Multi-Coil MRI Reconstruction Challenge-Assessing Brain MRI Reconstruction Models and Their Generalizability to Varying Coil Configurations

Abstract

Deep-learning-based brain magnetic resonance imaging (MRI) reconstruction methods have the potential to accelerate the MRI acquisition process. Nevertheless, the scientific community lacks appropriate benchmarks to assess the MRI reconstruction quality of high-resolution brain images, and evaluate how these proposed algorithms will behave in the presence of small, but expected data distribution shifts. The multi-coil MRI (MC-MRI) reconstruction challenge provides a benchmark that aims at addressing these issues, using a large dataset of high-resolution, three-dimensional, T1-weighted MRI scans. The challenge has two primary goals: (1) to compare different MRI reconstruction models on this dataset and (2) to assess the generalizability of these models to data acquired with a different number of receiver coils. In this paper, we describe the challenge experimental design and summarize the results of a set of baseline and state-of-the-art brain MRI reconstruction models. We provide relevant comparative information on the current MRI reconstruction state-of-the-art and highlight the challenges of obtaining generalizable models that are required prior to broader clinical adoption. The MC-MRI benchmark data, evaluation code, and current challenge leaderboard are publicly available. They provide an objective performance assessment for future developments in the field of brain MRI reconstruction.

Keywords: benchmark; brain imaging; image reconstruction; inverse problems; machine learning; magnetic resonance imaging (MRI).

Figure 1

Representative reconstructions of the different models submitted to Track 01 (i.e., 12-channel) of the challenge for R = 5. Note that the reconstructions from the top four methods, ResoNNance 1.0 and 2.0, and The Enchanted 1.0 and 2.0, try to match the noise pattern seen in the background of the reference image, while ML-UNICAMP, Hybrid-cascade, WW-net, and TUMRI seem to have partially filtered this background noise.

Figure 2

Quality assessment comparing the fully sampled reference and the reconstruction obtained by team ResoNNance 2.0. (A) The top row shows the border of the left putamen, where the reconstructed image has a discrepancy in shape compared to the reference image (highlighted with red circles). The bottom row shows that changes in the shape of the structure are also visible in the next slice of the same subject (highlighted with red arrows). It is important to emphasize that these discrepancies are not restricted to the putamen, but a systematic evaluation of where these changes occur is out of scope for this work. (B) Illustration of a case where the expert observed rated that the deep-learning-based reconstruction improved image quality. In this figure, we can see smoothening of cortical white matter without loss of information as no changes appeared in the pattern of gyrification within cortical gray matter.

Figure 3

Representative reconstructions of the different models submitted to Track 02 of the challenge for R = 5 using the 32-channel coil.

Figure 4

Sample reconstruction illustrating artifacts (highlighted in red boxes) that seem to be present on images reconstructed by models that used coil sensitivity estimation as part of their method.

Figure 5

Three sample reconstructions, one per row, for the top two models. The Enchanted 2.0 and ResoNNance 2.0 and the reference are illustrated. The arrows in the figure indicate regions of interest that indicate deviations between the deep-learning-based reconstructions and the fully sampled reference.