Deep Learning Image Reconstruction for Enhancing Image Quality and Acquisition Speed in 3T EDGE-MP2RAGE.

BACKGROUND AND PURPOSE: The three-dimensional edge-enhancing gradient echo (3D EDGE) MRI sequence is valuable in epilepsy imaging due to improved visualization of cortical lesions by enhancing the gray matter-white matter boundary contrast. However, EDGE suffers from low SNR and long acquisition times at 3T. This study evaluated a 3T deep learning (DL)-based reconstruction implemented in the EDGE-MP2RAGE sequence and compared its image quality with standard 3T and ultra-high-field 7T EDGE-MP2RAGE MRI. MATERIALS AND METHODS: In this retrospective study, 117 patients underwent 3T DL, standard 3T, or 7T EDGE-MP2RAGE MRI (n=39 per group). The EDGE images underwent quantitative and qualitative comparison of image quality. Quantitative analysis included contrast-to-noise ratio (CNR) and image noise. Qualitative image quality was independently and blindly scored by two neuroradiologists using a Likert scale. Statistical comparisons used Mann-Whitney U tests with Bonferroni correction. RESULTS: Compared with standard 3T EDGE-MP2RAGE, the 3T DL EDGE-MP2RAGE sequence reduced scan time by 47.7%, increased CNR by 85.9% (P<.001), and decreased image noise by 22.6% (P=.005). Qualitatively, 3T DL EDGE demonstrated significantly higher overall image quality, sharpness, structural conspicuity, and contrast uniformity (P<.001) without introducing additional artifacts. Compared with 7T, 3T DL EDGE achieved comparable CNR (P=.17) and noise (P=.77), while showing superior contrast uniformity (P<.001). CONCLUSIONS: DL-based reconstruction improves image quality in 3T EDGE-MP2RAGE while simultaneously decreasing scan time by nearly half versus conventional 3T EDGE-MP2RAGE.