Motion-aware deep learning reconstruction for accelerated free-breathing ZTE lung MRI

Motivation: Lung structural changes are important for respiratory disease diagnosis. ZTE imaging is suitable for ultra-short T2 or T2* MRI and has applications in lung imaging. Challenges in ZTE imaging include long scan time and respiratory motion. Goal(s): To develop a DL-based motion-resolved reconstruction method for faster ZTE lung imaging. Approach: Spiral phyllotaxis interleaves are sorted into four motion states using respiratory motions extracted from low-resolution ZTE images. A deep-learning network was trained to reconstruct images from accelerated acquisitions. Results: The proposed ZTE-Movienet enables 1.72-fold acceleration with scan time of ~3.5 minutes with image quality comparable to a conventional ZTE acquisition of 6.5 minutes . Impact: Deep learning motion-aware reconstruction of ZTE imaging enables accelerated motion-robust acquisition, which has the potential to promote the use of lung MRI in clinical practice.