How do we operate a large monthly red blood cell exchange program.

BACKGROUND: Red blood cell (RBC) exchange for sickle cell disease presents unique difficulties due to RBC phenotyping, complex antibody work-ups, large number of RBC units required, and vascular access considerations, any of which can delay the procedure. Multidisciplinary coordination and systemic processes ensure that monthly appointments remain on schedule. STUDY DESIGN AND METHODS: A high-volume chronic RBC exchange program is described, highlighting the importance of multidisciplinary coordination and process improvement strategies involving initial referral, vascular access, order sets, and allocation of antigen-negative or phenotypically matched RBCs. RESULTS: Approximately 50 outpatient RBC exchanges are performed each month with an 82% kept-appointment rate. Specific factors for program success include open communication across services and improvements to referrals and standardized order sets. CONCLUSION: A combination of multidisciplinary coordination and process improvement can ensure the success of a high volume RBC exchange program. Frequent communication of upcoming appointments between the referring hematologists, the hemapheresis clinic, transfusion service, and interventional radiology is critical. Advance notice to the immunohematology reference lab of upcoming appointments is needed to allow enough time for allocating antigen-negative RBCs. Order sets can be leveraged to standardize and streamline RBC exchanges. Lastly, numerous mechanisms help patients compensate for the cognitive sequelae of stroke.

Improving the Automatic Classification of Brain MRI Acquisition Contrast with Machine Learning.

Automated quantification of data acquired as part of an MRI exam requires identification of the specific acquisition of relevance to a particular analysis. This motivates the development of methods capable of reliably classifying MRI acquisitions according to their nominal contrast type, e.g., T1 weighted, T1 post-contrast, T2 weighted, T2-weighted FLAIR, proton-density weighted. Prior studies have investigated using imaging-based methods and DICOM metadata-based methods with success on cohorts of patients acquired as part of a clinical trial. This study compares the performance of these methods on heterogeneous clinical datasets acquired with many different scanners from many institutions. RF and CNN models were trained on metadata and pixel data, respectively. A combined RF model incorporated CNN logits from the pixel-based model together with metadata. Four cohorts were used for model development and evaluation: MS research (n = 11,106 series), MS clinical (n = 3244 series), glioma research (n = 612 series, test/validation only), and ADNI PTSD (n = 477 series, training only). Together, these cohorts represent a broad range of acquisition contexts (scanners, sequences, institutions) and subject pathologies. Pixel-based CNN and combined models achieved accuracies between 97 and 98% on the clinical MS cohort. Validation/test accuracies with the glioma cohort were 99.7% (metadata only) and 98.4 (CNN). Accurate and generalizable classification of MRI acquisition contrast types was demonstrated. Such methods are important for enabling automated data selection in high-throughput and big-data image analysis applications.

Utilizing a Digital Swarm Intelligence Platform to Improve Consensus Among Radiologists and Exploring Its Applications.

Radiologists today play a central role in making diagnostic decisions and labeling images for training and benchmarking artificial intelligence (AI) algorithms. A key concern is low inter-reader reliability (IRR) seen between experts when interpreting challenging cases. While team-based decisions are known to outperform individual decisions, inter-personal biases often creep up in group interactions which limit nondominant participants from expressing true opinions. To overcome the dual problems of low consensus and interpersonal bias, we explored a solution modeled on bee swarms. Two separate cohorts, three board-certified radiologists, (cohort 1), and five radiology residents (cohort 2) collaborated on a digital swarm platform in real time and in a blinded fashion, grading meniscal lesions on knee MR exams. These consensus votes were benchmarked against clinical (arthroscopy) and radiological (senior-most radiologist) standards of reference using Cohen's kappa. The IRR of the consensus votes was then compared to the IRR of the majority and most confident votes of the two cohorts. IRR was also calculated for predictions from a meniscal lesion detecting AI algorithm. The attending cohort saw an improvement of 23% in IRR of swarm votes (k = 0.34) over majority vote (k = 0.11). Similar improvement of 23% in IRR (k = 0.25) in 3-resident swarm votes over majority vote (k = 0.02) was observed. The 5-resident swarm had an even higher improvement of 30% in IRR (k = 0.37) over majority vote (k = 0.07). The swarm consensus votes outperformed individual and majority vote decision in both the radiologists and resident cohorts. The attending and resident swarms also outperformed predictions from a state-of-the-art AI algorithm.

Hyperpolarized 13 C MRI data acquisition and analysis in prostate and brain at University of California, San Francisco.

Based on the expanding set of applications for hyperpolarized carbon-13 (HP-13 C) MRI, this work aims to communicate standardized methodology implemented at the University of California, San Francisco, as a primer for conducting reproducible metabolic imaging studies of the prostate and brain. Current state-of-the-art HP-13 C acquisition, data processing/reconstruction and kinetic modeling approaches utilized in patient studies are presented together with the rationale underpinning their usage. Organized around spectroscopic and imaging-based methods, this guide provides an extensible framework for handling a variety of HP-13 C applications, which derives from two examples with dynamic acquisitions: 3D echo-planar spectroscopic imaging of the human prostate and frequency-specific 2D multislice echo-planar imaging of the human brain. Details of sequence-specific parameters and processing techniques contained in these examples should enable investigators to effectively tailor studies around individual-use cases. Given the importance of clinical integration in improving the utility of HP exams, practical aspects of standardizing data formats for reconstruction, analysis and visualization are also addressed alongside open-source software packages that enhance institutional interoperability and validation of methodology. To facilitate the adoption and further development of this methodology, example datasets and analysis pipelines have been made available in the supporting information.

Rhythmic Interventions, Musical Returns: The Health Benefits of Music as an Embodied, Everyday Practice.

In this essay, I explore the intermodal nature of well-being by considering the health benefits of music as an embodied, everyday practice in the unrelenting tempo of contemporary society. I draw from Henri Lefebvre's concept of rhythmanalysis to perform an embodied appraisal of neoliberal ideology as I experience it within the everyday spaces of higher education, and to promote an appreciation for, and utilization of musical and rhythmic interventions that restore balance and well-being amid the compressing agenda of the doctoral-degree chase. I weave together autoethnographic accounts of my own experience as I alternate between the challenges of graduate school and my involvement in various musical projects. My oscillations between these differently rhythmed worlds prompts the imagining of new possibilities for wellness and social relations through their coalescence.

Red blood cell exchange to avoid intubating a COVID-19 positive patient with sickle cell disease?

As the COVID-19 pandemic continues to claim lives across the globe, insufficient data exists regarding the optimal treatment. It is well known that patients 55 years of age or older and patients with certain chronic diseases are at higher risk of severe illness, including acute respiratory distress syndrome and death. A potentially fatal pulmonary complication of sickle cell disease, acute chest syndrome, can be precipitated by acute infections, including respiratory viruses. We report the case of a patient with sickle cell disease (HbSC) who developed COVID-19 pneumonia and acute chest syndrome who was treated with emergent red blood cell exchange in order to avoid endotracheal intubation.

Hemostasis testing and therapeutic plasma exchange: Results of a practice survey.

INTRODUCTION: Performing therapeutic plasma exchange (TPE) with albumin replacement decreases coagulation factor and platelet levels. No defined guidelines exist regarding laboratory testing to assess hemostasis in patients undergoing TPE. MATERIALS AND METHODS: A survey to evaluate hemostasis testing with TPE was distributed using online survey software. One response per institution was analyzed based on a hierarchical algorithm, excluding membrane filtration users, resulting in a maximum of 120 respondents per question. Descriptive analysis was performed with results reported as the number and/or frequency (%) of respondents to each question. RESULTS: The practices represented vary by institution type, number of apheresis procedures per year, and performance of TPE on children. Prior to TPE planned with albumin replacement, many respondents obtain laboratory studies for almost all patients (54.9% outpatients and 68.7% inpatients); however, some do not routinely obtain laboratory studies (9.7% outpatients and 4.4% inpatients). Hemoglobin/hematocrit, platelet count, fibrinogen, partial thromboplastin time (aPTT), and international normalized ratio (INR) are obtained prior to all TPE by 62.5%, 53.4%, 31.0%, 18.1%, and 17.7% of respondents, respectively; however, 1.0%, 8.7%, 29.0%, 38.3%, and 35.4%, respectively, do not routinely obtain these studies. Variation was observed in laboratory threshold values for action; the most common reported were hemoglobin/hematocrit <7 g/dL or 21% (31.0%), platelet count <50 × 109 /L (24.1%), fibrinogen <100 mg/dL (65.3%), aPTT >reference range and >1.5 times reference range (tied, 28.1%), and INR >1.5 (20.7%). CONCLUSIONS: Practice variation exists in hemostasis laboratory testing and threshold values for action with TPE. Further studies are needed to determine optimal hemostasis testing strategies with TPE.

Fully automated atlas-based method for prescribing 3D PRESS MR spectroscopic imaging: Toward robust and reproducible metabolite measurements in human brain.

PURPOSE: To implement a fully automated atlas-based method for prescribing 3D PRESS MR spectroscopic imaging (MRSI). METHODS: The PRESS selected volume and outer-volume suppression bands were predefined on the MNI152 standard template image. The template image was aligned to the subject T1 -weighted image during a scan, and the resulting transformation was then applied to the predefined prescription. To evaluate the method, H-1 MRSI data were obtained in repeat scan sessions from 20 healthy volunteers. In each session, datasets were acquired twice without repositioning. The overlap ratio of the prescribed volume in the two sessions was calculated and the reproducibility of inter- and intrasession metabolite peak height and area ratios was measured by the coefficient of variation (CoV). The CoVs from intra- and intersession were compared by a paired t-test. RESULTS: The average overlap ratio of the automatically prescribed selection volumes between two sessions was 97.8%. The average voxel-based intersession CoVs were less than 0.124 and 0.163 for peak height and area ratios, respectively. Paired t-test showed no significant difference between the intra- and intersession CoVs. CONCLUSION: The proposed method provides a time efficient method to prescribe 3D PRESS MRSI with reproducible imaging positioning and metabolite measurements. Magn Reson Med 79:636-642, 2018. © 2017 International Society for Magnetic Resonance in Medicine.

Hemostasis management and therapeutic plasma exchange: Results of a practice survey.

BACKGROUND: Patients undergoing therapeutic plasma exchange (TPE) may present with risks for hemorrhage or thrombosis. Use of replacement fluids devoid of coagulation factors will decrease factor levels and platelet levels. There are no established guidelines for hemostasis management in these situations. MATERIALS AND METHODS: A survey to evaluate current hemostasis management practice during TPE was conducted using online survey software. One response per institution was analyzed based on a hierarchical algorithm, excluding membrane filtration users, resulting in a maximum of 107 respondents. Descriptive analysis was performed with results reported as the number and frequency (%) of respondents to each question. RESULTS: Apheresis Medicine physicians, alone (59.4%) or jointly with the requesting provider (29.2%), choose the replacement fluid. Based on a theoretical patient case receiving five TPEs approximately every other day, the percent of respondents who would use albumin with or without normal saline was 94.7% with no history of a bleeding or clotting disorder, 1.1% with active bleeding, and 8.8% with hypofibrinogenemia (<100 mg/dL) due to recent TPE. More respondents would use albumin with or without normal saline for replacement fluid when a minor invasive procedure (49.5%) vs a major surgery (8.9%) was performed 1 day before TPE. Replacement fluid selection varied among respondents for several other clinical conditions. The most frequent use for cryoprecipitate by respondents (14.3%) was hypofibrinogenemia. CONCLUSIONS: These survey results demonstrate wide interinstitutional variation in replacement fluid selection to manage hemostasis in patients undergoing TPE. Further studies are needed to guide optimal hemostasis management with TPE.

Detection of localized changes in the metabolism of hyperpolarized gluconeogenic precursors 13 C-lactate and 13 C-pyruvate in kidney and liver.

PURPOSE: The purpose of this study was to characterize tissue-specific alterations in metabolism of hyperpolarized (HP) gluconeogenic precursors 13 C-lactate and 13 C-pyruvate by rat liver and kidneys under conditions of fasting or insulin-deprived diabetes. METHODS: Seven normal rats were studied by MR spectroscopic imaging of both HP 13 C-lactate and 13 C-pyruvate in both normal fed and 24 h fasting states, and seven additional rats were scanned after induction of diabetes by streptozotocin (STZ) with insulin withdrawal. Phosphoenolpyruvate carboxykinase (PEPCK) expression levels were also measured in liver and kidney tissues of the STZ-treated rats. RESULTS: Multiple sets of significant signal modulations were detected, with graded intensity in general between fasting and diabetic states. An approximate two-fold reduction in the ratio of 13 C-bicarbonate to total 13 C signal was observed in both organs in fasting. The ratio of HP lactate-to-alanine was markedly altered, ranging from a liver-specific 54% increase in fasting, to increases of 69% and 92% in liver and kidney, respectively, in diabetes. Diabetes resulted in a 40% increase in renal lactate signal. STZ resulted in 5.86-fold and 2.73-fold increases in PEPCK expression in liver and kidney, respectively. CONCLUSION: MRI of HP 13 C gluconeogenic precursors may advance diabetes research by clarifying organ-specific roles in abnormal diabetic metabolism. Magn Reson Med 77:1429-1437, 2017. © 2016 International Society for Magnetic Resonance in Medicine.

Power estimation for non-standardized multisite studies.

A concern for researchers planning multisite studies is that scanner and T1-weighted sequence-related biases on regional volumes could overshadow true effects, especially for studies with a heterogeneous set of scanners and sequences. Current approaches attempt to harmonize data by standardizing hardware, pulse sequences, and protocols, or by calibrating across sites using phantom-based corrections to ensure the same raw image intensities. We propose to avoid harmonization and phantom-based correction entirely. We hypothesized that the bias of estimated regional volumes is scaled between sites due to the contrast and gradient distortion differences between scanners and sequences. Given this assumption, we provide a new statistical framework and derive a power equation to define inclusion criteria for a set of sites based on the variability of their scaling factors. We estimated the scaling factors of 20 scanners with heterogeneous hardware and sequence parameters by scanning a single set of 12 subjects at sites across the United States and Europe. Regional volumes and their scaling factors were estimated for each site using Freesurfer's segmentation algorithm and ordinary least squares, respectively. The scaling factors were validated by comparing the theoretical and simulated power curves, performing a leave-one-out calibration of regional volumes, and evaluating the absolute agreement of all regional volumes between sites before and after calibration. Using our derived power equation, we were able to define the conditions under which harmonization is not necessary to achieve 80% power. This approach can inform choice of processing pipelines and outcome metrics for multisite studies based on scaling factor variability across sites, enabling collaboration between clinical and research institutions.

Serial analysis of 3D H-1 MRSI for patients with newly diagnosed GBM treated with combination therapy that includes bevacizumab.

Interpretation of changes in the T1- and T2-weighted MR images from patients with newly diagnosed glioblastoma (GBM) treated with standard of care in conjunction with anti-angiogenic agents is complicated by pseudoprogression and pseudoresponse. The hypothesis being tested in this study was that 3D H-1 magnetic resonance spectroscopic imaging (MRSI) provides estimates of levels of choline, creatine, N-acetylaspartate (NAA), lactate and lipid that change in response to treatment and that metrics describing these characteristics are associated with survival. Thirty-one patients with newly diagnosed GBM and being treated with radiation therapy (RT), temozolomide, erlotinib and bevacizumab were recruited to receive serial MR scans that included 3-D lactate edited MRSI at baseline, mid-RT, post-RT and at specific follow-up time points. The data were processed to provide estimates of metrics representing changes in metabolite levels relative to normal appearing brain. Cox proportional hazards analysis was applied to examine the relationship of these parameters with progression free survival (PFS) and overall survival (OS). There were significant reductions in parameters that describe relative levels of choline to NAA and creatine, indicating that the treatment caused a decrease in tumor cellularity. Changes in the levels of lactate and lipid relative to the NAA from contralateral brain were consistent with vascular normalization. Metabolic parameters from the first serial follow-up scan were associated with PFS and OS, when accounting for age and extent of resection. Integrating metabolic parameters into the assessment of patients with newly diagnosed GBM receiving therapies that include anti-angiogenic agents may be helpful for tracking changes in tumor burden, resolving ambiguities in anatomic images caused by non-specific treatment effects and for predicting outcome.

Precision medicine in chronic disease management: The multiple sclerosis BioScreen.

We present a precision medicine application developed for multiple sclerosis (MS): the MS BioScreen. This new tool addresses the challenges of dynamic management of a complex chronic disease; the interaction of clinicians and patients with such a tool illustrates the extent to which translational digital medicine-that is, the application of information technology to medicine-has the potential to radically transform medical practice. We introduce 3 key evolutionary phases in displaying data to health care providers, patients, and researchers: visualization (accessing data), contextualization (understanding the data), and actionable interpretation (real-time use of the data to assist decision making). Together, these form the stepping stones that are expected to accelerate standardization of data across platforms, promote evidence-based medicine, support shared decision making, and ultimately lead to improved outcomes.

Data Format Standardization and DICOM Integration for Hyperpolarized 13C MRI.

Hyperpolarized (HP) 13C MRI has shown promise as a valuable modality for in vivo measurements of metabolism and is currently in human trials at 15 research sites worldwide. With this growth it is important to adopt standardized data storage practices as it will allow sites to meaningfully compare data. In this paper we (1) describe data that we believe should be stored and (2) demonstrate pipelines and methods that utilize the Digital Imaging and Communications in Medicine (DICOM) standard. This includes proposing a set of minimum set of information that is specific to HP 13C MRI studies. We then show where the majority of these can be fit into existing DICOM Attributes, primarily via the "Contrast/Bolus" module. We also demonstrate pipelines for utilizing DICOM for HP 13C MRI. DICOM is the most common standard for clinical medical image storage and provides the flexibility to accommodate the unique aspects of HP 13C MRI, including the HP agent information but also spectroscopic and metabolite dimensions. The pipelines shown include creating DICOM objects for studies on human and animal imaging systems with various pulse sequences. We also show a python-based method to efficiently modify DICOM objects to incorporate the unique HP 13C MRI information that is not captured by existing pipelines. Moreover, we propose best practices for HP 13C MRI data storage that will support future multi-site trials, research studies and technical developments of this imaging technique.

Data Format Standardization and DICOM Integration for Hyperpolarized 13C MRI.

Hyperpolarized (HP) 13C MRI has shown promise as a valuable modality for in vivo measurements of metabolism and is currently in human trials at 15 research sites worldwide. With this growth, it is important to adopt standardized data storage practices as it will allow sites to meaningfully compare data. In this paper, we (1) describe data that we believe should be stored and (2) demonstrate pipelines and methods that utilize the Digital Imaging and Communications in Medicine (DICOM) standard. This includes proposing a set of minimum set of information that is specific to HP 13C MRI studies. We then show where the majority of these can be fit into existing DICOM attributes, primarily via the "Contrast/Bolus" module. We also demonstrate pipelines for utilizing DICOM for HP 13C MRI. DICOM is the most common standard for clinical medical image storage and provides the flexibility to accommodate the unique aspects of HP 13C MRI, including the HP agent information but also spectroscopic and metabolite dimensions. The pipelines shown include creating DICOM objects for studies on human and animal imaging systems with various pulse sequences. We also show a python-based method to efficiently modify DICOM objects to incorporate the unique HP 13C MRI information that is not captured by existing pipelines. Moreover, we propose best practices for HP 13C MRI data storage that will support future multi-site trials, research studies, and technical developments of this imaging technique.

Implementation and prospective real-time evaluation of a generalized system for in-clinic deployment and validation of machine learning models in radiology.

The medical imaging community has embraced Machine Learning (ML) as evidenced by the rapid increase in the number of ML models being developed, but validating and deploying these models in the clinic remains a challenge. The engineering involved in integrating and assessing the efficacy of ML models within the clinical workflow is complex. This paper presents a general-purpose, end-to-end, clinically integrated ML model deployment and validation system implemented at UCSF. Engineering and usability challenges and results from 3 use cases are presented. A generalized validation system based on free, open-source software (OSS) was implemented, connecting clinical imaging modalities, the Picture Archiving and Communication System (PACS), and an ML inference server. ML pipelines were implemented in NVIDIA's Clara Deploy framework with results and clinician feedback stored in a customized XNAT instance, separate from the clinical record but linked from within PACS. Prospective clinical validation studies of 3 ML models were conducted, with data routed from multiple clinical imaging modalities and PACS. Completed validation studies provided expert clinical feedback on model performance and usability, plus system reliability and performance metrics. Clinical validation of ML models entails assessing model performance, impact on clinical infrastructure, robustness, and usability. Study results must be easily accessible to participating clinicians but remain outside the clinical record. Building a system that generalizes and scales across multiple ML models takes the concerted effort of software engineers, clinicians, data scientists, and system administrators, and benefits from the use of modular OSS. The present work provides a template for institutions looking to translate and clinically validate ML models in the clinic, together with required resources and expected challenges.

Interactive Effects of Molecular, Therapeutic, and Patient Factors on Outcome of Diffuse Low-Grade Glioma.

PURPOSE: In patients with diffuse low-grade glioma (LGG), the extent of surgical tumor resection (EOR) has a controversial role, in part because a randomized clinical trial with different levels of EOR is not feasible. METHODS: In a 20-year retrospective cohort of 392 patients with IDH-mutant grade 2 glioma, we analyzed the combined effects of volumetric EOR and molecular and clinical factors on overall survival (OS) and progression-free survival by recursive partitioning analysis. The OS results were validated in two external cohorts (n = 365). Propensity score analysis of the combined cohorts (n = 757) was used to mimic a randomized clinical trial with varying levels of EOR. RESULTS: Recursive partitioning analysis identified three survival risk groups. Median OS was shortest in two subsets of patients with astrocytoma: those with postoperative tumor volume (TV) > 4.6 mL and those with preoperative TV > 43.1 mL and postoperative TV ≤ 4.6 mL. Intermediate OS was seen in patients with astrocytoma who had chemotherapy with preoperative TV ≤ 43.1 mL and postoperative TV ≤ 4.6 mL in addition to oligodendroglioma patients with either preoperative TV > 43.1 mL and residual TV ≤ 4.6 mL or postoperative residual volume > 4.6 mL. Longest OS was seen in astrocytoma patients with preoperative TV ≤ 43.1 mL and postoperative TV ≤ 4.6 mL who received no chemotherapy and oligodendroglioma patients with preoperative TV ≤ 43.1 mL and postoperative TV ≤ 4.6 mL. EOR ≥ 75% improved survival outcomes, as shown by propensity score analysis. CONCLUSION: Across both subtypes of LGG, EOR beginning at 75% improves OS while beginning at 80% improves progression-free survival. Nonetheless, maximal resection with preservation of neurological function remains the treatment goal. Our findings have implications for surgical strategies for LGGs, particularly oligodendroglioma.

Aurora-A kinase is differentially expressed in the nucleus and cytoplasm in normal Müllerian epithelium and benign, borderline and malignant serous ovarian neoplasms.

BACKGROUND: Aurora-A kinase is important for cellular proliferation and is implicated in the tumorigenesis of several malignancies, including of the ovary. Information regarding the expression patterns of Aurora-A in normal Müllerian epithelium as well as benign, borderline and malignant epithelial ovarian neoplasms is limited. METHODS: We investigated Aurora-A expression by immunohistochemistry in 15 benign, 19 borderline and 17 malignant ovarian serous tumors, and 16 benign, 8 borderline, and 2 malignant ovarian mucinous tumors. Twelve fimbriae from seven patients served as normal Müllerian epithelium controls. We also examined Aurora-A protein expression by western blot in normal fimbriae and tumor specimens. RESULTS: All normal fimbriae (n = 12) showed nuclear but not cytoplasmic Aurora-A immunoreactivity by immunohistochemistry. Benign ovarian tumors also showed strong nuclear Aurora-A immunoreactivity. Forty-eight percent (13/27) of borderline tumors demonstrated nuclear Aurora-A immunoreactivity, while the remainder (52%, 14/27) lacked Aurora-A staining. Nuclear Aurora-A immunoreactivity was absent in all malignant serous tumors, however, 47% (8/17) demonstrated perinuclear cytoplasmic staining. These results were statistically significant when tumor class (benign/borderline/malignant) was compared to immunoreactivity localization or intensity (Fisher Exact Test, p < 0.01). Western blot analysis confirmed the greater nuclear Aurora-A expression in control Müllerian epithelium compared to borderline and malignant tumors. CONCLUSION: Aurora-A kinase is differentially expressed across normal Müllerian epithelium, benign and borderline serous and mucinous ovarian epithelial neoplasms and malignant serous ovarian tumors., with nuclear expression of unphosphorylated Aurora-A being present in normal and benign neoplastic epithelium, and lost in malignant serous neoplasms. Further studies of the possible biological and clinical implications of the loss of nuclear Aurora-A expression in ovarian tumors, and its role in ovarian carcinogenesis are warranted.

Federated Learning used for predicting outcomes in SARS-COV-2 patients.

'Federated Learning' (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the "EXAM" (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.

Federated learning for predicting clinical outcomes in patients with COVID-19.

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.