Improving accelerated MRI by deep learning with sparsified complex data.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Accelerated MRI acquisition reduces scan times but often introduces artifacts like aliasing and ghosting.
• Complex-valued reconstruction from undersampled k-space data is crucial for preserving phase information in MRI.

Purpose of the Study:

• To develop and evaluate a complex-valued deep learning method for high-quality accelerated MRI reconstruction.
• To address ghosting artifacts in undersampled k-space data using a novel convolutional neural network architecture.

Main Methods:

• Retrospective undersampling of MRI data with skipped phase encoding.
• Application of a complex difference transform to create sparsified complex-valued edge maps.
• Training a complex-valued U-type convolutional neural network (SCU-Net) for deghosting.
• K-space inverse filtering on predicted deghosted edge maps for final image reconstruction.

Main Results:

• SCU-Net demonstrated effectiveness in deghosting aliased edge maps, even at high acceleration factors.
• High-quality complex MR images were successfully reconstructed.
• SCU-Net outperformed zero-filling, GRAPPA, RAKI, FDCU-Net, and CU-Net in qualitative and quantitative evaluations.

Conclusions:

• SCU-Net provides superior reconstruction quality for regularly undersampled k-space data using sparsified complex data.
• The proposed method is particularly advantageous for phase-sensitive MRI applications requiring accurate phase information.