SSFD: Self-Supervised Feature Distance as an MR Image Reconstruction Quality Metric
Evaluation of accelerated magnetic resonance imaging (MRI) reconstruction methods is imperfect due to the discordance between quantitative image quality metrics and radiologist-perceived image quality. Self-supervised learning (SSL) has become a popular pre-training tool due to its ability to capture generalizable and domain-specific feature representations of the underlying data for downstream tasks. In this study, we use SSL to extract image-level feature representations of MR images, and use those features to compute a self-supervised feature distance (SSFD) metric to assess MR image reconstruction quality. We demonstrate preliminary results showing the superiority of SSFD to common image quality metrics such as PSNR and SSIM, its robustness to image perturbations, and its ability to capture both pixel-level and global image quality information.
PDF Abstract
