Fig. 6

Workflow of neural network-based MRI reconstruction from undersampled k-space data: This figure outlines the workflow for reconstructing high-quality MRIs from undersampled k-space data using a neural network. Starting with undersampled Cartesian k-space data, an Inverse Fourier Transform (iFFT) generates an artefact-laden complex image. This image is fed into a neural network to produce a reconstructed complex image. The network’s output undergoes a Fast Fourier Transform (FFT) to form output k-space, which is combined with the original sampled k-space data using a missing mask. A final Inverse Fourier Transform (iFFT) converts the combined k-space back to the spatial domain, resulting in the final output image. The image is then compared to the ground truth for loss calculation and evaluation, guiding the network to improve its reconstruction accuracy.