A Hybrid Vision Transformer-BiRNN Architecture for Direct k-Space to Image Reconstruction in Accelerated MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Long scan times are a major limitation in Magnetic Resonance Imaging (MRI).
• Accelerated MRI techniques undersample k-space data, necessitating advanced reconstruction methods to address the ill-posed inverse problem.
• Current image-domain processing methods capture spatial context but often neglect crucial sequential k-space data characteristics for artifact disentanglement.

Purpose of the Study:

• To introduce a novel hybrid, dual-domain deep learning architecture for accelerated MRI reconstruction.
• To leverage both image-domain features and sequential k-space data characteristics for improved artifact suppression.
• To evaluate the proposed architecture against existing state-of-the-art methods.

Main Methods:

• Developed a hybrid deep learning architecture combining a Vision Transformer (ViT)-based autoencoder with Bidirectional Recurrent Neural Networks (BiRNNs).
• The ViT component learns features from image patches, while BiRNNs model sequential dependencies directly from k-space data.
• Evaluated the model on retrospectively undersampled neuro-MRI data with acceleration factors R=4 and R=8, using regular and random sampling patterns.

Main Results:

• The proposed dual-domain architecture significantly outperformed a standard ViT, an image-domain-only ViT autoencoder, and a UNet baseline.
• Superior performance and robustness were observed, particularly in high-acceleration and random-sampling scenarios.
• The integration of sequential k-space processing via BiRNNs proved critical for enhanced artifact suppression.

Conclusions:

• The novel hybrid, dual-domain deep learning architecture offers a robust solution for accelerated MRI.
• Processing sequential k-space data alongside image-domain features is essential for superior artifact reduction.
• This approach holds significant promise for improving the efficiency and diagnostic quality of MRI scans.