A data-driven approach to accelerate IR-prepared UTE Bone Imaging of the Knee

Motivation: (IR)-UTE MRI allows for (quantitative) assessment of bony tissue. However, imaging often suffers from poor SNR and time-consuming protocols. Goal(s): To develop a reconstruction method that can transfer undersampled IR-UTE-based scans of the knee into high-quality images. Approach: A physics-based reconstruction algorithm was implemented, which incorporates a denoising convolutional neural network (DnCNN) specifically trained for the presented application. Results: Exploiting the tailored DnCNN within the iterative reconstruction algorithm leads to significant noise suppression and an overall improved image quality compared to incorporating a generic DnCNN or applying the networks alone. Impact: The proposed reconstruction technique demonstrates potential for MR-based assessment of osseous tissue within clinically appealing scan time.