Acceleration of IR-prepared Ultra Short Echo Time Imaging of Bones exploiting a Plug-and-Play Denoising Prior

Motivation: (IR-)UTE MRI enables (quantitative) investigation of bony tissue. Imaging protocols, however, are still time consuming. Goal(s): To develop a reconstruction method, which can transfer undersampled / accelerated IR-UTE scans of bone into high quality images. Approach: A thresholded Landweber algorithm was implemented, which uses both an L1-sparsity model and a pre-trained denoising convolutional network as regularizers of the physical MR model. Results: The reconstruction method was capable of delivering superior image quality compared to reconstructions based on straightforward NUFFT or iterative SENSE, especially in the case of significant undersampling. Impact: IR-UTE imaging accelerated by our proposed reconstruction based on L1-sparsity and a pre-trained denoising convolutional neural network shortens investigations by a factor of up to five, thereby facilitating further research on the topic as well as clinical transfer.