Meeting Banner
Abstract #4223

Evaluation of Compressed Sensing MR Reconstruction Quality Using Signed Just Noticeable Difference (JND) Analysis

Michelle Yan1, Jeff Johnson1, Xiao Chen2, Li Pan2, Ti-Chiun Chang1, Yunqiang Chen1, Tong Fang1

1Siemens Corporate Research, Princeton, NJ, United States; 2Center for Applied Medical Imaging, Siemens Corporate Research, Baltimore, MD, United States

We propose to incorporate human perception model into assessing the quality of compressed sensing MR reconstruction algorithms. More specifically, we seek to measure the perceptual changes or degradation in reconstructed MR images using the signed just noticeable difference (JND) analysis, providing an overall quality color map or score for each MR image under consideration. Red color indicates positive (added) changes; blue color indicates negative (lost) changes. The lighter a JND map in color, the better a reconstruction method performs. The result has demonstrated potential value of JND maps in assessing image quality and comparing the overall performance of MR reconstruction methods.

Keywords

absolute accepted addition alter alternatives amount anatomical appearance applied artifacts assess assessing attention better blue boundaries burden challenge channel channels clear clearly clinical color complex compressed computational computed computing consideration containing corporate decreasing define defined definition degradation desired detection developed discrimination distinguish effort evaluating evaluation every exploits explored fail fast features frequency full great happens human idea identical illustrated implementation implies important improving independently indicated indicating individual introduce introduced involving just leads lighter location loss magnitude many maps meanings measures medical metrics mind model models much natural negative noticeable observation offers offs often option orientation original output overall partial perceived percentile perception perceptual performance person physical pixel pooling potential practicality predict primarily properly propose proposed providing quality quite recognition recognize reconstructed reconstruction reconstructions relatively require response score sensing sets signed smooth spatial specifically spite stimulus structural structures summation system target tong typically ultimate undesired variation vessel visibility visual ways widely yields