Accelerated Phase Contrast MRI with Use of K-Space Restoration and Resolution Enhancement Generative Adversarial Neural Network

Motivation: Phase-contrast (PC) MRI evaluates blood flow in cardiovascular disease. However, the prolonged scan times limit its efficiency. Goal(s): We sought to develop a highly accelerated PC technique based on omitting high-frequency k-space regions along the phase encoding direction. Approach: A deep learning k-space restoration and enhancement strategy for training (KREST) was developed to improve resolution while maintaining phase information. KREST was trained and tested with PC images from 1600 patients. Results: In a prospective study of 16 patients, KREST reduced breath-hold time relative to parallel imaging (19 vs 6 s). Impact: Our k-space restoration and enhancement strategy enables resolution-enhancement while providing k-space data consistency. Deep learning accelerated phase-contrast imaging showed similarly accurate quantification of peak mean velocity to a standardized parallel imaging method.