Selective evaluation of noise, blur, and aliasing artifacts in fast MRI reconstructions using a weighted perceptual difference model: Case-PDM

Abstract

The perceptual difference model (Case-PDM) is being used to quantify image quality of fast, parallel MR acquisitions and reconstruction algorithms as compared to slower, full k-space, high quality reference images. To date, most perceptual difference models average image quality over a wide range of image degradations. Here, we create metrics weighted to different types of artifacts. The selective PDM is tuned using test images from an input reference image degraded by noise, blur, or aliasing. Using an objective function based on the computation of diffusivity and edges applied to the output perceptual difference map, cortex channels in the PDM are arranged in a matrix and weighted by a 2D Gaussian function to ensure maximal response to each artifact in turn. PDM scores were compared to human ratings across a large set of MR reconstruction test images of varying quality. Human ratings (i.e. overall, noise, blur, and aliasing ratings) were obtained from a modified Double Stimulus Continuous Quality Scale experiment. For 3 different image types (brain, cardiac, and phantom), averaged r values [PDM, noise-PDM, blur-PDM, aliasing-PDM] were [0.933±0.018, 0.938±0.015, 0.727±0.106, 0.500±0.193], which with the possible exception of aliasing compared favorably inter-subject correlation [0.936±0.028, 0.856±0.064, 0.539±0.230, 0.767±0.125], respectively. With continued fine tuning, we believe that the weighted Case-PDM score will be useful for selectively evaluating artifacts in fast MR imaging.