Abstract #1039

Super-Resolution Musculoskeletal MRI using Deep Learning

Akshay S Chaudhari^1,2, Zhognan Fang³, Feliks Kogan¹, Jeff P Wood¹, Kathryn J Stevens^1,4, Jin Hyung Lee^2,3,5,6,7, Garry E Gold^1,2,4, and Brian A Hargreaves^1,2,6

¹Radiology, Stanford University, Palo Alto, CA, United States, ²Bioengineering, Stanford University, Palo Alto, CA, United States, ³LVIS Corporation, Palo Alto, CA, United States, ⁴Orthopaedic Surgery, Stanford University, Palo Alto, CA, United States, ⁵Neurology & Neurological Sciences, Stanford University, Palo Alto, CA, United States, ⁶Electrical Engineering, Stanford University, Palo Alto, CA, United States, ⁷Neurosurgery, Stanford University, Palo Alto, CA, United States

Near-isotropic high-resolution magnetic resonance imaging (MRI) of the knee is beneficial for reducing partial volume effects and allowing multi-planar image analysis. However, previous methods exploring isotropic resolutions, typically compromised in-plane resolution for thin slices, due to intrinsic signal-to-noise ratio (SNR) limitations. Even computer-vision-based super-resolution methods have been rarely been used in medical imaging due to limited resolution improvements. In this study, we utilize deep-learning-based 3D super-resolution for rapidly generating high-resolution thin-slice knee MRI from slices originally 2-8 times thicker. Through quantitative image quality metrics and a reader study, we demonstrate superior performance to both conventionally utilized and state-of-the-art super-resolution methods.

This abstract and the presentation materials are available to members only; a login is required.

Join Here