| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Letters to the Editor |
Fidler, MD * and
Botjan Likar, MD 
* Faculty of Medicine, Department of Oral Medicine, University of Ljubljana, Hrvatski trg 6, Ljubljana 1000, Slovenia
Faculty of Electrical Engineering, Laboratory of Imaging Technologies, University of Ljubljana, Hrvatski trg 6, Ljubljana 1000, Slovenia
e-mail: ales.fidler{at}mf.uni-lj.si
We read with interest the article by Dr Ringl and colleagues (1) in the September 2006 issue of Radiology, in which they evaluate the effect of using the lossy image-compression algorithm, Joint Photographic Experts Group (JPEG) 2000, on computed tomographic (CT) images. We found the most interesting part of the article to be in the Discussion section, in which the authors stated that application of the highest acceptable compression ratio (CR) might be restricted because the quality of lossy compressed images depends on the level of image noise and the structural content of the image. We would like to further highlight this important issue that has been often neglected in many studies evaluating lossy image compression in radiology.
For diagnostic accuracy, it is of utmost importance that the amount of image detail preservation in lossy image compression is constant, predictable, and controlled. Image compression can be performed by means of two compression modes—namely, a constant quality factor (QF) and a constant CR. The effects of these modes on image detail degradation were studied recently (2). It was demonstrated that image detail degradation greatly depends on image content in a constant CR compression mode. In other words, more image details would be lost on the more complex images and vice versa. On the other hand, constant degree of image detail degradation, which is crucial for diagnostic accuracy, was achieved only with a QF compression mode. However, the scales for adjusting QF with the most often used JPEG and JPEG 2000 lossy compression methods are not standardized, although the methods themselves are ISO (International Organization for Standardization) standards, which makes comparative evaluation of different lossy compression methods infeasible (3). This fact could be the reason that CR has become a de facto standard for expressing the degree of image data reduction in radiology, and it is even used in the U.S. Food and Drug Administration guidance (4). Nevertheless, it has to be emphasized that CR depends not only on the degree of data reduction but also on the image content of original image (2) and noise (5), which is in accordance with the theory of data compression (6). In conclusion, CR is by definition a measure of file size reduction and as such is not suitable to expressing the degree of image detail degradation in lossy image compression.
 |
References |
|---|
 TOP References References |
|---|
Medical Diagnostic Division of Radiology, Medical University of Vienna, Währinger Gürtel 18-20, Vienna 1090, Austria
e-mail: helmut.ringl{at}meduniwien.ac.at
We thank Drs Fidler and Likar for their interest in our article (1) and for their thoughtful comments about some of the problems of lossy image compression. As the authors point out, image compression can be performed by using a constant QF or by using a constant CR. It is true that the quality of lossy-compressed images depends on the level of image noise and the basic geometry of the content of the image, which would favor the authors' approach of a constant QF. However, as these authors state in a previous article (2), there is no standardized scale for the JPEG 2000 compression mode that would maintain in a constant manner the image quality. So we, like many other authors working on this subject (3–5), used CRs, which can be standardized.
Our study, by design, accounts for the different image noise. We used different scanners to include different levels of noise, and the latter was also addressed with the use of a large patient collective with a wide range of weights. The basic geometry of the images is also kept constant by the simple fact that the images of all patients are zoomed to the degree that the torso fits the size of the computed tomographic (CT) image; therefore, the most important high-attenuation steps are roughly in the same position for all patients. The variety of different abnormalities in the lung and, therefore, the different levels of entropy, were also taken into consideration in our study, both by means of the number of patients and by means of the inclusion of all major disease patterns that could increase the entropy of the image. And, finally, the difference in geometry within a single patient is covered by the fact that the test images were obtained at different levels in the patients.
It is further likely that image quality needs are different for different imaging modalities, which would further imply that the required QF would also vary. The statement by Drs Fidler and Likar that CR is a measurement of file-size reduction is well known, but it does not conflict in any way with the findings in our article. CRs are not perfect, but, to date, they are the most rapid and feasible way to control lossy image compression. To answer the authors' statement, there is nothing wrong with CRs, as long as they are applied for a specific image modality and for a specific organ or body part.
|
References |
|---|
|
TOPReferencesReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  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]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
