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

Abstract

‘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.

Rapid Artificial Intelligence Solutions in a Pandemic - The COVID-19-20 Lung CT Lesion Segmentation Challenge

“…Thus, a multitude of AI approaches have been developed, published and indicated great potential for clinical support, but they were often overfit, being trained using proprietary data or from a single site [14][15][16][17][18][19] . Alternatively, federated approaches allow algorithms to access data from multiple sites without the need of sharing raw data, but through this paradigm access is granted to a single algorithm and consortium, with sharing of model weights instead of raw data 20,21 . In particular, deep neural networks were used for the identification and segmentation of abnormal lung regions affected by SARS-CoV-2 infection.…”

User-Level Label Leakage from Gradients in Federated Learning

“…Both, FL as well as the more superordinate setting of distributed ML are used in many applications where labels can contain highly sensitive information. For example, in the medical sector, hospitals employ distributed learning to collaboratively build ML models for disease diagnosis and prediction [11,17]. In some cases, the medical data is collected directly from the patients' personal devices [8], e.g., mobile phones [9], where an application of FL could introduce many potential benefits.…”

Development of a federated learning approach to predict acute kidney injury in adult hospitalized patients with COVID-19 in New York City

“…FL avoids the need to collect and store private data at a centralized location by allowing institutions to download a preliminary ML model, refine it locally, and then upload updated model parameters to an aggregator. [5][6][7] FL models have been utilized to identify COVID-19 through computed tomography scans and to predict clinical outcomes in COVID-19 patients -published reports indicate that FL models not only outperform those trained on single-institution data but also approach the performance of traditional ML models trained on data pooled from multiple institutions. 6,[8][9][10] COVID-19 has diverse clinical manifestations, and a common complication in hospitalized COVID-19 patients is acute kidney injury (AKI).…”

“…[5][6][7] FL models have been utilized to identify COVID-19 through computed tomography scans and to predict clinical outcomes in COVID-19 patients -published reports indicate that FL models not only outperform those trained on single-institution data but also approach the performance of traditional ML models trained on data pooled from multiple institutions. 6,[8][9][10] COVID-19 has diverse clinical manifestations, and a common complication in hospitalized COVID-19 patients is acute kidney injury (AKI). [11][12][13][14] Studies have reported AKI prevalence up to 46%, and mortality rates in AKI cohorts vary from 35%-71%.…”