Deep Learning-Based Chemical Shift Artifact-Corrected Zero-Echo Time MRI for the Cervical Spine

Motivation: Zero echo-time (ZTE) MRI can allow for assessing bone morphology and abnormality, but low SNR and chemical shift artifacts can make analysis of high-resolution bone structure challenging. Goal(s): Our goal was to evaluate whether deep learning-based chemical shift artifact-correction (DLCSC) reconstruction for ZTE can allow for better assessment of bone morphology and pathological changes in the cervical spine. Approach: DLCSC reconstruction was applied to ZTE raw data from patients with cervical spine pathology, and the ability to visualize bone pathology was assessed qualitatively and quantitatively in comparison to CT. Results: DLCSC reconstruction greatly improved ZTE image quality and enhanced diagnostic performance. Impact: Deep learning-based chemical shift artifact-correction reconstruction can greatly improve SNR and reduce chemical shift artifacts for ZTE imaging, and allow for bone visualization close to CT. ZTE can be a radiation-free alternative to CT.