Deep Learning Accelerated 3D SPACE DIXON for Improved Fat Suppression in Musculoskeletal MRI

Motivation: The application of 3D SPACE-DIXON techniques for musculoskeletal MRI has advantages in terms of fat suppression capacities, but drawbacks with regard to substantially longer acquisition times compared with SPAIR fat suppressed 3D SPACE. Goal(s): To reduce the acquisition time of 3D SPACE-DIXON to that of 3D SPACE-SPAIR while maintaining superior fat suppression capabilities. Approach: Six-fold CAIPIRINHA acceleration was combined with a deep learning-based image reconstruction algorithm to decrease the acquisition time and maintain image quality. Results: Image quality parameters in volunteer scans were rated equivalent to the conventional 3D SPACE-DIXON sequence and superior to the 3D SPACE-SPAIR sequence. Impact: We present a 3D SPACE-DIXON technique with improved fat suppression performance and similar acquisition time compared with 3D SPACE-SPAIR for musculoskeletal 3D MRI. Preliminary in vivo results indicate the clinical utility of this technique for proton density-weighted 3D MRI.