. Grist Varun Buch Watsamon Jantarabenjakul Weichung Wang Won Young Tak Xiang Li Xihong Lin Young Joon Kwon Abood Quraini Andrew Feng Andrew N. Priest Baris Turkbey Benjamin Glicksberg Bernardo Bizzo Byung Seok Kim Carlos Tor-Díez Chia-Cheng Lee Chia-Jung Hsu Chin Lin Chiu-Ling Lai Christopher P. Hess Colin Compas Deepeksha Bhatia Eric K. Oermann Evan Leibovitz Hisashi Sasaki Hitoshi Mori Isaac Yang Jae Ho Sohn Krishna Nand Keshava Murthy Li-Chen Fu Matheus Ribeiro Furtado de Mendonça Mike Fralick Min Kyu Kang Mohammad Adil Natalie Gangai Peerapon Vateekul Pierre Elnajjar Sarah Hickman Sharmila Majumdar Shelley L. McLeod Sheridan Reed Stefan Gräf Stephanie Harmon Tatsuya Kodama Thanyawee Puthanakit Tony Mazzulli Vitor Lima de Lavor Yothin Rakvongthai Yu Rim Lee Yuhong Wen Fiona J. Gilbert Mona G. Flores Quanzheng Li 저자 : 이타이 다이안 Holger R. Roth Aoxiao Zhong에 관해 Ahmed Harouni 애니메이션 친절 Anas Z. Abidin Andrew Liu Anthony Beardsworth 코스타 브래드포드 J. 우드 Chien-Sung Tsai Chih-Hung Wang에 관하여 C. K. Lee Peiying Ruan Daguang Xu Dufan Wu Eddie Huang 펠리페 캠포스 키타무라 그리핀 라세이 Gustavo César de Antônio Corradi 부근의 호텔 Gustavo Nino Hirofumi Obinata Hui Ren 제이슨 C. 크레인 Jesse Tetreault Jiahui Guan John W. Garrett 조슈아 D. 카기 Jung Gil Park Keith Dreyer 크리슈나 유루루 Kristopher Kersten Marcio Aloisio Bezerra Cavalcanti Rockenbach Marius George Linguraru Masoom A. Haider 미나 아브델마세 Nicola Rieke Pablo F. Damasceno Pedro Mario Cruz e Silva Pochuan Wang Sheng Xu Shuichi Kawano 시라 스리사우디 Soo Young Park Thomas M. Grist Varun Buch 저녁 식사 웨이브 Weichung Wang Won Young Tak Xiang Li Xihong Lin 청년 Joon Kwon Abood Quraini Andrew Feng Andrew N. Priest Baris Turkbey Benjamin Glicksberg Bernardo Bizzo 김정은 김정은 Carlos Tor-Díez Chia-Cheng Lee Chin Lin 링 라이 Christopher P. Hess Colin Compas Deepeksha Bhatia Eric K. Oermann Evan Leibovitz Hisashi Sasaki Hitoshi Mori 이사크 얀 제이 호 아들 Krishna Nand Keshava Murthy Li-Chen Fu Matheus Ribeiro Furtado de Mendonça Mike Fralick 미니 키우 칸 Mohammad Adil Natalie Gangai Peerapon Vateekul 피에르 Elnajjar 사라 히크먼 Sharmila Majumdar Shelley L. McLeod 샤리단 리드 Stefan Gräf Stephanie Harmon 타츠야 코다마 Thanyawee Puthanakit Tony Mazzulli Vitor Lima de Lavor Yothin Rakvongthai Yu Rim 리 Yuhong Wen 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. 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 , , , , , , 임상시험 모집은 국가 규제 기관과 국제 협력 정신에 의해 가속화되고 촉진되었습니다. , , . The data analytics and AI disciplines have always fostered open and collaborative approaches, embracing concepts such as open-source software, reproducible research, data repositories and making available anonymized datasets publicly , 전염병은 급속하게 진화하고 광범위한 글로벌 도전에 대응할 때 임상 및 과학 커뮤니티를 권한을 부여하는 데이터 공동 작업을 진행할 필요성을 강조했습니다.데이터 공유에는 윤리적, 규제적, 법적 복잡성이 있으며, 최근 대형 기술 회사들이 의료 데이터 세계에 들어가면서 조금은 복잡해졌습니다. , , . 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 A concrete example of these types of collaboration is our previous work on an AI-based SARS-COV-2 clinical decision support (CDS) model. This CDS model was developed at Mass General Brigham (MGB) and was validated across multiple health systems’ data. The inputs to the CDS model were chest X-ray (CXR) images, vital signs, demographic data and laboratory values that were shown in previous publications to be predictive of outcomes of patients with COVID-19 , , , . 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 · 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 , , 잠재적으로 과잉 조립 및 낮은 일반화 가능성으로 이어질 수 있습니다.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 연방 학습은 데이터의 추적 가능성을 향상시키고 알고리즘 변화와 영향을 평가하는 중앙 조정 실험의 빠른 시작을 지원합니다. . 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 , FL는 최근의 의료 이미지 응용 프로그램에서 이미 약속을 보였습니다. , , , , including in COVID-19 analysis , , 주목할만한 예는 SARS-COV-2로 감염된 환자의 사망률 예측 모델이지만, 모드 및 규모의 수와 관련하여 제한된 임상 특징을 사용한다. . 38 39 40 41 42 43 8 44 45 46 CDS 모델은 임상 실습에서 비교적 일반적인 데이터 입력의 사용을 고려하여 성공적으로 연합될 수 있다는 이론을 제시했으며, 이는 환자의 상태 (예: 임상 인상 또는 보고 된 증상)에 대한 운영자에 의존하는 평가에 크게 의존하지 않는 것이었다. 대신 실험실 결과, 생명 징후, 이미지 연구 및 일반적으로 캡처 된 인구 통계 (즉, 연령)가 사용되었습니다. 따라서 우리는 클라이언트-서버 FL 접근법을 사용하여 다양한 데이터로 CDS 모델을 재훈련하여 새로운 글로벌 FL 모델을 개발하여 CXR 및 EMR 기능을 입력으로 사용했습니다. FL를 활용함으로써 참가 기관은 중앙 저장소로 데이터를 전송하지 않아야하지만 분산된 데이터 프레임워크를 활용해야합니다. Our hypothesis was that EXAM would perform better than local models and would generalize better across healthcare systems. Results EXAM 모델 아키텍처 The EXAM model is based on the CDS model mentioned above . 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). 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. The model therefore fuses information from both EMR and CXR features, using a 34-layer convolutional neural network (ResNet34) to extract features from a CXR and a Deep & Cross network to concatenate the features together with the EMR features (for more expanded details, see ). The model output is a risk score, termed the EXAM score, which is a continuous value in the range 0–1 for each of the 24- and 72-hour predictions corresponding to the labels described above. 방법 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. ). 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. , 각 기관 또는 사이트가 기여한 사례 수 (클라이언트 1은 가장 많은 사례를 기여하는 사이트를 나타냅니다). , 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 We compared locally trained models with the global FL model on each client’s test data. Training the model through FL resulted in a significant performance improvement ( « 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 , 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. , 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. ) and client 14 had cases only with RA treatment, such that the evaluation metric (av. AUC) was not applicable in either of these cases ( 클라이언트 14의 데이터는 로컬 모델에서 평균 일반화 가능성의 계산에서 제외되었다. 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 ( ) 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 비교적 작은 데이터 세트를 가진 클라이언트 사이트의 경우, 가장 좋은 FL 모델은 로컬 모델뿐만 아니라 미국 보스턴 지역의 5 개 클라이언트 사이트에서 더 큰 데이터 세트를 훈련받은 모델을 상당히 뛰어넘었습니다. ). 3b 글로벌 모델은 COVID 긍정적이고 부정적 인 환자에서 24/72 시간에 산소 요구를 예측하는 데 잘 작동했습니다. ). 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 , , Performance (ROC) (top) and confusion matrices (bottom) of the EXAM FL model on the CDH dataset for prediction of oxygen requirement at 24 h ( ) and 72 h ( ). ROCs for three different cutoff values ( ) of the EXAM risk score are shown. a b a b t For MV at CDH at 72 h, EXAM had a low false-negative rate of 7.1%. Representative failure cases are presented in Extended Data Fig. , showing two false-negative cases from CDH where one case had many missing EMR data features and the other had a CXR with a motion artifact and some missing EMR features. 4 Use of differential privacy A primary motivation for healthcare institutes to use FL is to preserve the security and privacy of their data, as well as adherence to data compliance measures. For FL, there remains the potential risk of model ‘inversion’ 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 COVID-19 환자들의 의사 결정을 지원하는 200가지 이상의 예측 모델이 발표되었습니다. . 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. 미래에는 질병 진화의 다른 단계로 인한 ‘인구 흐름’의 가능성을 조사할 계획입니다.우리는 20개 지역의 다양성으로 인해 이러한 위험이 완화되었을 수 있다고 생각합니다. 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 각 클라이언트 사이트에 대한 최적의 훈련 매개 변수를 더 효율적으로 찾기 위한 접근법입니다. 55 56 57 Known issues of batch normalization (BN) in FL motivated us to fix our base model for image feature extraction to reduce the divergence between unbalanced client sites. Future work might explore different types of normalization techniques to allow the training of AI models in FL more effectively when client data are nonindependent and identically distributed. 58 49 Recent works on privacy attacks within the FL setting have raised concerns on data leakage during model training . Meanwhile, protection algorithms remain underexplored and constrained by multiple factors. While differential privacy algorithms , , show good protection, they may weaken the model’s performance. Encryption algorithms, such as homomorphic encryption , maintain performance but may substantially increase message size and training time. A quantifiable way to measure privacy would allow better choices for deciding the minimal privacy parameters necessary while maintaining clinically acceptable performance , , . 59 36 48 49 60 36 48 49 Following further validation, we envision deployment of the EXAM model in the ED setting as a way to evaluate risk at both the per-patient and population level, and to provide clinicians with an additional reference point when making the frequently difficult task of triaging patients. We also envision using the model as a more sensitive population-level metric to help balance resources between regions, hospitals and departments. Our hope is that similar FL efforts can break the data silos and allow for faster development of much-needed AI models in the near future. 방법 Ethics approval 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 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 , , . Data from three independent sites were used for independent validation: CDH, MVH and NCH, all in Massachusetts, USA. These three hospitals had patient population characteristics different from the training sites. The data used for the algorithm validation consisted of patients admitted to the ED at these sites between March 2020 and February 2021, and that satisfied the same inclusion criteria of the data used to train the FL model. 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. ). 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 input 모델 결과(즉, 지상 진실) 라벨은 ED에 초기 입학 후 24시간 및 72시간 후 환자의 요구 사항에 따라 할당되었습니다. . 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 모델 개발 시험 COVID-19의 증상으로 병원에 입원하는 환자의 임상 과정에는 광범위한 변화가 있으며, 일부는 호흡 기능의 급속한 악화를 경험하여 hypoxemia를 예방하거나 완화시키기 위해 다른 개입이 필요합니다. , . 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 was developed to help triage such patients. 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. 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 Arguably the most established form of FL is implemention of the federated averaging algorithm as proposed by McMahan et al. 이 알고리즘은 각 참가 사이트가 클라이언트로 작용하는 클라이언트-서버 설정을 사용하여 실현할 수 있습니다. FL는 각 사이트에서 추정되는 로컬 손실 기능의 집합을 줄여서 글로벌 손실 기능을 최소화하는 방법으로 생각할 수 있습니다. 각 클라이언트 사이트의 로컬 손실을 최소화함으로써 중앙 집합 서버에 배운 클라이언트 사이트 무게를 동기화함으로써, 중앙 집합 사이트에서 전체 데이터 세트에 액세스 할 필요없이 글로벌 손실을 최소화 할 수 있습니다. 각 클라이언트 사이트는 로컬로 학습하고 보안 소켓 레이어 암호화 및 통신 프로토콜을 사용하여 기여를 집합하는 중앙 서버와 모델 무게 업데이트를 공유합니다. 서버는 집합 ). 72 9c A pseudoalgorithm of FL is shown in 추가 노트 . 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 이 계획의 변형에는 대규모 gradients 또는 차별적 개인 정보 보호 계획의 추가 클리핑이 포함될 수 있습니다. 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 피어슨의 상관 관계는 지역 데이터 세트 크기에 대한 지역적으로 훈련 된 모델의 일반화 가능성 (다른 클라이언트 사이트의 테스트 데이터에 대한 평균 AUC 값의 안정성)을 평가하는 데 사용되었습니다. = 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)/ , where is the standardized difference, is the standard deviation of the bootstrap differences and AUC1 and AUC2 are the corresponding bootstrapped AUC series. By comparing 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 값은 pROC 라이브러리와 함께 R에서 수행되었습니다. . 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 샘플 사이의 변동으로 계산된 통계는 샘플 내의 변동으로 나누어 샘플 사이의 분산 정도를 나타내는 샘플 사이의 변동으로 계산되었으며, 모델을 정량화하는 데 사용되었습니다. -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 Data availability 이 연구에 참여한 20개 연구소의 데이터 세트는 그들의 관리하에 남아있다.이 데이터는 각 지역 사이트에서 교육을 위해 사용되었으며 다른 참가 기관이나 연방 서버와 공유되지 않았으며 공개적으로 사용할 수 없습니다.독립적 인 검증 사이트의 데이터는 CAMCA에 의해 유지되며 Q.L.에 연락하여 액세스를 요청할 수 있습니다.CAMCA의 결정에 따라 MGB 연구 관리 및 MGB IRB 및 정책에 따라 연구 목적을위한 IRB의 데이터 공유 검토 및 수정이 수행 될 수 있습니다. Code availability 이 연구에서 사용되는 모든 코드와 소프트웨어는 NGC에서 공개적으로 사용할 수 있습니다. 액세스하고, 손님으로 로그인하거나 프로필을 만들려면 아래의 URL 중 하나를 입력하십시오.교육된 모델, 데이터 준비 지침, 교육 코드, 모델 검증, readme 파일, 설치 지침 및 라이선스 파일은 NVIDIA NGC에서 공개적으로 사용할 수 있습니다. : : 연방 학습 소프트웨어는 Clara Train SDK의 일부로 사용할 수 있습니다: . Alternatively, use this command to download the model “wget --content-disposition 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 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, 의학부를 통해 Chulalongkorn University는 Ratchadapisek Sompoch Endowment Fund RA (PO) (No. 001/63)에 COVID-19 관련 임상 데이터 및 생물학 표본의 수집 및 관리에 대해 의학부, Chulalongkorn University에 대한 감사를 전합니다. NIHR Cambridge Biomedical Research Center는 NIHR (Cambridge University Hospitals NHS Foundation Trust의 Cambridge Biomedical Research Center)에 의해 지원되는 A. Priest에게 감사를 전합니다. National Taiwan University MeDA Lab 및 MAHC 및 Taiwan National Health Insurance Administration는 AI 기술에 대한 MOST Joint Research Center, All Vista Healthcare National Health Insurance Administration, Taiwan, Ministry of Science and Technology, Taiwan National Center for Theoretical Sciences Mathematics Division에 대한 감사를 전합니다. National Institutes of Health ( https://data.ucsf.edu/covid19 이 논문은 CC by 4.0 Deed (Attribution 4.0 International) 라이선스 아래 자연에서 사용할 수 있습니다. 이 논문은 CC by 4.0 Deed (Attribution 4.0 International) 라이선스 아래 자연에서 사용할 수 있습니다.
