| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Thoracic Imaging |
1 From the Department of Radiology, University Hospital Vienna, AKH Wien, Währinger Gürtel 18-20, A-1090, Vienna, Austria (H.R., R.E.S., A.A.B., M.W., C.J.H., C.S.); and Department of Radiology, Utrecht Medical Center, Utrecht, the Netherlands (M.P.). Received March 29, 2005; revision requested May 25; revision received September 2; accepted September 12; final version accepted November 23. Address correspondence to H.R. (e-mail: helmut.ringl{at}meduniwien.ac.at).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Materials and Methods: In this institutional review board–approved investigation (protocol 238/2004), thin-section CT images were subjected to irreversible JPEG2000 compression by using five compression ratios (3:1, 5:1, 7:1, 9:1, and 11:1). Three radiologists independently evaluated 60 thin-section CT images, of various diseases, that were obtained with single-detector (weighted dose index, 14.4 mGy) and multidetector (weighted dose index, 9.8 mGy) CT. Toggling between the original and compressed images, readers had to identify the original image by using a forced-choice two-alternative model and to subjectively rank the quality of what they believed to be the compressed image. To assess the reader's ability to distinguish the compressed from the original image, a binomial test was used. Bonferroni correction was applied for all multiple tests.
Results: Images compressed with a ratio of 3:1 were not distinguishable from original images (P > .2 for all readers). With use of the 5:1 ratio, minor differences in appearance between the compressed and original images were seen by one of the three readers. With use of higher compression ratios (
7:1), all readers (P < .001) recognized the original image. The quality of more than 90% of the images compressed with a 7:1 or higher ratio was substantially degraded. Single-detector and multidetector CT results were not significantly different.
Conclusion: The highest ratio that yielded visually lossless compression of thin-section CT images was 3:1. With the 5:1 ratio, there was minor image quality loss, while use of higher compression ratios (7:1) caused substantial degradation of image quality and potential loss of diagnostic information.
© RSNA, 2006
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
In two studies (7,8), investigators studied the influence of JPEG2000 compression ratios on the detectability of lung nodules on CT images of the lung. Both studies revealed that compression ratios of up to 10:1 still yield diagnostically sufficient image quality. It is conceivable that the high contrast between lung nodules and the surrounding lung parenchyma enabled this relatively high compression ratio. This theory implies that more subtle parenchymal abnormalities that have lower contrast than nodules to the surrounding lung parenchyma require accordingly lower compression rates to remain detectable. These subtle parenchymal abnormalities are usually examined by using thin-section CT; however, to our knowledge, no information about the JPEG2000 compression ratios required for the accurate visualization of these anomalies exists. Thus, the aim of our study was to assess retrospectively the effect of JPEG2000 compression ratios on the quality of thin-section CT images.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Image Selection
For this study, a set of 60 images was compiled. These images had been obtained at thin-section CT examinations performed between January and December 2004 according to the following criteria: Thirty of the 60 images were obtained in 13 men and 17 women (mean age, 52.7 years ± 16.2 [standard deviation]; range, 19–83 years) with a mean weight of 68.9 kg ± 18.9 (range, 42.8–102.0 kg) at examinations performed with a single-detector CT unit (Tomoscan AV E1 7000; Philips, Best, the Netherlands). The other 30 images were obtained in 15 men and 15 women (mean age, 56.7 years ± 15.7; range, 25–90 years) with a mean weight of 69.9 kg ± 14.6 (range, 46–95 kg) at examinations performed with a multidetector CT unit (Somatom Plus 4 VolumeZoom; Siemens Medical Solutions, Forchheim, Germany).
In each of the two subsets of 30 images, the following predominant pathologic findings were equally distributed on five images each: (a) ground-glass opacity, (b) septal thickening, (c) bronchial wall thickening, (d) focal decrease in lung attenuation, (e) micronodules 1–10 mm in size, and (f) small calcifications. The definitions and descriptions of these findings were in accordance with the thin-section CT terminology defined by the Fleischner Society (9). For each of these six morphologic findings, each group of five images included two images showing extensive disease, one showing moderate disease, and two showing subtle disease. Classification of disease severity was based on the subjective evaluations of the three individuals (H.R., R.E.S., C.S.) involved in the case selection. The extent of disease was comparable between the subsets of images acquired with single-detector and multidetector CT. Moreover, the patients in whom the two subsets of images were acquired did not substantially differ in regard to sex, age, or weight.
Image Acquisition
The thin-section single-detector CT protocol consisted of 1.00-mm-thick sections acquired at full inspiration with 140 kV, 125 mAs, and a rotation time of 1 second in the sequential mode. The multidetector CT protocol consisted of 1.25-mm-thick sections acquired at full inspiration with 120 kV, 100 mAs, and a rotation time of 0.5 second in a 4 x 1-mm spiral mode and with the dose modulation program (CARE Dose; Siemens Medical Solutions) enabled. The effective tube-current–time product settings for the multidetector CT examinations varied between 81 and 94 mAs (mean, 84 mAs). The weighted CT dose index was 9.8 mGy per section for multidetector CT and 14.4 mGy per section for single-detector CT. For both thin-section CT protocols, a high-spatial-resolution kernel was used for reconstruction. The field of view was individually set for each patient such that the maximal transverse diameter of the lung matched the image size.
Image Compression
The compression process involved the following three steps: First, all images were exported as noncompressed DICOM 3.0 images to an external workstation. Second, images were irreversibly compressed and saved on the external hard disk. Third, these images were reloaded back into our picture archiving and communication system as DICOM files. The DICOM header of each file was modified such that all images could be displayed in the order of our viewing protocol.
For image compression, we used a DICOM-certified JPEG2000 algorithm (Aware JPEG SDK 3.42; Aware, Bedford, Mass) with compression ratios of 3:1, 5:1, 7:1, 9:1, and 11:1, as compared with a ratio of 1:1 for the original image file (12 bits, 384 kB). This algorithm compresses all 12 bits of the CT image (–1024 to 3072 HU) and allows altering of the image window settings after compression. The software provided by Aware included a full software developer kit for different programming languages and a command line tool that facilitates the automated generation of test series by using a few small scripts. The JPEG2000 encoder was set to standard settings for lossy compression (I97, with I meaning irreversible and 97 characterizing the wavelet filter). Although for a given original thin-section CT image, 12 bits per pixel is used as the depth resolution, images are frequently stored with 16 bits (4 bits are unused) for historical and technical reasons. All compression ratios used in this study were calculated in reference to an original image with a 12-bit gray-scale depth (384 kB for the entire image).
The achievable lossless JPEG2000 compression ratio can be used as an indicator of the composition of an image in terms of not only the amount of structures and image noise but also the areas on the image that contain no image information (eg, black frames or corners used for collimation), which are particularly amenable to image compression. The achievable compression ratio for lossless JPEG2000 compression of the thin-section CT images in this series amounted, on average, to 1.4:1.0 for the multidetector data and to 1.8:1.0 for the single-detector data. The higher lossless compression rate for the single-detector CT images was due to the fact that these images were acquired with a higher radiation dose, which resulted in less image noise, and they contained small black corners caused by the limited field of view of the scanner. These corners accounted for 14% of the compressible image data and consequently resulted in a slight overestimation of 14% of the compression ratio in favor of the single-detector CT images.
Image Reading Methods
The images were independently read by three board-certified radiologists with 10 (C.S., reader 1), 4 (H.R., reader 2), and 11 (A.A.B., reader 3) years experience in reading thin-section CT images. They evaluated the images by using standard Food and Drug Administration–approved medical picture archiving and communication system equipment (Agfa, release 4.1; Agfa Healthcare, Mortsel, Belgium) with a 21-inch cathode-ray tube monitor (Agfa/HB2183L; Agfa Healthcare) that conformed to the DICOM 3.0 standard. Theoretically, the maximal luminance was 600 cd/m2 and the minimal luminance was set as low as possible—at 0.3 cd/m2. Gray-scale CT images were displayed with 256 (8-bit) shadings and an on-screen spatial resolution of 1600 x 1200 pixels. The display was fixed at a commonly accepted thin-section CT window setting (window width, 1500 HU; window level, –650 HU). To mask the extrathoracic air and the chest wall soft tissues, all images were collimated to the lung parenchyma after compression by using a collimation feature provided with the picture archiving and communication system software. Compressed and original images were displayed at identical sizes by using a full-screen format. The viewing distance could be chosen arbitrarily by the reader, and the reading time was not limited. Reading conditions were kept constant for all reading sessions and included subdued ambient lighting. All annotations except a four-digit number were removed from the displayed images. The readers were instructed not to alter the window settings of the displayed soft-copy images.
Single-detector and multidetector CT images were evaluated during two separate reading sessions for each reader; thus, each reader viewed all of the images independently. All images were evaluated pairwise, with each pair consisting of the compressed and original versions of the same image. Accordingly, there were six pairs for each original image. Five pairs consisted of one compressed image (compressed with ratios of 3:1, 5:1, 7:1, 9;1, and 11:1) and one original image; the sixth pair consisted of two original images (1:1 compression ratio) and served as the control pair. Thirty CT images were obtained with the single-detector scanner, and 30 images were obtained with the multidetector scanner; a total of 360 image pairs were analyzed.
The two images in each pair were arranged such that the reader viewed them sequentially on a single monitor display by using the mouse-wheel or the keypad to toggle between the two images. Previous study (10) investigators reported that toggling between the two images was more time efficient and more sensitive for the detection of subtle differences in image quality than comparing the images side by side. The order of the image pairs, as well as the order of the two images (original then compressed image or vice versa) within one pair, was randomized differently for each reader by using the RND (randomize) function (linear-congruential method [11]) of the programming language Visual Basic 6 (Microsoft). During the reading session, no information about the compression ratio or the position of the original image (ie, whether first or second) was known to the reader or the study coordinator (H.R.).
Image Analysis
Image analysis involved two processes. First, the readers were asked to respond according to a two-alternative forced-choice model: For each image pair, the readers had to select the image with the superior image quality, which was thus considered the original. They were asked to do so even when they could not appreciate a difference in appearance between the images in a given pair. Second, the readers were asked to categorize images according to the differences between the compressed image and the original image by using a scale of A–C as follows: Category A included images that were identical or almost identical in appearance such that the original image could not be determined. Category B included images for which minimal differences in image quality between the two images, which allowed a definite identification of the original image, were observed. The differences between the images, however, were not considered to be diagnostically relevant. Category C included images for which there were considerable differences in image quality between the compressed and original images such that the degradation of image quality was likely to affect diagnoses to the extent that the image with inferior quality was unacceptable for diagnostic purposes, in the opinion of the reader.
We should note that the assignment of image pairs to category A, B, or C was based on a subjective ranking of the severity of image degradation, with regard of whether readers considered the image artifacts to have no, minimal, or considerable relevance to diagnostic performance. Our study setup did not include specific evaluation of the diagnostic accuracy of the thin-section CT images. Readers were instructed to not only focus on the abnormality of the individual case but also consider the appearance of the entire lung parenchyma.
Statistical Analyses
We applied a forced-choice two-alternative model to assess the readers' ability to differentiate the original image from the compressed image. The visually equivalent image appearance of the compressed and original images was indicated by an even distribution of the observed relative frequency of correct and false identifications of the original image (observed relative frequency approximately equal to 0.5, or 50%). The reader's ability to differentiate the compressed and original images was expressed as the number of correct responses that differed significantly from 0.5. We used a binomial test to assess the significance of the difference between the observed relative frequency and 0.5. The significance level was set at P = .05.
The advantage of using a forced-choice two-alternative model rather than offering the readers three alternatives, including one for equal image appearance, is that all answers (n = 30 in our study) are considered in the binomial test, whereas the use of a third alternative (equal image appearance) would have facilitated a decrease in the number of answers by the number of equal answers because these answers are ignored in binomial testing (12).
We used descriptive statistics to analyze the readers' visual assessments of the differences in image quality between the original and compressed images. The significance of differences between the multidetector and single-detector CT reading data with use of both the forced-choice two-alternative model and the three-category rating test was analyzed by performing 
2 tests (Fisher exact test) separately for each reader and each compression step. Because of the large number of tests (six compression steps) performed, to minimize 
(ie, type I) error, Bonferroni correction (13,14) was generally applied to all analyses in which multiple tests were performed. The significance level (P = .05) was therefore reduced to an -adjusted P level of .0083.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
|
|
|
|
|
|
All three readers assigned the differences in image appearance to category A in 100% of the single-detector CT image pairs. Thus, none of them observed differences between the compressed and original images. Readers 1, 2, and 3 assigned the appearance differences to category A in 97%, 100%, and 100% of the multidetector CT image pairs, respectively. No significant difference in subjective ratings between the single-detector and multidetector CT data (P > .99 for all three readers) was observed.
5:1 Compression Ratio
For reader 3 only, the observed relative frequency of correct responses was 0.73 with the single-detector CT protocol and 1.00 with the multidetector CT protocol. Only the observed relative frequency achieved with multidetector CT was significantly different from 0.5 (P < .001) and thus indicated an uneven distribution of responses. With single-detector CT, the P value (.016) was higher than the -adjusted P level of .0083; therefore, the difference did not reach significance. The uneven distribution of responses regarding the multidetector CT image pairs suggests that this reader (A.A.B.) correctly discriminated the original image from the compressed image in a statistically significant manner. He correctly identified the original image in 22 single-detector CT image pairs and in all 30 multidetector CT image pairs. This was the only significant difference observed between the single-detector and multidetector CT data (P = .005), whereas for readers 1 and 2, the P value was greater than .99 and .299, respectively.
Reader 3 assigned the differences in image appearance in 50% of the single-detector CT image pairs to category A and the differences in 47% of these pairs to category B. Thus, in nearly half of the image pairs, minimal differences between the original and compressed images could be seen. These minimal differences resulted in the discrimination between the two images in a substantial although not significant number of image pairs. These differences were not considered diagnostically relevant. Reader 3 assigned the differences between the original and compressed images to category A in 23% of the multidetector CT image pairs and to category B in 70% of these pairs. In only a few image pairs, the reader observed substantial differences between the images and could not exclude loss of diagnostic information. He therefore assigned the differences to category C in one (3%) single-detector CT image pair and in two (7%) multidetector CT image pairs. No significant difference in the subjective ratings of reader 3 was observed between the single-detector and multidetector CT data (P = .109).
For readers 1 and 2, the observed relative frequencies of correct responses were not significantly different from 0.5 and thus indicated an even distribution of responses. For differentiation of the original and compressed single-detector CT images, P values were .362 and .856 for readers 1 and 2, respectively. For differentiation of the multidetector CT images, P values were .585 and .2 for readers 1 and 2, respectively. The even distributions indicate that these two readers were unable to distinguish between the compressed and original images. They assigned the differences between the two images to category A in 93% and 90% of the single-detector CT image pairs and in 77% and 83% of the multidetector CT image pairs. No significant difference in the subjective ratings of readers 1 (P = .145) and 2 (P = .509) were observed between the single-detector and multidetector CT data.
7:1 or Higher Compression Ratio
For all three readers, the observed relative frequencies were significantly different from 0.5, indicating an uneven distribution of responses. For differentiation in both the single-detector and the multidetector CT image pairs, P values approached a value of less than .001. This uneven distribution of responses suggests that all three readers correctly discriminated the compressed image from the original image in almost 100% of the image pairs. No significant difference between the single-detector and multidetector CT data (P > .99, P = .052, and P > .99 for the three readers) was observed.
With 7:1 compression, readers 1, 2, and 3 assigned the differences between the original and compressed images to category C in 53%, 27%, and 70% of the single-detector CT image pairs, respectively, and in 70%, 77%, and 83% of the multidetector CT image pairs, respectively. Thus, the readers observed substantial differences between the two images such that loss of diagnostic information could not be excluded. All three readers assigned images compressed with ratios higher than 7:1 to category C in more than 90% of the image pairs. Results of 2 tests (Fisher exact tests) revealed the following two significant differences between the single-detector and multidetector CT data: Reader 2 judged the differences between the original images and the images compressed at ratios of 7:1 and 9:1 to be more substantial with the multidetector CT protocol than with the single-detector CT protocol. He assigned differences between the original and compressed single-detector CT images to category A or B significantly more frequently and assigned differences between the multidetector CT images to category C more frequently (P < .001 for 7:1 compression group, P < .002 for 9:1 compression group).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
The results achieved with a compression ratio of 5:1 were more ambiguous. Two of the three readers found that the images compressed at a ratio of 5:1 were indistinguishable from the original images. Reader 3 significantly differentiated the original from the compressed multidetector CT images. However, this reader was able to discriminate the original from the compressed images in a substantial yet not significant number of single-detector CT image pairs. With both protocols, he judged the degradation of image quality to be minor without diagnostic relevance (category B). On the basis of these findings, we conclude that a compression ratio of 5:1 induces no substantial loss of diagnostic information but might cause a visually appreciable loss of image quality.
The readers were asked to determine only which of the two images they considered to be the original image. They based this decision on any type of image quality degradation but were not asked to specifically qualify the finding on which they based their decision. The most apparent artifact that enabled discrimination between the original and compressed images was blurring of relatively homogeneous low-attenuating areas within the lung parenchyma, whereas distortion of sharply defined contours—those of septal lines or nodules, for example—occurred with only higher compression ratios. This blurring occurs because the JPEG2000 algorithm efficiently summarizes areas with only a few structural details (eg, air, nonopacified lung parenchyma, muscle, fat) to result in blurring in such areas after application of low compression ratios (3,5). However, well-defined structures with high contrast contours, such as interlobular septa or nodules, receive a higher priority during encoding and therefore appear preserved—even when higher compression ratios are used.
Because distortion of any image area implies loss of confidence and potential loss of diagnostic accuracy, we decided to consider not only the specific thin-section CT pattern but also the appearance of the entire lung parenchyma. We did not separately evaluate the effect of image compression on specific thin-section CT patterns because it seems unrealistic to apply different compression ratios for different thin-section CT patterns. In addition, thin-section CT images usually show a variety of patterns, with different contrast and spatial resolution requirements for any given image.
It should be noted again that the goal of our study was not to determine the threshold for the compression ratio that will or is likely to lead to loss of diagnostic accuracy. Such a determination involves many more factors, such as individual reader experience and/or the extent and type of disease, which are much more challenging to control. We therefore decided to determine the ratio of compression with which loss of diagnostic accuracy is unlikely, which applies if compressed and original images cannot be reliably discriminated visually in a setup that is very sensitive for the detection of even small differences. This, however, represents a rather conservative approach that most likely leads to an underestimation of the compression ratio with which unaltered diagnostic accuracy can be maintained.
We failed to establish a significant difference between the single-detector and multidetector CT images with respect to compression ratios, which resulted in no or minimal alterations in image quality. This finding does not mean that no difference actually exists. Because the multidetector CT scans had been obtained by using a radiation dose that was approximately 35% lower than the dose used to obtain the single-detector scans, they had higher noise. Differences between the compressed and original multidetector CT images were more frequently assigned to categories B and C than to category A and thus indicated minor or considerable degradation of image quality of the compressed images. In addition, one reader was able to discriminate the original image from the image compressed with a 5:1 ratio in a higher number of multidetector CT image pairs than single-detector CT image pairs. These findings appear to reflect the higher vulnerability to compression artifacts of multidetector CT images obtained with a relatively lower radiation dose.
Investigators in two studies (7,8) evaluated the performance of the JPEG2000 algorithm for compression of chest CT images. Results of both studies suggested that compression ratios of 10:1 were acceptable. Both studies, however, were focused on the detection of nodules obtained with low-dose screening protocols. Ko et al (7) used images obtained with a 1-mm section thickness, 20–40 mAs, and 120 kV. These authors found no reduction in performance with 10:1 compression at receiver operating characteristic analysis, but they reported markedly decreased sensitivity and emphasized the need for further studies. In the second study (8), a mobile CT scanner was used, and images were obtained with a 5-mm section thickness, a pitch of 2, and 50 mAs. The authors found no significantly different rate of detection of cancers 6–15 mm in diameter between noncompressed images and images compressed at 10:1 and 20:1 ratios. However, they observed the images compressed at a 20:1 ratio with the wavelet algorithm to have inferior quality.
In both studies, the authors did not specify the original image size, in kilobytes, that they were referring to for the calculation and description of the compression ratio. Most CT images with a matrix of 512 x 512 pixels are stored in a 512-kB file so that 16 bits are reserved for each pixel. Most scanners only use 12 of the 16 bits for the Hounsfield unit information per pixel (gray-scale depth, meaning 384 kB). Knowledge of the underlying bit depth, however, is crucial for interpretation the results.
Compared with our results, the higher compression ratios used in both previous studies (7,8) were found to be acceptable for the specific imaging task for which they were tested. This difference is most likely due to the following factors: First, the study goals were different. Although the detection of a particular structure (eg, nodules) was tested in the previous two studies, we sought to determine the compression ratio that would still yield indistinguishable images. The latter determination has to be considered the more conservative approach and thus, not surprisingly, yielded lower compression ratios.
Second, the ratio of acceptable compression is highly dependent on the noise on the original image because the binary information for a homogeneous area is relatively small and can be effectively compressed, as compared with an area that mainly contains image noise, which is less amenable to compression although it does not contain structural information. Thin-section (1-mm) CT images, however, have intrinsically higher image noise than do the images evaluated in one of the previous studies (8).
Third, the ratio of acceptable compression is highly dependent on the structural composition of the image. Screening examinations performed solely to detect nodules are likely to be amenable to relatively high compression ratios (7,8), although the detection of ground-glass nodules would be expected to be negatively affected. The detection of ground-glass nodules, however, was not specifically addressed in these studies.
Last, compared with noisier and edge-enhanced thin-section CT images, 5-mm-thick CT sections of the lung—like those previously evaluated (8)—are more affected by partial volume effects and are characterized by a higher intrinsic blurriness that is already on the original images.
To our knowledge, there is only one previous study in which the influence of image compression on the depiction of lung parenchyma on thin-section CT scans was evaluated (15). The authors compared compressed images of 2- and 10-mm section thickness with the original images and found that the thicker scans could be compressed with ratios of up to 6:1 or 7:1, while the thinner scans could be compressed with ratios of only 4:1 or 5:1. Although this previous study and our study have concordant reported results, the two investigations are not directly comparable: In the previous study, an older JPEG algorithm was used, the study group was small (eight patients), and only the regions of interest (64 x 64 pixels, 8-bit depth) were compressed. These regions of interest were evaluated by means of subjective assessment of quality and objective measurements of the peak signal-to-noise ratio. The effect of compression, however, is largely influenced by whether only parts of the image or the entire image is compressed and, likewise, whether the image evaluation includes only parts of the image or the entire image.
A limitation of our study was that the application of our findings might have been restricted because the effect of compression on image quality depends on the level of image noise on the image and the structural content of the image. From this perspective, our results are primarily valid for the scanning protocols that we used. We tried to counterbalance these effects by using the CT protocols for two scanners (single-detector and multidetector CT) and a variety of thin-section CT patterns. We therefore believe that our results could be extrapolated to a variety of thin-section CT protocols that yield comparable image quality. Also, it could be argued that, compared with other CT protocols, discontinuous thin-section CT does not produce an enormous data load and thus does not represent the type of CT examination that requires image compression. However, we think that the option available with multisection CT scanners to obtain every chest scan with thin-section quality will result in an increasing number of chest CT examinations that serve both purposes: continuous evaluation of the mediastinum and lungs and high-spatial-resolution evaluation of the lung parenchyma.
The binomial test that we applied to assess statistical significance has relatively low power in the detection of small yet statistically significant differences. Nevertheless, it is the correct test for analysis of the data obtained in a forced-choice experiment. A forced-choice experiment, however, is designed also to reveal small differences in image quality that may compensate for the low power of the binomial test.
In summary, our study results show that the compression of thin-section CT images of the lung with use of JPEG2000 compression at a ratio of 3:1 does not cause visually noticeable differences in image quality. At a compression ratio of 5:1, minimal alterations in image quality can be seen, but loss of diagnostically relevant information is unlikely. Ratios of 7:1 or higher cause substantial losses in image quality with an increased risk of loss of diagnostically relevant information. Therefore, we do not recommend using ratios of 7:1 or higher for compression of thin-section CT images.
|
ADVANCES IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
FOOTNOTES |
|---|
Abbreviations: DICOM = Digital Imaging and Communications in Medicine • JPEG2000 = Joint Photographic Experts Group 2000
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, H.R., C.S.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, H.R., R.E.S.; experimental studies, H.R., R.E.S., A.A.B., M.P., C.J.H., C.S.; statistical analysis, M.W., C.S.; and manuscript editing, all authors
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  K. J. Kim, B. Kim, K. H. Lee, T. J. Kim, R. Mantiuk, H.-S. Kang, and Y. H. Kim Regional Difference in Compression Artifacts in Low-Dose Chest CT Images: Effects of Mathematical and Perceptual Factors Am. J. Roentgenol., August 1, 2008; 191(2): W30 - W37. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
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]
|
|
|
|
|
|
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]
|
|
|
|
|
|
B. Kim, K. H. Lee, K. J. Kim, R. Mantiuk, S. Hahn, T. J. Kim, and Y. H. Kim Prediction of Perceptible Artifacts in JPEG 2000 Compressed Chest CT Images Using Mathematical and Perceptual Quality Metrics Am. J. Roentgenol., February 1, 2008; 190(2): 328 - 334. [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]
|
|
|
|
|
|
A. Fidler, B. Likar, and H. Rupert Ringl What Is Wrong with Compression Ratio in Lossy Image Compression? Radiology, October 1, 2007; 245(1): 299 - 300. [Full Text] [PDF]
|
|
|
|
|
|
H. S. Woo, K. J. Kim, T. J. Kim, S. Hahn, B. Kim, Y. H. Kim, and K. H. Lee JPEG 2000 Compression of Abdominal CT: Difference in Tolerance Between Thin- and Thick-Section Images Am. J. Roentgenol., September 1, 2007; 189(3): 535 - 541. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
