SCAMPI -- database-free neural network reconstruction for undersampled Magnetic Resonance Imaging

Purpose: Reducing acquisition time in MRI is mainly based on vast undersampling of the data measured in the spatial frequency domain. The resulting artifacts are reduced by various reconstruction algorithms. Here, we present SCAMPI (Sparsity Constrained Application of deep Magnetic resonance Priors for Image reconstruction), an untrained deep Neural Network, that utilizes the Deep Image Prior approach together with sparsity constraints to reconstruct undersampled MRI data without any previous training on external datasets. Methods: Two-dimensional MRI data from the FastMRI dataset with Cartesian undersampling in phase-encoding direction were reconstructed for different acceleration rates for single coil and multicoil data. Results: The performance of our architecture was compared to state-of-the-art Compressed Sensing methods and ConvDecoder, another untrained Neural Network for MRI reconstruction. SCAMPI outperforms these by better reducing undersampling artifacts and yielding lower error metrics in multicoil imaging. Moreover, it is a novel tool for reconstructing undersampled single coil k-space data. Furthermore, SCAMPI can reconstruct multicoil data without explicitknowledge of coil sensitivity profiles. Conclusion: The presented approach avoids overfitting to dataset features, that can occur in Neural Networks trained on databases. It allows better results and faster reconstruction than the baseline untrained Neural Network approach.