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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Global correlations are widely seen in human anatomical structures due tosimilarity across tissues and bones. These correlations are reflected inmagnetic resonance imaging (MRI) scans as a result of close-range protondensity and T1/T2 parameters. Furthermore, to achieve accelerated MRI, k-spacedata are undersampled which causes global aliasing artifacts. Convolutionalneural network (CNN) models are widely utilized for accelerated MRIreconstruction, but those models are limited in capturing global correlationsdue to the intrinsic locality of the convolution operation. Theself-attention-based transformer models are capable of capturing globalcorrelations among image features, however, the current contributions oftransformer models for MRI reconstruction are minute. The existingcontributions mostly provide CNN-transformer hybrid solutions and rarelyleverage the physics of MRI. In this paper, we propose a physics-basedstand-alone (convolution free) transformer model titled, the Multi-headCascaded Swin Transformers (McSTRA) for accelerated MRI reconstruction. McSTRAcombines several interconnected MRI physics-related concepts with thetransformer networks: it exploits global MR features via the shifted windowself-attention mechanism; it extracts MR features belonging to differentspectral components separately using a multi-head setup; it iterates betweenintermediate de-aliasing and k-space correction via a cascaded network withdata consistency in k-space and intermediate loss computations; furthermore, wepropose a novel positional embedding generation mechanism to guideself-attention utilizing the point spread function corresponding to theundersampling mask. Our model significantly outperforms state-of-the-art MRIreconstruction methods both visually and quantitatively while depictingimproved resolution and removal of aliasing artifacts.

Multi-branch Cascaded Swin Transformers with Attention to k-space Sampling Pattern for Accelerated MRI Reconstruction