Learning-based super-resolution technique significantly improves detection of coronary artery stenoses on 1.5T whole-heart coronary MRA

Background Whole-heart coronary MRA permits noninvasive detection of coronary artery disease. In the brain MRI, superresolution (SR) technique has emerged as a method to enhance image resolution and improve diagnostic performance. In the current study, we developed a learning-based SR technique optimized for whole-heart coronary MRA (Figure 1). Summarizing briefly, this SR algorithm consists of (1) construction of the dictionary for lowand high-resolution pairs of image patches, and (2) generation of high-resolution coronary MRA images by embedding optimal patches selected form the dictionary by least square fitting. The purpose of this study was to determine if the SR technique can improve the diagnostic performance of 1.5T whole-heart coronary MRA in detecting significant stenosis on catheter coronary angiography.