Effects of Lossless and Lossy Image Compression and Decompression on Archival Image Quality in a Bone Radiograph and an Abdominal CT Scan
by

Michael Tobin, M.D., Ph.D.

ABSTRACT

The purpose of this study was to investigate the effects of lossless and lossy compression on two types of archived medical images for which interpretation depends on (1) high spatial resolution, e.g. a bone radiograph, and (2) high contrast resolution, e.g., abdominal computed tomography (CT) study.
The author found that image quality was preserved with lossless compression and relatively low levels of lossy decompression (~2.5 to 1) but that at higher levels of lossy compression, visible image degradation resulted, sooner for wavelet than for JPEG.
Because the level of compression that preserves clinically acceptable image quality may depend on the modality, the anatomy, and the pathology, the author recommends lossless wavelet or JPEG compression algorithms for medical image archiving.
KEY WORDS: image compression, lossless, lossy, wavelet, image quality, medical imaging, image archiving

INTRODUCTION

Radiology is undergoing a profound transition from interpretation of images displayed on film to reading images on high-resolution computer monitor screens. As this new era emerges, radiologists are also starting to see traditional film delivery replaced by electronic transfer of digital files, and film storage rooms replaced by archives of computer files.
The medical enterprise depends on a system that makes diagnostic images available for radiologic interpretation, that transmits images to physicians throughout the system, and that efficiently stores images pending retrieval for future medical or legal purposes. Computerized medical imaging generates large, data-rich electronic files. To speed electronic transfer and minimize computer storage space, medical images often undergo compression into smaller digital files.
The underlying assumption, that computer-based medical imaging remains sufficiently faithful to film display to maintain diagnostic accuracy, raises some important questions. Does compression of digitally archived images affect image quality? Does the process introduce artifacts or result in lost image detail? Do the effects of compression differ among various types of radiology images?
The level of diagnostic detail needed for clinical interpretation of medical images varies according to modality. In general, nuclear medicine scans require less detail than computed tomography (CT) or magnetic resonance (MR). Because interpretation of mammography and radiography depends on high spatial resolution, these images demand more detail than CT or MR, which need high contrast resolution for diagnostic interpretation. The same compression process may affect images that require high spatial resolution differently than images that depend on high contrast resolution To document the effects of compression on diagnostic detail, the author tested two types of radiologic images with various compression algorithms.

MATERIALS AND METHODS

The author selected two radiology images to study under various compression algorithms: (1) a bone radiograph of the hand of a patient with scleroderma (Fig 1); and (2) an abdominal CT scan of a patient with adult-onset polycystic kidney disease (Fig 2).
The bone digital image was created by scanning the hand radiograph at 300 dots per inch (dpi) with an Epson 636 scanner equipped with a transparency adapter and connected to an Amiga 1200 computer. The resulting greyscale image was 2136 x 3196 pixels in size and 256 bits deep.
The CT digital image was created by scanning the CT film at 320 dpi using the Epson scanner/Amiga computer system described above. The resulting greyscale image was 1163 x 1038 pixels in size and 256 bits deep.
Five board certified radiologists at the author's institution assessed the degree of image degradation resulting from various types and amounts of compression associated with several different digital image file formats. A qualitative, rather than a quantitative, approach was chosen because radiologists typically evaluate images qualitatively in their day-to-day practice and, also, because common metrics used for comparing images pre- and post compression, e.g., mean pixel error, root mean square error, maximum error, etc., may not correlate well with visual assessment of image quality. (1)

File Formats

There are more than 100 image file formats for computerized digital images, which can be divided into those which, in the process of storage, compress the original image and those which do not.
Image formats using compression can be further divided into (1) those that maintain full fidelity of the original digital data during the process of compression ("lossless" compression); and (2) those which do not ("lossy" compression). The degree of "lossiness" is usually under operator control.

Non-compressed Format

The TIFF (Tagged Image File Format) format, originally developed by the Aldus Corporation, is well suited for graphics and publishing applications and is widely accepted in business and industry. Files are large but stored data is not subject to compression/decompression noise or other artifacts. TIFF files accommodate 24-bit Images. Non-compressed TIFF images were taken as the standard against which other images were compared.

Lossless Compression Formats

TIFF

As a user option, TIFF files can be losslessly compressed at the time of image storage using lossless compression algorithms such as RLE (run length encoding) or LZW (Lempel-Ziv-Welch). Data integrity is maintained and resulting file size is smaller.

GIF

GIF (Graphics Interchange Format) was developed by Unisys and licensed by Compuserve as a cross-platform image standard for its users on the Internet. Lossless compression is achieved with the proprietary LZW (Lempel-Ziv-Welch) algorithm. GIF files are limited to 256 different colors or shades of grey.
GIF files do not store actual grey scale values in the image matrix. Instead, single numbers are used, each one of which corresponds to a specific grey scale value in the image. The one-to-one correspondence between a given (index) number and its greyscale value is kept in a look-up table (lut), or palette, which is stored with the image. It is the matrix of index numbers that is compressed with the LZW algorithm. (2)

PNG

The PNG (Portable Network Graphics) format is intended as a replacement for the GIF file format whose copyright owner, Unisys, requests royalty payment from commercial developers for use of the LZW algorithm.
PNG uses the lossless "Deflate" algorithm, which is based on LZ77, the predecessor to LZW. Image processing programs such as those that first "filter" an image so that repeating patterns can be recognized vertically as well as horizontally, for example, can make PNG's use of the "Deflate" algorithm more effective. Although the details need not concern the radiologist, they do help explain why the PNG usually achieves higher compression than GIF.
The Deflate algorithm used by PNG is also used by the well-known pkZIP compression program, which itself can be used to compress files losslessly, including image files. The PNG format can accommodate images with 512 levels of grey, a decided benefit when storing medical images of that depth.
PNG and GIF formats are particularly well suited for Internet graphics such as logos, where uniformity of color etc. leads to significant redundancy of data and high degrees of compression. PNG has additional advantages that are beyond the scope of this paper, although it has not had the widespread acceptance initially predicted.

BMP

BMP (bitmapped picture) is Microsoft Windows(tm) device-independent bitmap standard. Users of this format can depend on their images being displayed on any Windows device. BMP supports 24-bit images. Lossless compression is possible, using BMP's run length encoding (RLE) algorithm, but the resulting image file supports only 256 levels of grey. Again, this limitation becomes important for medical images with more than 256 greyscale levels.

Lossy File Formats

JPEG
Developed by the Joint Photographics Expiratory group, JPEG is a compression scheme, with JFIF as the associated file format. At present, JPEG compression is based on the DCT (Discrete Cosine Transform) approach. JPEG compression is lossy, although it can be made to operate in lossless mode.
With JPEG compression, the degree of lossiness is under operator control. Because the current implementation of JPEG operates on 8 x 8 pixel segments, images can appear blocky at high compression ratios.

Wavelet

The wavelet compression algorithm has features similar to, yet different from, the Fourier transform. (3). Although often used for lossy compression, the wavelet algorithm can be operated in a lossless mode. An important point is that wavelet compression operates on an entire image at once, thus avoiding the "blockiness" associated with JPEG methodology.
Compression using wavelets may to offer advantages over current compression techniques and is anticipated as the basis of JPEG 2000. (4)

RESULTS

Tables 1 and 2 summarize results of various compression techniques on the hand radiograph and the abdominal CT image.

Hand Radiograph (AP)

TIFF
When stored as an uncompressed 8-bit deep TIFF image, the hand radiograph required 6.82 MB (megabytes) of space on a Syquest Syjet (removable) hard drive.
However, when compressed with the lossless LZW algorithm, the same TIFF image required only 4.00 MB, a 1.7:1 compression ratio, with no loss in quality.

BMP
As an uncompressed BMP file, the hand radiograph in Figure 1 required 6.82 MB, which was the same size -- and quality -- of the uncompressed TIFF file. When the BMP image was losslessly compressed according to a Run Length Encoding (RLE) algorithm, the resulting file size is 5.69 MB, a 1.2:1 compression ratio while maintaining image fidelity.

GIF
A GIF file of the hand radiograph was 4.46 MB, 1.5 times smaller than the uncompressed TIFF image. Lossless compression was achieved with the LZW algorithm at the time of GIF file creation. As with other formats using lossless compression, digital image quality was fully maintained.

PNG
Using the "Deflate" algorithm, the PNG image format squeezes the original, uncompressed TIFF file to 3.69 MB, a 1.8:1 compression ratio, a slight improvement over GIF, with no loss of image quality.

JPEG
At Quality Factor (QF) =100 (least compressed; best image), the hand radiograph file size was 2.82 MB (2.4:1 compression ratio), with image quality visually indistinguishable from the original.
At QF=3.2, the digital file size was 0.36 MB (a 19:1 compression ratio). Image quality was still extremely good, with loss of detail evident visually only on magnified images.
At QF=1.0, the file size was only 0.17 MB (40:1 compression ratio), but the "block" artifact characteristic of JPEG compression was readily visible.
Finally, at QF=0.1, the file size was reduced to 0.08 MB (85:1 compression ratio) but the image became unreadable. (Fig 3)

Wavelet (lwf and others)
Using the evaluation software provided by LuraTech, Inc., the highest quality, lowest compression (Q=100) resulted in a 0.98 MB file (a 7:1 compression ratio) and excellent image quality.
However, changing the quality to Q=99 (0.58 MB, compression ratio of 12:1) or Q=98 (0.37 MB, compression ratio 18:1), led to loss of trabecular detail that was readily visible. (Fig 4)
Similar results were obtained with the freely downloadable software from LizardTech, Inc..

Abdominal CT (Single Axial Slice)

TIFF
As an uncompressed TIFF file, the CT image in Figure 2 required 1.21 MB. However, when compressed with the lossless LZW algorithm, the same TIFF image required only 0.90 MB, a 1.3:1 compression ratio. Image quality was maintained.

BMP
The Windows(tm) BMP (uncompressed) CT digital file was the same size (1.21 MB) and quality as the TIFF digital image. With RLE (lossless) compression, the image file was actually larger, 1.41 MB, with a compression ratio of 0.86:1, an 8% gain in size, but withot loss in image quality.

GIF
The GIF CT digital image file, automatically compressed by the LZW algorithm, is 1.02 MB. Compared to the uncompressed TIFF image, this is 1.2:1 compression ratio achieved without image degredation.

PNG
The PNG CT digital image file is 0.79 MB, a 1.5:1 compression when compared to the uncompressed TIFF image. Image quality is maintained and file size is smaller than the corresponding GIF image.

JPEG
At QF =100 (least compressed; highest quality), the CT digital image file size was 0.77 MB (1.6:1 compression ratio), with image quality indistinguishable from the original.
Using QF=3.2, the CT image digital image file was only 0.15 MB -- a compression ratio of 8:1 -- with excellent image quality, without block artifact.
Using QF=1.0, the CT digital image file size decreased to 0.06 MB (a compression ratio of 22:1). Digital image quality was very good, only slightly inferior to that produced by QF=3.2, with subtle artifact visible at 2x magnification, primarily in the more continuous tone areas and in the alphanumerics.
Using QF=0.1, the CT digital image file was further reduced to 0.02 MB (a compression ratio of 67:1), but the image was unreadable. (Fig 5)

Wavelet
Using the evaluation software provided by LuraTech, Inc., highest quality, lowest compression (Q=100) led to a file size of 0.76 MB -- a compression ratio of 1.6:1 -- with excellent image quality.
At Q=89, the file size was 0.12 MB, a 10:1 compression ratio. Although image quality was still very good, there was subtle smoothing of liver parenchyma.
At Q=85, the file size was 0.07 MB, a 28:1 compression ratio. Artifactual smoothing of liver parenchyma was more obvious.
At Q=80, file size was reduced to 0.03 MB, a 45:1 compression ratio. The image appears out of focus, although major pathology remained evident (Fig 6).
Regardless of whether the algorithm used was from LuraTech, LizardTech, or SPIHT, wavelet compression ratios of less than 10:1 gave excellent images, while smoothing artifacts were present at higher ratios.

DISCUSSION

Images compressed losslessly occupy less space than the originals, but space-saving gains are modest, with compression ratios in the 2.5:1 range. And as we observed from the RLE (Run Length Encoded) BMP file of the CT image, it was actually possible to increase the file size of complex images with compression. No image data are lost during lossless compression. Decompression restores the original image without loss of fidelity. Images stored in the GIF and PNG formats are compressed automatically, whereas for TIFF and BMP files, the user decides whether or not to compress the file.
Lossy compression achieves higher compression ratios than lossless, but at the expense of image quality, with the degree of lossiness under user control. The artifact introduced during lossy compression depends on the compression scheme and how it is implemented. (5)
JPEG has been used on the Internet for many years. (2, 6). Because the standard implementation of JPEG operates on 8 x 8 pixel segments, images can appear blocky at high compression ratios although in our test images, we did not observe this artifact, even at compression ratios of 20:1 and higher.
If the JPEG DCT algorithm were applied to an entire image, rather than to 8 x 8 pixel sub-units of it, even higher levels of acceptable compression might be achieved. Toney et al. found that DCT compression ratios of 10:1 were suitable for detection of subtle fractures in the pediatric population. (7) Sayre et al., however, showed that there was no loss in diagnostic accuracy in the detection of subperiosteal resorption at compression ratios of 20:1 when full-frame DCT was used. (8)
Wavelet transforms are explained in a highly readable book (3), in a book chapter (9) and in numerous papers (10-13). When used to achieve high compression, wavelets can cause images to appear smooth, with a wavy appearance, sometimes described as looking similar to grains of rice. Because wavelets operate on a entire image at once, they avoid the "blockiness" associated with JPEG methodology.
Compression with wavelets may offer advantages over current compression techniques, such as the ability to progressively zoom in on an image. This scalability allows a single wavelet file to produce images with different resolutions for different needs. In the popular press, there is excitement about wavelet compression, with one author enthusing, "Save space! Shrink image files with little or no quality lost!" (14) However, the results of this author's study led to a different conclusion. It was JPEG with DCT, not wavelet, that was able to compress the hand radiograph to high compression ratios without visual loss of quality. (Fig 7)
Our study showed that, when applying the LuraTech wavelet implementation to the hand radiograph, compression ratios greater than 7:1 led to over-smoothed images with loss of trabecular detail. This rapid fall-off in quality, at even modest wavelet compression ratios, was also reported in studies of ultrasound images by Persons et al. (5,15) There may be good theoretical reasons for their findings. As noted by Ericson et al., the trabecular pattern in bone radiographs, and speckle in ultrasound examinations, are particularly sensitive to blurring from compression because they, like random noise, are characterized by numerous, high-frequency coefficients, which wavelet compression removes. (11) Indeed, slightly compressed images, precisely because they tend to have less noise, are sometimes preferred by observers. (1, 16)
Because wavelets are a class of functions, it is possible for one implementation to outperform another. (17) However, regardless of whether this author used the wavelet implementation from LuraTech, SPIHT (Set Partitioning in Hierachical Trees) (18) or LizardTech, only modest image compression could be achieved before loss of detail became visually evident. The consumer version of the LizardTech's image compression program is currently limited to files no greater than 1600 x 2100 pixels, smaller than the digitized hand radiograph used in our study, making it less useful than LuraTech's software, which can process images up to 4096 x 4096 pixels.

Comparing Images Compressed with Wavelets vs. JPEG with DCT

Our study found that JPEG with DCT achieved higher compression than wavelets with less loss of detail visually. (Fig 6) The reasons for this result are not entirely clear. However, Persons et al. suggest that, although wavelet transforms may be more mathematically correct, JPEG with DCT compression is optimized for visual perception and, therefore, may be more visually correct. (10)
Interestingly, it was easier to compress the hand radiograph than the CT slice, regardless of whether lossless wavelet (compression ratio 7:1 vs 1.6:1) or lossless JPEG (compression ratio 2.4:1 vs. 1.6:1) was used. Erickson obtained similar results when comparing the higher wavelet compressibility of chest radiographs with CT and MR images. (11)
Goldberg et al. (19), citing Gillespy et al. (20) and Chan et al. (21), suggest that lower compressibility relates to smaller matrix size and less pixel-to-pixel correlation. These authors would therefore predict that the CT slice being smaller than the hand radiograph (1.2 MB vs. 6.82 MB, respectively) should be less compressible, as observed in this study.
A second explanation specific for the wavelet transform, which filters the original image into high- and low-frequency sub-band is that higher frequencies, which contain progressively more detail, are usually more subject to compression than the low-frequency sub-band, which contains much of the information needed to reconstruct the image. Persons et al. found that the most compressible images have the greatest percentage of their information (energy) in the lowest frequency sub-band. (22) Chest radiographs, for example, which have 99.69% of their energy in the lowest-frequency sub-band, are the most compressible, whereas CT and MR images, which have, respectively, only 92.12% and 78.03% in the same sub-band, are less compressible. As pointed out by Persons et al., this analysis ignores issues such as the differing levels of compressibility of various structures within an image. Nonetheless, the findings discussed in this paper may account for the different levels of acceptance of, and enthusiasm for, wavelet compression found in the literature, depending on the imaging modality studied and the pathology to be detected.
Although studies in the literature differ in the way images are obtained (e.g., analog vs. digital), the imaging modality evaluated (e.g., chest radiography vs. CT), the medium used for interpretation (e.g., film vs. computer monitor), the specific pathology sought, the endpoint of the study (diagnostic accuracy vs. loss of image quality), etc., there does appear to be general agreement with our results. For example, Slone et al., studying compressed images on a workstation, found not only that JPEG and wavelet compression ratios needed to be in the 8:1 to 10:1 range to be visually lossless, but concluded similar to this author's:
"Despite expectations for improved performance with wavelet-based algorithms, we found that the JPEG baseline algorithm resulted in performance that was as good as, if not better than performance with the WTCQ at low compression ratios." (23) (WTCQ = wavelet-based trellis-coded quantization)

CONCLUSION

For archival purposes, lossless wavelet compression of medical images may be acceptable, but high levels of lossy compression are not. This is true both for images that depend on spatial resolution for their interpretation (e.g., radiographs) as well as for those that depend on contrast resolution (e.g., CT scans).
Specific applications, such as teaching or Internet publishing, may allow quite high compression with adequate results. Indeed, when the author compressed the original CT image 40:1 using the LuraTech wavelet algorithm, and then scaled it to 50% of its original size, the resulting image still demonstrated the important clinical findings. Although this may be suitable for educational purposes (Fig 8), smoothing artifact was still present and one would hesitate to use it for diagnostic purposes. Therefore, when compressing medical images, one must consider not only of the uses that were intended by also on those that might be made.
In response to studies that show that compressed images retain sufficient detail to diagnose a specific pathology, this author would argue that diagnostic quality must be maintained for every possible pathology, now and in the future, which is a much more stringent requirement.
This study also suggests that, for greyscale medical images, JPEG compression with DCT may be more maligned than warranted. Nonetheless, this author recommends lossless or minimal compression unless the user conducts the extensive testing required for lossy compression. Blume asks the question, "Are you afraid of data compression?" and concludes that you should not fear it. (24) Based on this study, however, the author is more inclined to agree with Kivijari et al. when they state:
"There is always the possibility that a vague detail might give a reason to suspect some critical changes in a patient's condition. For this reason, the lossy techniques, which tend to give high compression ratios, such as 1:10 and 1:30, are not acceptable in medical image compression." (25)
(Readers who would like to participate in an active discussion of image compression are referred to Internet newsgroups, such as comp.compression. Also available are basic introductions to image file formats (2, 26) and image compression. (27, 28)

ACKNOWLEDGEMENT

The author would like to thank Linda E. Ketchum for careful review of this manuscript and helpful advice.
September, 2001

Table 1.

Hand Radiograph: Results of Compression

File Format Size (MB) Mode Compression Ratio Perceived Quality

TIFF 6.82 N 1.0:1 ***

TIFF LZW 4.00 LL 1.7:1 *

BMP 6.82 N 1.0:1 *

BMP RLE 5.69 LL 1.2:1 *

GIF 4.46 LL 1.5:1 *

PNG 3.69 LL 1.8:1 *

JPEG 100.0 2.82 L 2.4:1 *

JPEG 3.2 0.36 L 19:1

JPEG 1.0 0.17 L 40:1 *

JPEG 0.1 0.08 L 85:1 *

WAVELET 100 0.98 L 7:1 ***

WAVELET 99 0.58 L 12:1

WAVELET 98 0.37 L 18:1 *

Perceived Quality *** = Best
Mode N = No Compression
Mode L = Lossy Compression
Mode LL = Lossless Compression

Table 2.

CT of the Abdomen: Results of Compression

File Format Size (MB) Mode Compression Ratio Perceived Quality

TIFF 1.21 N 1.0:1 *

TIFF LZW 0.90 LL 1.3:1 *

BMP 1.21 N 1.0:1 *

BMP RLE 1.41 LL 0.8:1 *

GIF 1.02 LL 1.2:1 *

PNG 0.79 LL 1.5:1 *

JPEG 100.0 0.77 L 1.6:1 *

JPEG 3.2 0.15 L 22:1 *

JPEG 1.0 0.06 L 35:1 *

JPEG 0.1 0.02 L 67:1 *

WAVELET 100 0.76 L 1.6:1 ***

WAVELET 89 0.12 L 10:1

WAVELET 85 0.04 L 28:1 *

WAVELET 80 0.03 L 45:1 *

Perceived Quality *** = Best
Mode N = No Compression
Mode L = Lossy Compression
Mode LL = Lossless Compression

REFERENCES

Erickson BJ, Manduca A, Persons KR, et. al. Evaluation of irreversible compression of digitized posterior-anterior chest radiographs. J Digit Imaging 1997; 10(3):97-102.

Webster, T. "Web Designer's Guide to Graphics: PNG, GIF & JPEG." Indianapolis, IN: Hayden Books, 1997.

Hubbard, BB. The World According to Wavelets: The Story of a Mathematical Technique in the Making, 2nd ed. Natik, MA: A.K. Peters, Ltd., 1998.

Johnson RC, JPEG 2000 wavelet spec approved. EETimes.com, 1999 Dec 12. http://www.eetimes.com/story/industry/systems_and_software_news/OEG19991228S0028

Persons KR, Palisson PM, Manduca A, et al. Ultrasound grayscale image compression with JPEG and wavelet techniques. J Digit Imaging 2000; 13(1):25-32.

JPEG FAQ http://www.faqs.org/faqs/jpeg-faq/

Toney MO, Dominguez R, Dao HN, Simmons G. The effect of lossy discrete cosine transform compression on subtle fractures. J Digit Imaging 1997; 10(4):169-73.

Sayre JW, Ho BK, Boechat MI, Hall TR, Huang HK. Subperiosteal resorption: Effect of full-frame image compression of hand radiographs on diagnostic accuracy. Radiology 1992 Nov; 185(2):599-603.

Goldberg M. Image Compression. In: Siegel EL, Kolodner RM, ed. Filmless Radiology. New York: Springer Verlag, 1999: Chap. 15, 295-310.

Reiter, E. Wavelet Compression of Medical Imagery. Telemed J 1996; 2(2):131-137.

Erickson BJ, Manduca A, Palisson P, t. al. Wavelet Compression of Medical Images. Radiology 1998 Mar; 206(3): 599-607.

Schomer DF, Elekes AA, Huffman JC, et al. Introduction to wavelet-based compression of medical images. Radiographics 1998; 18(2):469-481.

Strintzis MG. A review of compression methods for medical images in PACS. Int J Med Inf 1998; 52(1-3):159-165.

Sawalich, W. LuraTech LuraWave. PC Photo 2000; July/August: 80.

Persons KR, Hangiandreou NJ, Charboneau JW, et al. Clinical evaluation of irreversible compression of ultrasound images using the JPEG algorithm at approximately 9:1. J Digit Imaging 2000; 13(2 Suppl 1):191-192.

Smith I, Roszkowski A, Slaughter R, Sterling D. Acceptable levels of digital image compression in chest radiology. Australas Radiol 2000; 44(1):32-35.

Mitra S, Yang S., Kustov V. Wavelet-based vector quantization for high-fidelity compression and fast transmission of medical images. J Digit Imaging 1998; 11(4 Suppl 2):24-30.

Center for Image Processing Research (CIPR), Rensselaer Polytechnic Institute (RPI), SPIHT

Goldberg MA, Gazelle GS, Boland GW. Focal hepatic lesions: effect of three-dimensional wavelet compression on detection at CT. Radiology 1997 Jan; 202(1):159-165.

Gillespy T 3rd, Rowberg AH. Displaying radiologic images on personal computers: image storage and compression: Part 1. J Digit Imaging 1993 Nov; 6(4):197-204.

Chan KK, Lou SL, Huang HK. Full-frame transform compression of CT and MR images. Radiology 1989 Jun; 171(3):847-851.

Persons KR, Palisson PM, Manduca A, et. al. An analytical look at the effects of image compression on medical images. J Digit Imaging 1997; 10(3 Suppl 1):60-65.

Slone RP, Foos DH, Whiting BR, et al. Assessment of visually lossless irreversible image compression: comparison of three methods by using and Image-comparison workstation. Radiology 2000; 215(2):543-553.

Blume H. DICOM (Digital Imaging and Communications in Medicine) state of the nation. Are you afraid of Image Compression? Adm Radiol J 1996; 15(11):36-40.

Kivijarvi J, Ojala T, Kaukoranta T, Kuba A, Nyul L, Nevalainen O. A comparison of lossless compression methods for medical images. Comput Med Imaging Graph 1998; 22(4):323-339.

Celeste A. The future of web design -- Parts I-III. designer.com

Schilling RB, ed. Understanding Compression. Great Falls, VA: Society for Computer Applications in Radiology, 1997.

Erickson BJ. Irreversible Compression of Medical Images. Great Falls, VA: Society for Computer Applications in Radiology, 2000.

[SUMMARY]

The author investigated the effects of lossless and lossy compression on two types of archived medical images for which interpretation depends on (1) high spatial resolution, e.g. a bone radiograph, and (2) high contrast resolution, e.g., abdominal computed tomography (CT) study.
A typical bone radiograph (anterior-posterior [AP] view of a hand with scleroderma) and abdominal CT study (7-mm contrast-enhanced transaxial slice from a patient with adult-onset polycystic kidney disease) were compressed and decompressed using commonly available file formats and lossless and lossy algorithms. Different levels of compression were used for lossy algorithms. Resulting image quality was assessed qualitatively by five board-certified radiologists.
We found that image quality was preserved with lossless compression, (e.g. TIFF, GIF, BMP), JPEG and wavelet compression in their lossless modes, and relatively low levels of lossy decompression (~2.5 to 1). Higher levels of lossy JPEG and wavelet compression, resulted in visible image degradation, sooner for wavelet than for JPEG.
Lossy wavelet compression of the bone radiograph led to loss of trabecular detail although pathology, such as calcification and erosion, remained visible. Overall, the abdominal CT image was more resistant to compression than the bone radiograph.
It is possible that no one lossy compression ratio with currently available algorithms will yield decompressed images of diagnostic quality for all types of radiology images. The level of compression that preserves clinically acceptable image quality may depend on the modality, the anatomy, and the pathology.
At present, the author recommends lossless wavelet or JPEG compression algorithms for medical image archiving.

List of Publications -->
Return To Home -->