Utility of Deep Learning Super-Resolution in the Context of Osteoarthritis MRI Biomarkers

Deep learning based super-resolution (SR) is a computer vision method that can enhance the resolution of low-resolution images, which has recently been applied to MRI. Despite enhancing the appearance of images, it is unclear what the impact of SR is in evaluating potential quantitative biomarkers of knee osteoarthritis. In this study, we evaluate the variations in biomarkers of the cartilage and osteophytes when compared with low-resolution images, the super-resolution (DeepResolve) images, and the original high-resolution images. We demonstrate that the SR images do not appear to bias quantitative biomarkers and offer a better alternative to traditional interpolation methods to enhance the appearance of MR images.

Chaudhari AS, Stevens KJ, Wood JP, Chakraborty AK, Gibbons EK, Fang Z, Desai AD, Lee JH, Gold GE, Hargreaves BA. Utility of deep learning super-resolution in the context of osteoarthritis MRI biomarkers. J Magn Reson Imaging. 2019 Jul 16. doi: 10.1002/jmri.26872.

Online Journal Article

A comparison of the 3x low-resolution tricubic interpolated images that are enhanced using the DeepResolve super-resolution algorithm, as well as the original high-resolution images.

Akshay Chaudhari
Assistant Professor (Research) of Radiology (Integrative Biomedical Imaging Informatics at Stanford) and, by courtesy, of Biomedical Data Science
Kate Stevens
Associate Professor of Radiology (Musculoskeletal Imaging)
Arjun Desai
Ph.D. Student in Electrical Engineering, admitted Autumn 2019
Jin Hyung Lee
Associate Professor of Neurology (Neurology Research Faculty), of Neurosurgery and of Bioengineering and, by courtesy, of Electrical Engineering
Garry Gold
Stanford Medicine Professor of Radiology and Biomedical Imaging
Brian A. Hargreaves
Professor of Radiology (Radiological Sciences Laboratory) and, by courtesy, of Electrical Engineering and of Bioengineering