k -Space Deep Learning for Accelerated MRI.

Area of Science:

• Medical Imaging
• Signal Processing
• Machine Learning

Background:

• Compressed sensing MRI accelerates data acquisition by reconstructing images from undersampled k-space data.
• The annihilating filter-based low-rank Hankel matrix approach (ALOHA) is a state-of-the-art method for k-space interpolation.
• ALOHA relies on the low-rank Hankel matrix completion in the k-space domain, exploiting signal sparsity.

Purpose of the Study:

• To propose a fully data-driven deep learning algorithm for k-space interpolation.
• To adapt the deep learning approach for non-Cartesian k-space trajectories.
• To demonstrate the superiority of the proposed method over existing deep learning techniques.

Main Methods:

• A novel deep learning network inspired by the link between convolutional neural networks and Hankel matrix decomposition.
• Integration of a regridding layer to handle non-Cartesian k-space data.
• Extensive numerical experiments for performance evaluation.

Main Results:

• The proposed deep learning method consistently outperforms existing image-domain deep learning approaches for k-space interpolation.
• The algorithm effectively interpolates missing k-space data, leading to improved image reconstruction.
• The network's adaptability to non-Cartesian trajectories is demonstrated.

Conclusions:

• The developed data-driven deep learning algorithm provides a powerful new tool for k-space interpolation in compressed sensing MRI.
• This approach offers significant advantages over current methods, particularly in terms of performance and flexibility.
• The findings pave the way for more efficient and accurate MRI reconstruction techniques.