Deep Learning-based Super-Resolution reconstruction for Fast T1 and T2-weighted Head and Neck MRI

Motivation: Current acceleration techniques for head and neck MRI face trade-offs between acquisition speed and image quality. Goal(s): To evaluate a novel deep learning-based reconstruction framework integrating compressed sensing and super-resolution techniques for T1- and T2-weighted head and neck MRI. Approach: We prospectively enrolled 54 patients who underwent paired conventional and DL-reconstructed sequences, with quantitative and qualitative assessment by two radiologists. Results: The framework achieved 46.3% and 26.9% reduction in acquisition time for T1WI and T2WI respectively, with significant improvements in SNR (both P<0.001), CR (both P<0.001), and superior qualitative scores in image sharpness, lesion conspicuity, and overall image quality (all P<0.05). Impact: This integrated deep learning framework offers a clinically viable solution for accelerated head and neck MRI acquisition while enhancing image quality, potentially improving workflow efficiency and patient comfort in routine clinical practice.