Abstract:To quantitatively study the subjective and objective quality of medical image, nuclear medicine images were lossy compressed using a JPEG2000 plug-in at 10: 1 to 50:1 with a 10: 1 step to obtain images of various quality. For subjective evaluation, 60 thyroid images, 15 with a I-cm diameter artificial cold lesion and 15 with a 1-cm diameter artificial hot lesion, were compressed and analyzed in random order by 3 nuclear medicine physicians independently, giving receiver operating characteristic (ROC) confidence rating to the likelihood of lesion presence. Data from the 3 physicians were merged for ROC analysis. The area under ROC curve (AUC) was used as index of image diagnostic quality. From this subjective evaluation, the AUCs for the original, 10: 1, 20:1,30:1, 40: 1, and 50:1 images were 0.978 +/- 0.013, 0.967 +/- 0.015, 0.917 +/- 0.024, 0.913 +/- 0.024, 0.879 +/- 0.028 and 0.857 +/- 0.030 respectively, decreasing with more compressing. Paired sample t test revealed that there was no significant difference (P>0.5) between the original and the 10: 1 images while the more compressed images were significantly different from the original images (P<0.05 for 20: 1, P<0.01 for 30 to 50: 1). For objective study, simulated line source images of various noise level were compressed. Peak to total count ratio (PTCR) of line spread function of the simulated line source images were analyzed quantitatively. In the objective study, PTCR decreased as compression ratio increased. It was also revealed that the higher the noise level, the faster PTCR decreased. Bivariate correlation of the AUC and PTCR of 5% and 10% noise level was significant at the 0.01 level, with a correlation coefficient of 0.937 and 0.943 respectively. The slight decrease of PTCR for compression ratio from 10: 1 to 20: 1 suggests that images quality should be studied for compression ratio from 10:1 to 20:1 with finer compression ratio steps to gain more compression, even though the change of subjective measure is statistically significant. The dramatic decrease of PTCR and statistically significant (though numerically slight) change of AUC for compression ratio from 20:1 to 30:1 indicates substantial loss of image quality that would nullify any effort for more compression without degrading image quality.

Quantitative Comparison Of Subjective And Objective Medical Image Quality: Applied To Lossy Jpeg2000 Compression Of Nuclear Medicine Image