Generation of high-resolution MPRAGE-like images from head MRI localizer images

Motivation: MPRAGE is useful for voxel-based morphometry (VBM) but is not a priority in routine clinical examinations. We hypothesized that converting MRI localizer (AutoAlignHead, AAH) to MPRAGE would be beneficial for subjects who did not have MPRAGE. Goal(s): We aimed to establish and evaluate a Pix2Pix-based model for generating MPRAGE-like images from AAH. Approach: Training, validation, and test datasets of 340 patients were included. The test set was evaluated with visual examination by radiologists and quality metrics of PSNR, SSIM, and LPIPS. VBM using Freesurfer was also performed. Results: Visual examination, image metrics, and VBM showed comparable scores between generated and original images. Impact: MPRAGE-like images generated from MRI localizer images and the reference MPRAGE image showed good agreement with respect to visual assessment of medial temporal lobe atrophy by radiologists. Voxel-based morphological analysis was also acceptable for evaluation of temporal lobe atrophy.