English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Multi-institutional efforts can facilitate training of deep MRIreconstruction models, albeit privacy risks arise during cross-site sharing ofimaging data. Federated learning (FL) has recently been introduced to addressprivacy concerns by enabling distributed training without transfer of imagingdata. Existing FL methods for MRI reconstruction employ conditional models tomap from undersampled to fully-sampled acquisitions via explicit knowledge ofthe imaging operator. Since conditional models generalize poorly acrossdifferent acceleration rates or sampling densities, imaging operators must befixed between training and testing, and they are typically matched acrosssites. To improve generalization and flexibility in multi-institutionalcollaborations, here we introduce a novel method for MRI reconstruction basedon Federated learning of Generative IMage Priors (FedGIMP). FedGIMP leverages atwo-stage approach: cross-site learning of a generative MRI prior, andsubject-specific injection of the imaging operator. The global MRI prior islearned via an unconditional adversarial model that synthesizes high-quality MRimages based on latent variables. Specificity in the prior is preserved via amapper subnetwork that produces site-specific latents. During inference, theprior is combined with subject-specific imaging operators to enablereconstruction, and further adapted to individual test samples by minimizingdata-consistency loss. Comprehensive experiments on multi-institutionaldatasets clearly demonstrate enhanced generalization performance of FedGIMPagainst site-specific and federated methods based on conditional models, aswell as traditional reconstruction methods.

Federated Learning of Generative Image Priors for MRI Reconstruction