Multi-Image Reconstruction in Multi-Contrast MRI

The acquisition of multiple contrast magnetic resonance images (MRI) has an important role in clinical diagnosis by increasing diagnostic knowledge. Long scanning durations, in which the patient must remain immobile, limits the acquisition of multiple contrast MRIs. Scanning times can be reduced by undersampled acquisitions and reconstruction of undersampled images. Common methods produce a fully sampled MR image of a single contrast from undersampled MR image of the same contrast. However, limited prior information of a single contrast MR image in the input data limits the reconstruction performance. Hence, reconstruction performance can be increased with the use of multiple contrast MRI as input data. In this study, a multi-contrast MRI reconstruction method is proposed which simultaneously produces fully sampled images from undersampled images of more than one contrast. This method is applied using generative adversarial network that produce highly realistic images by better recovering the high frequency details. The proposed method has been tested on a dataset containing multiple contrast brain MR images, and it has been demonstrated that it provides superior performance compared to the alternative single contrast reconstruction method as a result of numerical and visual evaluations.