An international study presenting a federated learning AI platform for pediatric brain tumors.

in Nature communications by Edward H Lee, Michelle Han, Jason Wright, Michael Kuwabara, Jacob Mevorach, Gang Fu, Olivia Choudhury, Ujjwal Ratan, Michael Zhang, Matthias W Wagner, Robert Goetti, Sebastian Toescu, Sebastien Perreault, Hakan Dogan, Emre Altinmakas, Maryam Mohammadzadeh, Kathryn A Szymanski, Cynthia J Campen, Hollie Lai, Azam Eghbal, Alireza Radmanesh, Kshitij Mankad, Kristian Aquilina, Mourad Said, Arastoo Vossough, Ozgur Oztekin, Birgit Ertl-Wagner, Tina Poussaint, Eric M Thompson, Chang Y Ho, Alok Jaju, John Curran, Vijay Ramaswamy, Samuel H Cheshier, Gerald A Grant, S Simon Wong, Michael E Moseley, Robert M Lober, Mattias Wilms, Nils D Forkert, Nicholas A Vitanza, Jeffrey H Miller, Laura M Prolo, Kristen W Yeom

TLDR

  • FL-PedBrain is a platform that helps doctors use AI to better diagnose and treat brain tumors in children. The platform works by training AI models on data from many hospitals around the world, without sharing the data directly. This way, doctors can use the AI models to help them make better decisions, even if they don't have access to all the data. The study found that the AI models trained using federated learning performed almost as well as those trained using all the data in one place. The study also found that federated learning can help doctors make better decisions when there isn't enough data to train the AI models. The study supports the use of federated learning in medical AI applications, particularly in tertiary care hospitals where data sharing is limited due to privacy concerns. The study also highlights the importance of addressing data heterogeneity in future federated learning studies to improve performance and robustness.

Abstract

While multiple factors impact disease, artificial intelligence (AI) studies in medicine often use small, non-diverse patient cohorts due to data sharing and privacy issues. Federated learning (FL) has emerged as a solution, enabling training across hospitals without direct data sharing. Here, we present FL-PedBrain, an FL platform for pediatric posterior fossa brain tumors, and evaluate its performance on a diverse, realistic, multi-center cohort. Pediatric brain tumors were targeted due to the scarcity of such datasets, even in tertiary care hospitals. Our platform orchestrates federated training for joint tumor classification and segmentation across 19 international sites. FL-PedBrain exhibits less than a 1.5% decrease in classification and a 3% reduction in segmentation performance compared to centralized data training. FL boosts segmentation performance by 20 to 30% on three external, out-of-network sites. Finally, we explore the sources of data heterogeneity and examine FL robustness in real-world scenarios with data imbalances.

Overview

  • FL-PedBrain is an FL platform for pediatric posterior fossa brain tumors. The study evaluates its performance on a diverse, realistic, multi-center cohort. The platform orchestrates federated training for joint tumor classification and segmentation across 19 international sites. The primary objective is to assess the performance of FL-PedBrain in comparison to centralized data training and explore its robustness in real-world scenarios with data imbalances. The study aims to answer the question: Can federated learning improve the performance of AI models for pediatric posterior fossa brain tumors?
  • Comparative Analysis & Findings:
  • FL-PedBrain exhibits less than a 1.5% decrease in classification and a 3% reduction in segmentation performance compared to centralized data training. FL boosts segmentation performance by 20 to 30% on three external, out-of-network sites. The study identifies data heterogeneity as a source of performance differences between centralized and federated training. The findings suggest that federated learning can improve the performance of AI models for pediatric posterior fossa brain tumors, especially in scenarios with data imbalances. The study supports the use of federated learning in medical AI applications, particularly in tertiary care hospitals where data sharing is limited due to privacy concerns. The findings also highlight the importance of addressing data heterogeneity in future federated learning studies to improve performance and robustness.

Comparative Analysis & Findings

  • FL-PedBrain exhibits less than a 1.5% decrease in classification and a 3% reduction in segmentation performance compared to centralized data training. FL boosts segmentation performance by 20 to 30% on three external, out-of-network sites. The study identifies data heterogeneity as a source of performance differences between centralized and federated training. The findings suggest that federated learning can improve the performance of AI models for pediatric posterior fossa brain tumors, especially in scenarios with data imbalances. The study supports the use of federated learning in medical AI applications, particularly in tertiary care hospitals where data sharing is limited due to privacy concerns. The study also highlights the importance of addressing data heterogeneity in future federated learning studies to improve performance and robustness.

Implications and Future Directions

  • The study's findings demonstrate the potential of federated learning in improving the performance of AI models for pediatric posterior fossa brain tumors. The results suggest that federated learning can address data sharing and privacy concerns in medical AI applications. The study identifies data heterogeneity as a source of performance differences between centralized and federated training. Future studies should address data heterogeneity by developing robust algorithms and incorporating diverse data sources. The study also highlights the importance of exploring the performance of federated learning in real-world scenarios with data imbalances. Future research should investigate the impact of data imbalances on federated learning performance and develop strategies to mitigate their effects. The study's findings support the use of federated learning in medical AI applications, particularly in tertiary care hospitals where data sharing is limited due to privacy concerns. The study also highlights the importance of addressing data heterogeneity in future federated learning studies to improve performance and robustness.
Read Full Article