Up to four times accelerated musculoskeletal MRI at 0.4T using the CIRIM-network

Motivation: Low-field MRI scanners are gaining attention for being cost-effective, increasing the accessibility of MRI worldwide. Still, the scan times of low-field MRI are high, necessitating acceleration techniques. Goal(s): To accelerate 0.4T knee and spine MRI using deep learning reconstruction. Approach: A neural network (the CIRIM) was trained to reconstruct undersampled data. The undersampling pattern and loss function of the CIRIM were optimized, and different acceleration factors were explored. Results: The CIRIM successfully reconstructed accelerated knee and spine data. For 2D and 3D images, some minor blurring was seen beyond an acceleration of 3 and 4, respectively. Impact: Low-field MRI is cost-effective and can therefore increase the worldwide accessibility of MRI. By accelerating imaging using deep learning reconstruction on undersampled data, we realize time-efficient scanning.