Federated learning enables big data for rare cancer boundary detection

Although machine learning (ML) has shown promise across disciplines, out-of-sample generalizability is concerning. This is currently addressed by sharing multi-site data, but such centralization is challenging/infeasible to scale due to various limitations. Federated ML (FL) provides an alternative paradigm for accurate and generalizable ML, by only sharing numerical model updates. Here we present the largest FL study to-date, involving data from 71 sites across 6 continents, to generate an automatic tumor boundary detector for the rare disease of glioblastoma, reporting the largest such dataset in the literature (n = 6, 314). We demonstrate a 33% delineation improvement for the surgically targetable tumor, and 23% for the complete tumor extent, over a publicly trained model. We anticipate our study to: 1) enable more healthcare studies informed by large diverse data, ensuring meaningful results for rare diseases and underrepresented populations, 2) facilitate further analyses for glioblastoma by releasing our consensus model, and 3) demonstrate the FL effectiveness at such scale and task-complexity as a paradigm shift for multi-site collaborations, alleviating the need for data-sharing.

ArtifactID: Identifying artifacts in low-field MRI of the brain using deep learning

Low-field MR scanners are more accessible in resource-constrained settings where skilled personnel are scarce. Images acquired in such scenarios are prone to artifacts such as wrap-around and Gibbs ringing. Such artifacts negatively affect the diagnostic quality and may be confused with pathology or reduce the region of interest visibility. As a first step solution, ArtifactID identifies wrap-around and Gibbs ringing in low-field brain MRI. We utilized two datasets: 179 T₁-weighted pathological brain images from a 0.36 T scanner and 581 publicly available T₁-weighted brain images. Individual binary classification models were trained to identify through-plane wrap-around, in-plane wrap-around, and Gibbs ringing. Visual explanations obtained via the GradCAM method helped develop trust in the wrap-around model. The mean precision and recall metrics across the four implemented models were 97.6% and 92.83% respectively. Agreement analysis of the models and the radiologists' labels returned Cohen's kappa values of 0.768 ± 0.062, 1.00 ± 0.000, 0.89 ± 0.085, and 0.878 ± 0.103 for the through-plane wrap-around, in-plane wrap-around, and Gibbs ringing models, respectively.