Estimating the detectability of small lesions in high resolution MR compressed images

Abstract

Quality degradation in high resolution magnetic resonance (MR) images compressed with JPEG2000 is studied. The present study combines the results of ten quantitative quality criteria with four mathematical observer models to estimate the maximum achievable compression rate (CR) that does not affect an effective diagnosis of specific pathologies related with small lesions detection such as Multiple Sclerosis or Virchow Robins disease. The graphical behavior of metrics employed is presented. At bitrate = 0.062 bpp (CR=160:1) the most compression is achieved while images still preserve the information needed for a safe diagnosis. Images compressed using smaller values than this would be no longer useful for diagnosis tasks. This result facilitates a safer use of the JPEG 2000 codec when compressing high resolution images like the images tested, and assures a safer diagnosis task when dealing with small size lesions that define certain pathology. © 2008 Springer-Verlag Berlin Heidelberg.