Dec 9, 2022
UPenn-led study brought together almost 6,500 brain cancer MRIs from 71 sites on 6 continents to study the benefits of federated-learning-based AI on glioblastoma segmentation.
Neosoma today applauded the publication of an article in Nature Communications detailing a large, multi-site federated learning glioblastoma (GBM) imaging study led by researchers at the University of Pennsylvania.
The purpose of the study was to determine whether many institutions across the globe could effectively utilize imaging data, without the raw data leaving each site, to train an artificial-intelligence (AI)-based software model to segment GBM tumors, and whether the performance of this model would surpass a baseline model trained with a more traditional approach.
The study applied 6,314 GBM brain tumor MRIs from 71 sites across 6 continents, including Neosoma, to a decentralized AI algorithm training process called "federated learning."
Performance of the federated AI model exceeded that of the baseline model, proving that federated learning can be effectively applied to complex medical software use cases, and can ultimately benefit patients.
Read the full article in Nature Communications.