Accelerated cardiac cine with spatio-coil regularized deep learning reconstruction.

Area of Science:

• Medical Imaging
• Deep Learning
• Cardiovascular Imaging

Background:

• Accelerated imaging techniques are crucial for reducing scan times in cardiac cine MRI.
• Deep learning (DL) based reconstruction methods offer potential for improving image quality and acceleration.
• Current methods face challenges in balancing acceleration factors with diagnostic image fidelity.

Purpose of the Study:

• To develop an iterative deep learning (DL) reconstruction method with spatio-coil regularization and multichannel k-space data consistency for accelerated cine imaging.
• To evaluate the performance of the proposed Spatio-Coil Regularized DL (SCR-DL) approach against existing reconstruction techniques.

Main Methods:

• The Spatio-Coil Regularized DL (SCR-DL) approach was developed, incorporating multicoil information into data consistency and regularizer terms.
• SCR-DL utilizes shift-invariant convolutional kernels for k-space interpolation and individual coil image reconstruction.
• The method was compared with GRAPPA, SENSE-DL, and STR-DL at 8-fold acceleration using retrospective undersampled cine data, with quantitative evaluation via NMSE and SSIM.

Main Results:

• SCR-DL successfully reconstructed highly accelerated cine images with significant improvements in NMSE (0.03 ± 0.02) and SSIM (91.4% ± 2.7%) compared to GRAPPA, SENSE-DL, and STR-DL (p < 0.001).
• Prospectively accelerated cine images at 2-fold and 8-fold accelerations showed no significant difference in left-ventricular ejection fraction or mass, despite a 3-fold reduction in scan time.

Conclusions:

• SCR-DL enables highly accelerated cardiac cine imaging, substantially reducing breath-hold duration.
• The SCR-DL method demonstrates superior image quality (NMSE and SSIM) compared to GRAPPA and SENSE-DL for accelerated cardiac cine imaging.