Info

rate in bpp

FIGURE 6 Scatter plot of ROI SNR: Wavelet coding. Reprinted with permission from S.M. Perlmutter, P.C. Cosman, R.M. Gray, R.A. Olshen, D. Ikeda, C.N. Adams, B.J. Betts, M. Williams, K.O. Perlmutter, J. Li, A. Aiyer, L. Fajardo, R. Birdwell, and B.L. Daniel, Image Quality in Lossy. Compressed Digital Mammograms, Signal Processing, 59:189-210, 1997. © Elsevier.

rate in bpp

FIGURE 6 Scatter plot of ROI SNR: Wavelet coding. Reprinted with permission from S.M. Perlmutter, P.C. Cosman, R.M. Gray, R.A. Olshen, D. Ikeda, C.N. Adams, B.J. Betts, M. Williams, K.O. Perlmutter, J. Li, A. Aiyer, L. Fajardo, R. Birdwell, and B.L. Daniel, Image Quality in Lossy. Compressed Digital Mammograms, Signal Processing, 59:189-210, 1997. © Elsevier.

reasonably expect that radiologists would differ on this point, and a question that addressed overall subjective quality would therefore produce a variety of interpretations from the judges. By focusing the question on the specific measurement and the radiologists' confidence in it regardless of what portion of the image contributed to that confidence level, and then by examining the relationship between actual measurement error and these subjective opinions, we hoped to obtain data relevant to the question of whether radiologists can be asked to trust their diagnoses made on processed images in which they may lack full confidence. No attempt was made to link the five possible scores to specific descriptive phrases, as is done with the mean opinion score rating system for speech. However, the radiologists were asked to try to use the whole scale. The CT subjective assessment was performed separately from the diagnostic task by three different radiologists. The phrasing of the question was very similar.

Images compressed to lower bit rates received worse quality scores as was expected. Figure 7 shows subjective score vs bit rate for the CT mediastinum study. The data are fit with a quadratic spline with a single knot. Figure 8 shows the general trend of mean subjective score vs mean bit rate for the MR study. A spline-like function that is quadratic from 0 to 2.0 bpp and linear from 2.0 to 9.0 bpp was fit to the data. The splines have knots at 0.6, 1.2, and 2.0 bpp. Figure 9 shows a spline fit of subjective score plotted against actual bit rate for the compressed levels only for the MR study. The general conclusion from the plots is that the subjective scores at the higher levels were quite close to the subjective scores on the originals, but at lower levels there was a steep drop-off of scores with decreasing bit rate.

These scores can also be analyzed by the Wilcoxon signed rank test. The paired t-test may be slightly less applicable since the subjective scores, which are integers over a very limited

0 0

Post a comment