| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Computer Applications |
1 From the Department of Diagnostic Radiology, Yale University School of Medicine, 2-332 SP, 333 Cedar St, New Haven, CT 06520. Received October 27, 1999; revision requested December 7; revision received April 6, 2000; accepted April 20. Address correspondence to J.A.B. (e-mail: James.Brink@yale.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: A digital test pattern (Society of Motion Picture and Television Engineers, 512 x 512 matrix) was transmitted after JPEG or wavelet compression by using point-to-point and Web-based teleradiology, respectively. Lossless, 10:1 lossy, and 20:1 lossy ratios were tested. Images were evaluated for high- and low-contrast resolution, sensitivity to small signal differences, and misregistration artifacts. Three independent observers who were blinded to the compression scheme evaluated these image quality measures in 20 clinical cases with similar levels of compression.
RESULTS: High-contrast resolution was not diminished with any tested level of JPEG or wavelet compression. With JPEG compression, low-contrast resolution was not lost with 10:1 lossy compression but was lost at 3% modulation with 20:1 lossy compression. With wavelet compression, there was loss of 1% modulation with 10:1 lossy compression and loss of 5% modulation with 20:1 lossy compression. Sensitivity to small signal differences (5% and 95% of the maximal signal) diminished only with 20:1 lossy wavelet compression. With 10:1 lossy compression, misregistration artifacts were mild and were equivalent with JPEG and wavelet compression. Qualitative clinical findings supported these findings.
CONCLUSION: Lossy 10:1 compression is suitable for on-call electronic transmission of body CT images as long as original images are subsequently reviewed.
Index terms: Computed tomography (CT), image processing, **.121122 • Computed tomography (CT), image quality, **.1211 • Data compression • Images, storage and retrieval • Images, transmission • Phantoms • Teleradiology
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
Standard compression algorithms in current clinical use include the Joint Photographic Experts Group (JPEG) and wavelet techniques, which both belong to the general class of transform-based lossy compression techniques (1,2). With JPEG compression, the image is broken up into 8 x 8-pixel blocks, and each block is subjected to a discrete cosine transformation. The resultant spectral coefficients are quantized and then reordered for more efficient encoding (3). Wavelet compression, on the other hand, uses high- and low-pass filters applied repeatedly to the image to generate sub-band images on which low- and high-frequency information are separated along the x and y axes (4). Each compression algorithm is associated with its own pattern of artifact (1,5).
Study findings on the effects of compression on CT images of the liver (6) and thorax (7) have indicated that the diagnostic accuracy of CT is preserved with up to 10:1 lossy compression. However, specific guidelines for image compression have not emerged, as evidenced by the lack of image compression guidelines in the Standards for Teleradiology issued recently by the American College of Radiology (8). The purpose of this study was to quantitatively assess the effect of image compression on CT image quality and, thereby, to determine the levels of JPEG and wavelet compression acceptable for electronic transmission of body CT images by using teleradiology.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
Phantom Study
An initial electronic phantom study was performed by using the Society Of Motion Picture and Television Engineers (SMPTE) electronic test pattern (Fig 1). For the purpose of the study, the 512 x 512-matrix version was used with an 8-bit depth to simulate a prewindowed CT image prior to teleradiologic transmission. The phantom image was transmitted with 2:1 (lossless) and approximately 10:1 or 20:1 (both lossy) JPEG or wavelet compression (Figs 2, 3). A commercial implementation of JPEG compression designed by Cemax-Icon (Fremont, Calif) was used on their sending-end teleradiologic server. JPEG compression was tested during transmission of direct-digital CT images by using dual-channel bonded 64-kbyte integrated services digital network, or ISDN, lines that connected a satellite facility to our emergency radiology department located several miles away. Wavelet compression performed by Medweb (San Francisco, Calif) was applied on their Digital Imaging and Communications in Medicine (DICOM) image server. In this instance, automatic decompression was performed on the client computer by using a Web browser plug-in. Wavelet compression was tested during transmission from the Web-based image server by using a 10 base-T Ethernet connection.
|
|
|
|
|
The SMPTE phantom contained high-contrast-resolution test patterns at the peripheral aspects of the phantom (Fig 1) (9,10). The high-contrast-resolution test patterns are divided into three zones with different spatial frequencies. The signal modulation of the high-contrast-resolution pattern was 100% (alternating between the highest and lowest gray-scale values). The low-frequency pattern was composed of alternating bars measuring 3 pixels wide, the bars in the midfrequency pattern were 2 pixels wide, and the bars in the high-frequency pattern were 1 pixel wide; on a CT scan reconstructed with a 24-cm field of view, these patterns corresponded to 0.3, 0.5, and 1.0 line pair per millimeter, respectively.
The low-contrast-resolution patterns mimicked the format of the high-contrast-resolution patterns, except that the spatial frequency in all three zones was equal to that of the midfrequency zone in the high-contrast-resolution test patterns (the bars are 2 pixels wide). In the zone with the lowest contrast resolution, the modulation was 1% (ie, the lighter band equaled 51%, and the darker band equaled 50% of the maximal signal). In the middle zone, the modulation was 3% (51% and 48% of the maximal signal). In the zone with the highest contrast resolution, the modulation was 5% (53% and 48% of the maximal signal). On CT images displayed with a standard soft-tissue window width setting (400 HU), these zones with low-contrast resolution represented attenuation differences of 4, 12, and 20 HU.
The sensitivity to small signal differences was assessed by viewing gray squares superimposed on black and white larger squares. These smaller squares represented 5% of the maximum signal superimposed on 0% (ie, gray on black) or 95% of the maximal signal superimposed on 100% (ie, gray on white). On a CT image displayed with a standard soft-tissue window width setting (400 HU), these differences corresponded to an attenuation difference of 20 HU.
Misregistration artifacts were observed by noting the distortion of alphanumeric values present on the image and the displacement of the high-contrast-resolution test pattern bars relative to their original position. These misregistration artifacts were qualitatively rated as none, mild, moderate, or severe by the two reviewers in consensus.
Clinical Study
A clinical study was subsequently performed to confirm the findings of the phantom study for JPEG compression. The clinical material consisted of 20 CT studies performed at our local Veterans Administration hospital (West Haven, Conn). Twenty consecutive patients with electronically retrievable CT images served as the study group. These patients, who presented with acute onset of flank pain, underwent CT of the abdomen or pelvis for renal or ureteral calculi. With such studies, no oral or intravenous contrast material is used, which results in a uniform attenuation range afforded by the nonenhanced soft tissues in the abdomen and pelvis.
Image sets were transmitted from the Veterans Administration hospital to the emergency radiology department in our university hospital by using point-to point teleradiology (Cemax-Icon). Each image set, which consisted of 12-bit DICOM data, was transferred three times by using three compression ratios (2:1, 10:1, and 20:1 JPEG compression). The entire set of CT images was evaluated on a viewing workstation composed of four 1,024 x 1,024-pixel monitors. Three reviewers, two body imaging specialists (A.K., C.R.T.), and one general radiologist (V.P.N.) viewed the CT images by using a 1 x 2 display mode that displayed each image with full resolution without magnification. The reviewers were blinded to the compression scheme used with each set of images. The CT image sets were randomly ordered such that the compression ratio used with any set was unknown and unpredictable.
The reviewers were asked to assess the following parameters on a four-point scale (1 = excellent, 2 = good, 3 = fair, and 4 = poor). First, the reviewers were asked to rate the perceived high-contrast resolution by grading the edge sharpness of high-attenuating structures, such as calculi and phleboliths, in the abdomen and pelvis. Second, the reviewers were asked to assess the perceived low-contrast resolution by noting the sharpness and separability of subtle low-contrast-resolution structures, such as small vessels and inflammatory change that may be detected in the setting of urinary calculi, in the retroperitoneal fat. Finally, the reviewers were asked to assess the image quality, with regard to compression artifacts.
Qualitative data were analyzed by using the Wilcoxon signed rank test for each parameter and individual reader. Bonferroni correction for multiple (three) comparisons was used; statistical significance was accepted with a P value of .017 (.05/3). In addition, the differences in qualitative scores between 2:1 and 10:1, 10:1 and 20:1, and 2:1 and 20:1 were computed for each case and reader. The frequency histograms of these differences were plotted, and the median and mode for the differences were determined for each comparison.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
Sensitivity to small signal differences (at 5% and 95% of the maximal signal) was diminished with 20:1 wavelet compression but remained unaffected at all tested levels of JPEG compression. Misregistration artifacts were graded as moderate with 10:1 JPEG compression (Fig 2a, middle) and severe at 20:1 compression (Fig 2a, right). With wavelet compression, these artifacts were graded as mild or moderate with 10:1 (Fig 2b, middle) or 20:1 (Fig 2b, right) compression, respectively.
Clinical Study
Among the 20 patients in our clinical sample, renal calculi were present in 12, ureteric calculi in 10, hydronephrosis in six, perinephric stranding in 18 and periureteral stranding in eight.
Results of our clinical study were as follows (Figs 4–6). With reference to high-contrast resolution, the histogram of pooled differences in the qualitative assessment scores showed no difference between 2:1 and 10:1 compression in the majority of cases (median and mode, 0) (Fig 7a). When we compared 10:1 compression with 20:1 compression (Fig 7b) and 2:1 compression with 20:1 compression (Fig 7c), there was a leftward shift in the histogram of pooled differences such that, in most cases, the quality of the more compressed images was one level worse than that of the less compressed images (median and mode, -1). Statistical analysis revealed no significant difference in high-contrast resolution between 2:1 and 10:1 compression (Table). However, high-contrast resolution was significantly worse with 20:1 compression compared with both 2:1 and 10:1 compression.
|
|
|
|
|
|
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
The consequences of image compression have been previously evaluated with digitized radiographs (11–15), computed radiographs (16–18), mammograms (19), echocardiograms (20), CT images (6,7,21), and magnetic resonance images (21,22). Among these studies, only three focused on CT images.
In 1989, Chan et al (21) reported the use of a full-frame cosine-transform compression method with which good results were achieved at a compression ratio of 5:1. Subsequent study findings of Cosman et al (7) determined that the diagnostic accuracy of thoracic CT is maintained with 9:1 lossy compression by using tree-structured vector quantization. Most recently, Goldberg et al (6) concluded that detection of focal liver lesions remained satisfactory following 10:1 lossy wavelet compression compared with 15:1 and 20:1 wavelet compression.
We objectively assessed the effects of image compression on various measures of image quality at body CT by using a standard electronic imaging phantom. Recently, findings of a descriptive study (23) in a digital SMPTE test pattern was used in the evaluation of teleradiologic systems have been reported. The SMPTE phantom we used allows quantitative evaluation of high- and low-contrast resolution and sensitivity to small signal differences and misregistration artifacts (9,10). A limitation of this phantom is that it consists exclusively of linear patterns and does not contain a representation of curved surfaces such as those present in the body.
The phantom study findings revealed no loss of high-contrast resolution with any test level of JPEG or wavelet compression with test patterns composed of bars measuring as small as 1 pixel in width. This highest level of spatial frequency corresponds to resolutions of 0.8 line pair per millimeter on CT images reconstructed with a 32-cm field of view and 1.0 line pair per millimeter on a CT image reconstructed with 24-cm field of view.
Low-contrast resolution was unaffected with 10:1 JPEG compression, while loss of low-contrast resolution at 3% modulation was noted with 20:1 JPEG compression. This finding corresponds to an inability to resolve closely spaced structures that differ by 12 HU on a CT image displayed with a standard soft-tissue window width setting (400 HU).
With wavelet compression, low-contrast resolution was somewhat worse; loss of 1% modulation was noted with 10:1 compression, and loss of 5% modulation was noted with 20:1 compression. The loss with 10:1 compression corresponds to an inability to resolve small structures that differ by only 4 HU on a CT image displayed with a standard soft-tissue window width setting and is not thought to be important for most clinical body CT applications. However, the loss at 20:1 compression represents the inability to resolve small structures which differ by 20 HU. Differences in attenuation as small as 10 HU may be clinically important. For example, renal lesions with attenuation that increases by at least 10 HU following the intravenous administration of contrast material are thought to be enhancing lesions that may represent renal neoplasms (24). Thus, differences at 3% and 5% modulation may be clinically important, whereas differences at 1% modulation likely are not. Further studies are necessary to confirm these observations.
The sensitivity to small signal differences at the extremes of the contrast-resolution scale were not diminished with either 10:1 JPEG or wavelet compression. However, these signal differences were not appreciable with 20:1 wavelet compression. Conversely, misregistration artifacts were somewhat more severe with JPEG compression compared with wavelet compression. Such artifacts were graded as moderate with 10:1 JPEG compression compared with a grade of mild with 10:1 wavelet compression and as severe with 20:1 JPEG compared with a grade of moderate with 10:1 wavelet compression.
Although wavelet compression is commonly thought to be superior to JPEG compression, the results of our electronic phantom study do not fully support this conclusion (11,25). Low-contrast resolution was better preserved with JPEG compression compared with wavelet compression, although misregistration artifacts were more severe with JPEG compression. The results may have been influenced by the rectilinear nature of the SMPTE test pattern. Wavelet compression, in general, has been reported to perform better with curved interfaces than does JPEG compression (11); curved interfaces were not tested in this study. Conversely, the discrete nature of the JPEG algorithm has been reported to result in a more pixelated or edge-enhanced appearance; this appearance may account for the greater misregistration artifacts with the JPEG algorithm than with the wavelet algorithm, which tended to result in artifact with a smoother appearance (1). Finally, the results of this study may be related to the specific wavelet algorithm that we used. Wavelet algorithms continue to be developed that may improve on the results of this study.
Clinical acceptance testing of lossy compression algorithms should include both quantitative measurements of image degradation in vitro and qualitative confirmation of these results in vivo (26). The results of our clinical evaluation of JPEG compression generally supported the results of our electronic phantom study with regard to both high- and low-contrast resolution. Assessment of image artifacts revealed significant deterioration with progressive levels of JPEG compression in both the phantom and clinical studies. Such artifacts pose a threat to the reliable implementation of lossy compression with images that serve as the permanent record for archival or primary interpretation. Although they may not degrade the ability to depict gross disease in the abdomen or pelvis, compression artifacts may degrade the ability to resolve small structures such as tiny ureteral calculi. In addition, loss of low-contrast resolution may obscure ancillary findings such as subtle infiltration of periureteral soft tissue. Thus, these results lead us to conclude that 10:1 lossy compressed images may be suitable for on-call teleradiology. However, uncompressed or lossless images should be subsequently viewed, particularly with high-resolution applications. In addition, postprocessing of lossy-compressed images may enhance the appearance of artifacts (26).
With 10:1 compression (relative to lossless 2:1 compression), we observed no significant difference in high-contrast resolution, a marginal loss in low-contrast resolution, and a mild increase in image artifacts. With 20:1 compression (relative to lossless compression), there was significant deterioration in high- and low-contrast resolution and image quality. On the basis of the qualitative phantom and clinical image analysis, we conclude that 10:1 compression is suitable for use in direct digital transmission of body CT images for on-call teleradiology. Uncompressed or lossless images will remain the diagnostic standard until confirmatory study findings on the impact of compression on the accuracy of radiologic diagnosis are available.
|
FOOTNOTES |
|---|
Abbreviations: DICOM = Digital Imaging and Communications in Medicine, JPEG = Joint Photographic Experts Group, SMPTE = Society of Motion Picture and Television Engineers
Author contributions: Guarantors of integrity of entire study, all authors; study concepts and design, A.K., C.R.T., V.P.N., J.A.B.; definition of intellectual content, all authors; literature research, A.K., V.P.N., A.R.D., J.A.B.; clinical studies, all authors; experimental studies, all authors; data acquisition, all authors; data analysis, A.K., A.R.D., J.A.B.; statistical analysis, A.K., A.R.D., J.A.B.; manuscript preparation, A.K., A.R.D., J.A.B.; manuscript editing and review, all authors.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS DISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  V. Bajpai, K. H. Lee, B. Kim, K. J. Kim, T. J. Kim, Y. H. Kim, and H. S. Kang Differences in Compression Artifacts on Thin- and Thick-Section Lung CT Images Am. J. Roentgenol., August 1, 2008; 191(2): W38 - W43. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Ringl, R. Schernthaner, E. Sala, K. El-Rabadi, M. Weber, W. Schima, C. J. Herold, and A. K. Dixon Lossy 3D JPEG2000 Compression of Abdominal CT Images in Patients with Acute Abdominal Complaints: Effect of Compression Ratio on Diagnostic Confidence and Accuracy Radiology, August 1, 2008; 248(2): 476 - 484. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B. Kim, K. H. Lee, K. J. Kim, R. Mantiuk, H.-r. Kim, and Y. H. Kim Artifacts in Slab Average-Intensity-Projection Images Reformatted from JPEG 2000 Compressed Thin-Section Abdominal CT Data Sets Am. J. Roentgenol., June 1, 2008; 190(6): W342 - W350. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. Ringl, R. E. Schernthaner, C. Kulinna-Cosentini, M. Weber, C. Schaefer-Prokop, C. J. Herold, and W. Schima Lossy Three-dimensional JPEG2000 Compression of Abdominal CT Images: Assessment of the Visually Lossless Threshold and Effect of Compression Ratio on Image Quality Radiology, November 1, 2007; 245(2): 467 - 474. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Suryanarayanan, A. Karellas, S. Vedantham, S. M. Waldrop, and C. J. D'Orsi Detection of Simulated Lesions on Data-compressed Digital Mammograms Radiology, July 1, 2005; 236(1): 31 - 36. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Kalyanpur, V. P. Neklesa, D. T. Pham, H. P. Forman, S. T. Stein, and J. A. Brink Implementation of an International Teleradiology Staffing Model Radiology, August 1, 2004; 232(2): 415 - 419. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. P. Ko, H. Rusinek, D. P. Naidich, G. McGuinness, A. N. Rubinowitz, B. S. Leitman, and J. M. Martino Wavelet Compression of Low-Dose Chest CT Data: Effect on Lung Nodule Detection Radiology, July 1, 2003; 228(1): 70 - 75. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
