Benchmarking MRI Reconstruction Neural Networks on Large Public Datasets

Abstract

Deep learning is starting to offer promising results for reconstruction in Magnetic Resonance Imaging (MRI). A lot of networks are being developed, but the comparisons remain hard because the frameworks used are not the same among studies, the networks are not properly re-trained, and the datasets used are not the same among comparisons. The recent release of a public dataset, fastMRI, consisting of raw k-space data, encouraged us to write a consistent benchmark of several deep neural networks for MR image reconstruction. This paper shows the results obtained for this benchmark, allowing to compare the networks, and links the open source implementation of all these networks in Keras. The main finding of this benchmark is that it is beneficial to perform more iterations between the image and the measurement spaces compared to having a deeper per-space network.

Calibration-Less Multi-Coil Compressed Sensing Magnetic Resonance Image Reconstruction Based on OSCAR Regularization

“…The proposed method is also made of fully interpretable steps which make it reliable to perturbations in the data set (e.g., bad channel in the phased array coil with low SNR or small motion compensated in Non-Cartesian sampling due to variable density sampling) than could not be anticipated in recent deep learning approaches [ 59 , 60 , 61 ] proposed for medical image reconstruction. In the latter, sometimes a minor modification in the unseen test data set that was not met in the training and validation sets leads to major artifacts on reconstructed images [ 62 ].…”

Special Issue on Signal Processing and Machine Learning for Biomedical Data

“…The authors of [21] presented a benchmark of several deep neural networks for MRI reconstruction. The methods analyzed for the problem of MRI reconstruction made use of two databases "fastMRI" and "OASIS".…”

Machine learning in Magnetic Resonance Imaging: Image reconstruction

“…These include (but are not limited to) raw k-space data sets; mridata.org, NYU fastMRI [93] and Calgary-Campinas-359 [94], as well as DICOM imaging data sets; UK Biobank [95], Hunan Connectome Project [96], The Montreal Neurological Institute's Brain Images of Tumors for Evaluation (NTI BITE) [97] and OASIS-3 [98]. The availability of these datasets enables development of novel DL image reconstruction frameworks, as well as making it possible to benchmark and compare networks in the same setting [99].…”