To assess the advantages of the Joint Photographic Experts Group (JPEG)2000 3D (part 2) over JPEG2000 in compressing thin-section abdomen CTdata sets, 60 thin-section (0.67 mm) scans from 35 males and 25 females, ranging from 23 to 95 years of age (mean, 58 years), were compressed reversibly (as a negative control) and irreversibly to 4:1, 6:1, 8:1, 10:1, and 12:1 using JPEG2000 3D and JPEG2000 algorithms. Encoding and decoding times and peak signal-to-noise ratios (PSNRs) were measured. For 60 (one image per scan) representative sections containing abnormalities, three radiologists independently compared original and compressed images and graded compression artifacts as 0 (none, indistinguishable), 1 (barely perceptible), 2 (subtle), or 3 (significant). According to pooled radiologists’ responses, the range of visually lossless threshold (VLT, the highest compression ratio at which a compressed image is indistinguishable from its original) was determined as one of , 4:1–6:1, 6:1–8:1, 8:1–10:1, 10:1–12:1, and . Wilcoxon signed rank tests and exact tests for paired proportions were used for the comparisons between the two compressions. At each irreversible compression ratio, compared to JPEG2000, JPEG2000 3D required two- or threefold greater computing times and introduced less artifacts in terms of PSNR and the grade ( at 6:1 or higher) and the presence of perceived artifacts (, at 6:1 for all readers and at 8:1 for two readers). According to PSNR and readers’ responses, 6:1 and 8:1 JPEG2000 3D compressions showed more artifacts than 4:1 and 6:1 JPEG2000 compressions, respectively, and 10:1 and 12:1 JPEG2000 3D compressions showed similar artifacts to those of 8:1 and 10:1 JPEG2000 compressions, respectively. The determined VLT range was higher for JPEG2000 3D than for JPEG2000 : the 3D compression showed the VLT ranges of 4:1–6:1, 6:1–8:1, and 8:1–10:1 for 24 (40%), 30 (50%), and 6 (10%) of the 60 original images, respectively, while the 2D compression showed the VLT ranges of , 4:1–6:1, and 6:1–8:1 for 1 (1.7%), 40 (66.7%), and 19 (31.6%) images, respectively. Compared to JPEG2000, JPEG2000 3D increased the VLT range in 23 of the 60 original images by one or two ranges , while the remaining 37 images had the same VLT range between the two compressions. In conclusion, compared to JPEG2000 compression, JPEG2000 3D compression yields less artifacts in compressing thin-section abdomen CTimages but requires significantly greater computing times. For the tested data set compressed to the range from 4:1 to 12:1, JPEG2000 3D could increase compression level reasonably (by 2 or less in terms of compression ratio) compared to JPEG2000 for the same amount of artifacts.
This work was supported by the Brain Korea 21 Project in 2008 and by the Korea Science and Engineering Foundation (KOSEF) grant funded by Ministry of Science and Technology (MOST), Republic of Korea (No. R01-2008-000-10055-0). The ICT at Seoul National University provides research facilities for this study.
II. MATERIALS AND METHODS
II.A. Scan and image selection
II.B. CT scanning
II.C. Image compression
II.E. Radiologists’ visual analysis
II.F. Visually lossless threshold (VLT)
II.G. Artifact pattern
II.H. Statistical analysis
III.A. Reversible compression ratio
III.B. Encoding and decoding times (Fig. 2)
III.C. PSNR (Figs. 3 and 4)
III.D. Human visual analyses
III.F. Artifact pattern
Data & Media loading...
Article metrics loading...
Full text loading...