Related Experiment Video
Updated: Jun 14, 2025
08:05
Measuring Statistical Learning Across Modalities and Domains in School-Aged Children Via an Online Platform and Neuroimaging Techniques
Published on: June 30, 2020
7.5K
An international study presenting a federated learning AI platform for pediatric brain tumors
Edward H Lee1,2, Michelle Han3,4, Jason Wright5
1Department of Neurosurgery, Stanford University School of Medicine, Stanford, CA, USA. edward.heesung.lee@gmail.com.
Nature Communications
|September 2, 2024
Summary
Federated learning (FL) enables AI model training across hospitals without sharing patient data. FL-PedBrain improved pediatric brain tumor segmentation by 20-30% on external sites, demonstrating its real-world effectiveness.
Area of Science:
- Medical artificial intelligence
- Computational neuroscience
- Oncology
Background:
- Artificial intelligence (AI) in medicine is limited by small, non-diverse patient cohorts due to data privacy concerns.
- Federated learning (FL) offers a solution by enabling collaborative model training across institutions without direct data sharing.
Purpose of the Study:
- To introduce FL-PedBrain, a federated learning platform for pediatric posterior fossa brain tumors.
- To evaluate the performance of FL-PedBrain on a diverse, multi-center cohort, addressing data scarcity in pediatric brain tumor datasets.
Main Methods:
- Developed FL-PedBrain, an FL platform for joint tumor classification and segmentation.
- Orchestrated federated training across 19 international sites.
- Evaluated performance against centralized training and on external, out-of-network sites.
Main Results:
- FL-PedBrain showed minimal performance decrease (<1.5% classification, <3% segmentation) compared to centralized training.
- FL significantly boosted segmentation performance by 20-30% on three external sites.
- Investigated data heterogeneity and FL robustness in imbalanced, real-world scenarios.
Conclusions:
- FL-PedBrain effectively enables multi-institutional AI training for pediatric brain tumors.
- Federated learning demonstrates robustness and improved performance, especially on external datasets, overcoming data limitations.
