Question about Ground Truth


Refer to “fastMRI: An Open Dataset and Benchmarks for Accelerated MRI” and “fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction Using Machine Learning”, I am confused about the reconstruction method of the Ground Truth. Are any sensitivity map estimations taken into the consideration before applying root-sum-of-squares (RSS)? Papers seem to say no. However, “End-to-End Variational Networks for Accelerated MRI Reconstruction” and earlier “Learning a variational network for reconstruction of accelerated MRI data”, etc. all estimate the sensitivities during the accelerated reconstruction. Won’t this affect the evaluation?

Hello @yi_zhang, the sensitivity coil maps are not necessary for reconstructing the ground truth if you have fully-sampled data, but sensitivity coil information is useful for removing aliasing patterns that arise when reconstructing subsampled data. That’s why we use them in the neural network models (which work with subsampled data), but not the ground truth (which has access to fully-sampled data).