Results of the 2020 fastMRI Challenge for Machine Learning MR Image\n Reconstruction

Abstract

Accelerating MRI scans is one of the principal outstanding problems in the\nMRI research community. Towards this goal, we hosted the second fastMRI\ncompetition targeted towards reconstructing MR images with subsampled k-space\ndata. We provided participants with data from 7,299 clinical brain scans\n(de-identified via a HIPAA-compliant procedure by NYU Langone Health), holding\nback the fully-sampled data from 894 of these scans for challenge evaluation\npurposes. In contrast to the 2019 challenge, we focused our radiologist\nevaluations on pathological assessment in brain images. We also debuted a new\nTransfer track that required participants to submit models evaluated on MRI\nscanners from outside the training set. We received 19 submissions from eight\ndifferent groups. Results showed one team scoring best in both SSIM scores and\nqualitative radiologist evaluations. We also performed analysis on alternative\nmetrics to mitigate the effects of background noise and collected feedback from\nthe participants to inform future challenges. Lastly, we identify common\nfailure modes across the submissions, highlighting areas of need for future\nresearch in the MRI reconstruction community.\n