Study Published In Nature Medicine Shows Positive Impact Of Leveraging Federated Learning For Healthcare AI

The Rhino Health Platform is now building on the study findings, improving patient outcomes by facilitating global research collaboration and protecting data privacy

Rhino Health announced that the EXAM study led by its co-founder and CEO Ittai Dayan, MD, and Mona Flores, MD, global head of medical AI for NVIDIA, has been published in Nature Medicine. This is the world’s largest and most prominent study to-date utilizing federated learning (FL) to train a healthcare AI solution on diverse data across institutions. With FL, artificial intelligence models are trained using data from disparate sources — without sharing or aggregating data. This protects privacy, facilitates access to more diverse datasets, and makes it easier for AI researchers and developers around the world to collaborate.

Recommended AI News: Microsoft Announces Quarterly Dividend Increase and New Share Repurchase Program

“Federated learning promises to accelerate creation and adoption of AI-based healthcare solutions that work more effectively for more patients in more places,” said Dr. Flores. “EXAM demonstrates the positive impact of leveraging diverse and disparate datasets and underscores the potential for more widespread usage of the FL-based approach.”

The EXAM study, conducted in Fall 2020, brought together 20 institutions across North and South America, Europe, and Asia to create an AI model that could accurately predict future oxygen requirements of an individual seen in the emergency department with COVID-19 symptoms. Inputs included vital signs, laboratory data, and chest x-rays. EXAM achieved an average area under the curve (AUC) of over 0.92, an average improvement of 16% and a 38% increase in generalizability over local models. The model remained robust to a gold-standard validation conducted at three separate, independent sites. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as setting the stage for broader use of FL in healthcare. A full description and complete findings are available here.

“Federated learning is a promising tool that can accelerate the clinical translation of AI models by ensuring they retain a consistent level of performance across today’s real-world patient ecosystem,” said Christoph Wald, MD, PhD, MBA, FACR, who serves as Radiology Chair at Lahey Hospital and Medical Center and is Chair of the American College of Radiology (ACR) commission on informatics. “This large-scale study is the first to demonstrate the impact of federated learning on patient outcomes globally and provides a strong foundation for continued improvement and accelerated adoption of advanced AI solutions.”

Recommended AI News: Vista Equity Partners Expands Team With Experienced Communications Executive

“Seeing firsthand the transformative power of federated learning made it clear to me that the health ecosystem needs to rethink how we manage data-sharing,” said Fiona Gilbert, MD, chair of Radiology at the University of Cambridge School of Medicine. “The ability to collaborate globally while always prioritizing and protecting privacy is essential to innovations in diagnostics and treatment pathways that will improve outcomes for today’s large and increasingly diverse patient populations.”

Currently, Rhino Health is collaborating with several partners including the Center for Advanced Medical Computing and Analysis (CAMCA), Massachusetts General Hospital (MGH), University of Cambridge School of Clinical Medicine, and several other prominent academic medical centers and research consortia to implement the learnings from EXAM. The goal is to make global collaborations like EXAM repeatable, sustainable, much faster and more effective. Ultimately, this will accelerate creation and deployment of AI algorithms that help clinicians tackle some of the most complex challenges in healthcare — applying the learnings from this pandemic to pressing unmet patient needs and other epidemics.

“The only way to bring model improvement into the clinical workflow is through federated learning,” said Dr. Quanzheng Li of the Gordon Center for Medical Imaging at Massachusetts General Hospital, who developed the initial model on which the EXAM study was based. “This will allow us to learn deep insights from the longitudinal patient pathway, both locally at a specific site and across any place the model is deployed.”

Recommended AI News: Ping Identity Names Jason Kees Chief Information Security Officer