Initial clinical experience of mineral bone imaging using a ZTE sequence reconstructed with Deep-Learning and Chemical Shift Correction (DLCSC)

Motivation: Bone MRI using ZTE has the potential to provide clinically relevant information on mineral bone but its routine use remains limited by its low SNR and chemical shift artifacts especially at 3.0TGoal(s): We evaluated the impact of a deep learning reconstructions for ZTE which performs denoising, improves resolution and minimizes chemical shift artifacts (DLCSC). Approach: ZTE sequences prospectively obtained in 10 patients were reconstructed with the DLCSC algorithm. Native and processed images were compared for their performance to detect bone lesions, taking CT as gold standard. Results: DLCSC increased image quality and significantly improved bone lesion detection compared to native ZTE images. Impact: DLCSC reconstruction overcomes the pitfalls of the ZTE sequence (low SNR and chemical shift artifacts) and improves its diagnostic performance. A complete study of the skeleton, including mineral bone assessment, becomes possible within a single MRI examination.