Artifacts map-guided nonlocal mean denoising for abdominal 4-dimensional magnetic resonance imaging using view-sharing acquisition for radiotherapy motion management

The clinical adaptation of 4D-MRI in respiratory motion management is limited by the low image quality and motion artifacts of 4D-MRI sequences. This study aims to develop a novel artifact Map-guided Nonlocal mean (AM-NLM) technique that can be integrated into the clinical 4D-MRI workflow to suppress motion artifacts and enhance image quality. The AM-NLM technique was developed and tested on 4D-MR images of 28 liver cancer patients. A multiphase motion field was computed on the frames with the minimum average localized gradient entropy for each phase to generate a full set of improved quality 4D-MR images. Artifact maps were calculated based on the local image sharpness to guide nonlocal averaging, and a set of denoised eight-phase 4D-MR images was finally generated. The 4D-MR images were evaluated for image quality and motion accuracy. Conventional 4D-MRI approaches were also evaluated for comparison. AM-NLM 4D-MR images have significant improvements in SNR and CNR compared to the original 4D-MR images. High motion accuracy was achieved for AM-NLM 4D-MR images because the average deviation in the diaphragm position from the mean value for each phase was at the subvoxel level. Both qualitative and quantitative results suggested that the 4D-MR images generated by the AM-NLM technique had high image quality while maintaining image sharpness and motion accuracy. The AM-NLM technique has shown capability of suppressing motion artifacts and enhancing image quality of clinically acquired 4D-MR images, making it a promising technique in applications of 4D-MRI in radiotherapy.