Lossy JPEG compression: easy to compress, hard to compare

Dentomaxillofacial Radiology (2006) 35, 67–73 q 2006 The British Institute of Radiology http://dmfr.birjournals.org

REVIEW

Lossy JPEG compression: easy to compress, hard to compare A Fidler*,1, B Likar2 and U Skalericˇ3 1 University of Ljubljana, Faculty of Medicine, Department of Restorative Dentistry and Endodontics, Ljubljana, Slovenia; 2University of Ljubljana, Faculty of Electrical Engineering, Ljubljana, Slovenia; 3University of Ljubljana, Faculty of Medicine, Department of Oral Medicine & Periodontology, Ljubljana, Slovenia

Objectives: To review the literature on lossy compression in dental radiography and to discuss the importance and suitability of the methodology used for evaluation of image compression. Methods: A search of Medline (from 1966 to October 2004) was undertaken with the search expression “(Radiography, dental) and compression”. Inclusion criterion was that the reference should be evaluating the effect of lossy image compression on diagnostic accuracy. For all included studies, information in relation to mode of image acquisition, image content, image compression, image display, and method of image evaluation was extracted. Results: 12 out of 32 papers were included in the review. The design of these 12 studies was found to vary considerably. Parameters used to express the degree of information loss (DIL) were either or both compression ratio (CR) and compression level (CL). The highest acceptable CR reported in the studies ranged from 3.6% to 15.4%. Furthermore, different CR values were proposed even for the same diagnostic task, for example, for caries diagnosis CR ranged from 6.2% to 11.1%. Conclusion: Lossy image compression can be used in clinical radiology if it does not conflict with national law. However, the acceptable DIL is difficult to express and standardize. CR is probably not suitable to express DIL, because it is image content dependent. CL is also probably not suitable to express DIL because of the lack of compression software standardization. Dentomaxillofacial Radiology (2006) 35, 67–73. doi: 10.1259/dmfr/52842661 Keywords: radiography, dental; radiography, dental, digital; data compression Introduction Digital radiography offers several advantages over conventional radiography such as ease of image manipulation, computer analysis, storage, retrieval and transmission of images. Acquisition of digital images, as a first step in digital radiography, can be performed with digitization of conventional films or directly with digital systems. The latter offers additional advantages, such as dose reduction, reduction of time consumption, and the fact that there is no need for film processing.1 Digital images are stored as computer files in picture archive and communication systems (PACS). With increasing utilization of digital radiography, storage (hard disk and archival media size) and transmission (bandwidth of computer networks) requirements are also increasing. These requirements can be considerably reduced by image compression, which can be either lossless or lossy. Lossless image compression methods *Correspondence to: Alesˇ Fidler, Department of Restorative Dentistry and Endodontics, Hrvatski trg 6, SI-1000 Ljubljana, Slovenia; E-mail: [email protected] Received 14 December 2004; revised 20 June 2005; accepted 12 July 2005

preserve all image information and their use is not questioned. However, the approximate compression ratio (CR) is 1:2 to 1:4, depending on image characteristics.2 Lossy compression methods can achieve much higher CR, up to 1:40.3 However, increasing the degree of information loss (DIL) will eventually lead to erroneous diagnosis. Therefore, to efficiently reduce the file size the highest DIL still maintaining diagnostic accuracy should be utilized. Unfortunately, there is no single acceptable maximum CR for all diagnostic tasks, even for the images of same modality.4 Therefore, numerous studies are required for every specific diagnostic task.4 Several studies have sought to determine the highest acceptable CR for different diagnostic tasks in dental radiology. However, recommendations on this issue differ considerably. In early work describing digital dental radiography, the use of lossy compression was discouraged due to legal issues.5 In a review of digital imaging for caries diagnosis, lossy image compression of approximately 1:12 was recommended for this task.1 More recently, lossy image compression was not recommended as a default storing method for digital dentomaxillofacial

Lossy JPEG compression: a review A Fidler et al

68

radiography due to possible interference with automated image analysis systems.6 Use of lossy image compression was recommended for image transmission in the same paper,6 but no recommendations regarding CR were given. Since the year 2000, the use of lossy compression has been accepted by the Food and Drug Administration (FDA), but the compression method and ratio are left to the radiologist’s discretion.7 Therefore, recommendations for use of lossy image compression in dental radiology are required. However, it should be emphasised that because the legislation about the use of lossy compression varies between the countries, general scientific recommendations are applicable only if they are not in conflict with national law. The aim of this review is to discuss the importance and suitability of the methodologies used in previous lossy compression studies and to draw recommendations for further research and clinical use.

Methods A search of Pubmed Medline (1966 to October 2004) was undertaken with the search expression “(Radiography, dental) AND compression”. Studies were included in further analysis if they were evaluating the effect of lossy image compression on diagnostic accuracy. For all such included studies, information relating to the mode of image acquisition, image content, image compression, image display, and method of image evaluation was extracted.

Results The Medline search returned 32 references. After reading the titles and abstracts to assess suitability against the inclusion criteria, the complete text of 18 papers that investigated lossy image compression was retrieved. 12 of these papers fulfilled the critera8 – 19 (Table 1) and were included in review, while six of them were excluded1,5,6,20 – 22 (Table 2). In general, studies assessing the influence of image compression on diagnostic accuracy share a similar experimental design. A set of in vitro or in vivo images, obtained with one of the digitalization methods, is compressed with one or more compression methods at several levels of DIL. Compressed images are then evaluated with one or more methods.4 In other words, for each DIL (independent variable) the diagnostic value of the compressed image and the amount of file size reduction (dependent variables) are assessed, and the highest acceptable DIL with corresponding file size reduction is reported. However, despite the similarity of study designs, many features that could possibly affect the results and their clinical applicability differ between the studies. For this reason, information relating to image acquisition, image content, image compression, image display and image evaluation was extracted, presented in the following sections and summarized (Table 1). Dentomaxillofacial Radiology

Image acquisition Methods for image acquisition in the reviewed studies can be divided into three groups: scanning of conventional films,11,16,17 storage phosphor systems8 – 10,12,14,15,18, and charge coupled device based systems8,13,19 (Table 1). Image resolution and image noise are properties of a digital image that are determined by mode of image acquisition. Image resolution depends on the receptor size and on the spatial resolution of the detector. The size of the uncompressed image file increases with the square of spatial resolution, that is, a twofold increase of spatial resolution results in fourfold increase of file size. Increased spatial resolution also results in increased noise level relative to the signal23 and this can cause a larger error in caries diagnosis10,12 and a reduction in the efficiency of lossy image compression.10,12 Despite the effect of spatial resolution on compressibility of images,11 many of the compression studies neglect to report its value. Spatial resolution was reported for all studies utilizing scanned conventional film: 300 dpi16,17 and five different resolutions (100 dpi, 200 dpi, 300 dpi, 400 dpi and 600 dpi) in one study.11 Spatial resolution was reported in only three out of nine studies utilizing digital systems: 300 dpi utilizing DenOptix,18 and 360 dpi10 or 400 dpi14,15 utilizing Digora. Spatial resolution of digital systems increases with technological development. However, higher resolution does not necessarily increase diagnostic accuracy. In a study comparing three scanning resolutions (150 dpi, 300 dpi and 600 dpi) for caries diagnosis, it was demonstrated that 300 dpi performed best.24 A similar study comparing resolutions of 200 dpi, 400 dpi and 600 dpi reported better results with 200 dpi or 400 dpi than with 600 dpi.25 Therefore, the lowest spatial resolution adequate for diagnostic accuracy should be used to reduce file size efficiently. Image content Characteristics of the radiographed object, properties of the acquisition system and subsequent modifications of the image determine the image content. In as early as the second study on image compression in dental radiology in 1996,9 it was stated that compression efficiency depends on spatial frequency content in the image; in other words, CR depends on image content. This was confirmed with studies demonstrating that an increase of the level of added noise results in an increase of compressed file size.10,12 According to the authors’ knowledge, no other studies assessing the influence of image content on compressed file size have been published to date. Currently there is no database of standard images freely available for compression studies in dental radiography. Consequently, radiographed objects and projections utilized in the reviewed studies differ considerably between the studies. In vitro studies utilized periapicals of dry pig mandible specimens,14,15 periapicals of extracted human premolars and molars embedded in a plaster,8,9,18 bitewings simulated with extracted teeth embedded in plaster16 and periapicals of human jaw cadaver specimens19 In vivo studies utilized bitewings10,12 and periapicals.11,13,17 Image content differs considerably

Table 1 Summary of data extracted from reviewed studies relating to image acquisition, image content, compression algorithm and software, compression scale, compression levels and corresponding file size reduction (% of original), and evaluation of image compression Compression algorithm and software

Compression scale

extracted human molars and premolars, three in a line, embedded in plaster extracted human molars and premolars, three in a line, embedded in plaster

J; Adobe Photoshop 2.5.1

0 –9

CL 5 CR 8

diagnostic accuracy (caries)

J; Adobe Photoshop 2.5.1

0 –9

CL 3 5 7 9 CR 20.0 8.4 4.6 3.0

observer evaluation, diagnostic accuracy (caries)

CL 2 27 53 128 CR 30.6 7.0 4.6 2.9 CL 10 9 8 7 6 5 4 3 2 1 CR 39.8 24.8 15.7 11.3 8.6 6.4 6.9 4.7 4.9 4.4 4.0 CL 27 53 CR 7.1 4.6 (no noise) CR 11.1 6.0 (low noise) CR 18.1 9.7 (medium noise) CR 25.0 13.5 (high noise) CL CR 50.0 25.0 12.5 6.3 3.1 2.1 1.6 CL 90 70 50 30 CR 14.1 6.5 4.5 3.3 CL 90 70 50 30 CR 14.1 6.5 4.5 3.3 CL 27 CR 11.1

diagnostic accuracy (caries)

Image acquisition

Wenzel (1995)8

Digora RVG Sens-a-Ray Visualix

Wenzel (1996)9

Digora

Janhom (1999)10

Digora

clinical bitewings

J; Leadtools

1 –255

Yuasa (1999) #1

Film

clinical periapicals of lower molar region

J; Adobe Photoshop 4.0

10 – 0

Janhom (2000)12

Digora

clinical bitewings

J; Leadtools

1 –255

Eraso (2002)13

Schick

clinical periapical radiographs

J; DCTune

#2

Fidler (2002)14

Digora

periapicals of pig mandible

J; MS Photo Editor

100 –1

15

Digora

periapicals of pig mandible

100 –1

Janhom (2002)16

Film

Siragusa (2002)17

Film

extracted human premolars and molars, embedded in plaster, simulating bitewings clinical periapicals

J; MS Photo Editor W; LEAD Image Viewer þ J; Leadtools W; ViewMed

Pabla (2003)18

DenOptix

Koenig (2004)19

Schick CDR

Fidler (2002)

extracted human premolars and molars, embedded in plaster human jaw cadaver specimens

1 –255

J; Adobe Photoshop 5.0

10 – 0

J; VixWin 2000

100 –#3

not specified

100 –#3

CL 10 9 8 7 6 5 4 3 2 1 0 CR 44.2 31.6 21.4 15.4 11.9 9.0 10.1 8.1 7.0 6.2 5.2 CL 100 75 50 CR 39.7 9.0 6.2 CL 100 75 50 40 20 CR 50 7.1 4.3 3.6 2.1

Evaluation of image compression

observer evaluation numerical analysis, diagnostic accuracy (caries)

diagnostic accuracy (periapical lesions) diagnostic accuracy (digital subtraction radiography) numerical analysis, diagnostic accuracy (digital subtraction radiography) numerical analysis, diagnostic accuracy (caries) numerical analysis, observer evaluation

Lossy JPEG compression: a review A Fidler et al

Author (year)

11

Compression levels (CL) and corresponding compression ratios (CR) (% of original size)

Image contents

diagnostic accuracy (caries) diagnostic accuracy (periapical lesions)

69

Dentomaxillofacial Radiology

J, JPEG compression; W, wavelet compression; CL, compression level; CR, compression ratio; Bold, the highest acceptable CL and CR reported #1 Original file size was not reported in the paper, CL values were calculated on assuming an original bitmap file size is 40 mm £ 30 mm at resolution of 400 dpi. #2 Image dependent quantization table optimization27 was used to achieve adjusted CR #3 Lower end of compression scale was not reported in the paper

Lossy JPEG compression: a review A Fidler et al

70 Table 2

List of papers excluded from the review

Author (year)

Reason for exclusion

Godfredsen (1996)20 5

Veersteg (1997) Wenzel (1998)1 Wenzel (1999)22

Godfredsen (2001)6 Gurdal (2001)21

Evaluating observers’ use of image-enhancement facilities on digital images Review Review Methodological study on the effect of the choice of gold standard on the diagnostic outcome of approximal caries detection in original and compressed digital radiographs Paper on demonstrating effects of lossy JPEG compression and giving recommendation on use of lossy compression Technical report on variations of radiodensity data introduced by lossy JPEG compression and/or the use of three software programs

between the studies and this variability might have an influence on compressibility and diagnostic accuracy. Image compression Discrete cosine transformation based Joint Photographic Experts Group (JPEG) is by far the most often used image compression method in medical and dental radiology, with wavelet compression as an emerging method. Development of computer programs is rapid and, after the publication of the first study on image compression in dental radiology in 1995,8 a large number of compression programs and their versions have become available and utilized in the subsequent studies (Table 1). Commercial programs were mainly used: different versions of Adobe Photoshop (Adobe Systems Inc., San Jose, CA),8,9,11,17 Leadtools (Leadtools Technologies Inc., Charlotte, NC),10,12,16 MS PhotoEditor (Microsoft Corp., Redmond, WA),14,15 and VixWin 2000 (Gendex Imaging System, Milano, Italy).18 JPEG compression method with image content dependent quantization table26 was used in one study.13 In one study, JPEG compression software was not specified.19 In the majority of JPEG compression software, DIL can be manually adjusted on the compression scale. Even though JPEG is an ISO standard,27 its compression scale is not standardized. Consequently, different programs, or even different versions of the same program, have different or even opposing compression scales; that is, in some programs low value means low but in others high DIL. A value for the lowest and the highest DIL on compression scale of computer programs was extracted from the reviewed studies (Table 1). Compression level (CL) is a value selected on the compression scale that expresses the DIL. CL is also referred to as quality or compression factor but the term CL will be used throughout this review. In three studies, only one CL was used to show that lossy compression can be applied in dental radiology,8 to assess the influence of image noise on diagnosis and compressibility,10 and to compare JPEG and wavelet compression at the same CR.16 In the other studies, from two to 11 CL were arbitrarily chosen (Table 1). Dentomaxillofacial Radiology

File size reduction is expressed by CR, which is defined as a ratio between compressed and original file size. It can be expressed by ratio or percentage. The highest acceptable CRs reported for different tasks in dental radiography are different, and ranged from 3.6%19 to 15.4%17 (Table 1). Different CR values are proposed even for the same diagnostic task. For instance, for caries diagnosis the reported CR ranged between 6.2%18 and 11.1%.16 A large number of computer programs increases the possibility for errors in implementation of a compression method. For example, it has been demonstrated for one computer program that decompression of JPEG images is not performed correctly.21 This might lead to erroneous and misleading results in evaluating compressed images. In addition, a possible flaw of the compression scale in Adobe Photoshop was reported in a recent study.17 On the compression scale of this program, lower CL represent higher DIL and should result in higher CR. This was not true for CL 4 and 5, for which an opposite relationship between CR and CL was found. In one of the reviewed studies, adjusting CR instead of CL was utilized.13 In this implementation26 the DIL was not set solely by the user but also depended on the image content. Such an approach can be misleading because the same detail that would be preserved in a simple image could be lost in a more complex one in order to achieve the desired CR. Wavelet based compression algorithms were used in only two studies: LEAD Image Viewer (Leadtools Technologies Inc., Charlotte, NC) was used for JPEG2000 compression15 and Viewmed (Pegasus Imaging Corp., Tempa, FA) was used for another wavelet based compression method.16 In medical radiology, the use of wavelet compression seems promising,28 however, due to the lack of studies no relevant conclusion on wavelet compression can be made in dental radiology. Image display For human observer based evaluation, digital images must be displayed on a computer monitor. While the type of display does not seem to have an influence on observer performance,29,30 the opposite is true for image magnification.25,31 – 33 Spatial resolution of computer monitors is typically between 75 dpi and 100 dpi, which is three to five times lower than the spatial resolution of the image receptors used in dental radiology. Consequently, displaying all image pixels in one-to-one correspondence results in magnification of the displayed image. If images are displayed in their original size, then loss of diagnostic information occurs.31 Reviewed studies reported image display parameters inadequately. For example, in some studies no image display details were reported,8,9,17 or only the resolution of display was reported.10 – 13,16,32 In another study, resolution of display and image size in inches was reported,19 and in only one study were type and resolution of the display as well as pixel-to-pixel correspondence reported.18 In two studies, image display parameters were not applicable due to image analysis with digital subtraction radiography.14,15 Such a variety in image display parameters, important for human observer evaluation,

Lossy JPEG compression: a review A Fidler et al

makes objective comparison impossible. To utilize all the image information stored in a digital image, images should be displayed in one-to-one pixel correspondence. Otherwise, increased image resolution is not beneficial for diagnostic accuracy and only increases file size, as has been described previously. Image evaluation In the reviewed studies, images have been evaluated by one or more methods from all three groups:4 (1) numerical analysis of pixel values before and after compression, (2) subjective observer evaluation focusing on aesthetic acceptability and estimated diagnostic value, and (3) objective measurement of diagnostic accuracy using blinded evaluation methods. The first evaluation approach (numerical analysis) was used in four studies: standard deviation of pixel grey values of subtraction images,12,16 mean and standard deviation of pixel grey values of subtraction images,17 average pixel error and peak signal to noise ratio.15 Those parameters correlate poorly with visual quality or diagnostic value of images and cannot replace human observer evaluation.4 Nevertheless, one study recommended critical values for standard deviation of pixel value differences of 1.4 for diagnosis and 2.4 for illustration.17 The second evaluation approach (subjective observer evaluation focusing on aesthetic acceptability and estimated diagnostic value) was used in three studies. General image quality was assessed by the visual analogue scale11 or by scoring the images on an 11-point scale.9 Image quality of eight anatomical structures was scored on a 5-point scale.11 Image quality for endodontic diagnosis was assessed by scoring the images using a 3-point scale.17 The third evaluation approach (objective measurement of diagnostic accuracy for specific diagnostic tasks) was used in 11 studies. Specific diagnostic tasks evaluated in reviewed studies (Table 1) were presence of caries,8,9,18 presence and depth of caries,10,12,16 changes in alveolar bone measures with digital subtraction radiography14,15 and presence of periapical lesions on periapical radiographs.19 Type of changes evaluated in the reviewed studies can be divided into four groups: natural teeth with caries lesions,8 – 10,12,16,18 changes in alveolar bone simulated by adding bone chips with different masses14,15 and periapical lesions obtained from clinical radiographs17 or chemically created in human jaw cadaver specimens.19 The true diagnosis for caries was obtained either from histological evaluation,8,9,16,18 or a panel of experts, interpreting conventional radiographs.10,12 In the latter two studies, conventional films were exposed together with image plates, so not only the influence of compression, but also the influence of receptor was affecting the results. The true diagnosis of periapical lesions was obtained from original images.19 For the assessment of the alveolar bone loss with digital subtraction radiography,14,15 the true values were also obtained from original images. The importance of methodology for diagnostic test evaluation has been described extensively.34 Briefly, a

71

diagnostic method under evaluation should be validated against true diagnosis, usually named as “gold standard”, which should fulfil the following criteria:34 (1) it should be established by a method that is precise, (2) it should reflect the patho-anatomical appearance of disease, and (3) it should be established independently of the diagnostic method under evaluation. Those three requirements are difficult to fulfil, especially in in vivo studies. When no valid expression of the true state of the disease can be obtained, a study of precision (reproducibility/repeatability/replicability/intraobserver and interobserver variation) may be considered. 1 In such situations, an uncompressed image or conventional film radiograph may be used as a true diagnosis. As a consequence, a tested diagnostic method cannot outperform an established diagnostic method that was selected as a true diagnosis.34 The importance of the effect of the choice of gold standard on the diagnostic outcome of approximal caries detection in original and compressed digital radiographs was studied.22 It was demonstrated that using the observer as the gold standard instead of obtaining true gold standard by microscopy results in significantly higher accuracy and, paradoxically, higher accuracy of the compressed, degraded images compared with the originals.22 In summary, for evaluation of influence of image compression on diagnostic accuracy, a true diagnosis should be used whenever possible. Otherwise, a study should be described and interpreted as a study of precision (reproducibility/repeatability/replicability/intraobserver and interobserver variation). Discussion When lossy compression is to be used in PACS for clinical use, two parameters must be known. First, the highest acceptable DIL is required to adjust the compression software, and second, the expected file size of compressed image is required to plan hardware requirements, such as disk storage requirements and network bandwidth. Studies evaluating the influence of compression on diagnostic accuracy should therefore report parameters that are needed for the set up of PACS. Degree of information loss DIL in reviewed studies was expressed with CR and CL. In general, CR is used more frequently, probably due to the benefit of expressing DIL and file size reduction with one parameter. This parameter can also be compared between the studies. CR as a measure of information loss is even used in the FDA Guidance for the Submission of Premarket Notifications for Medical Image Management Devices, which states: “A message stating that irreversible compression has been applied and the approximate compression ratio should accompany images that have been subjected to irreversible compression.”7 However, it has to be emphasised that CR depends not only on the DIL, but also on the image content of the original image. For example, a simple image would achieve higher CR than a Dentomaxillofacial Radiology

Lossy JPEG compression: a review A Fidler et al

72

more complex one at the same DIL. Therefore, CR might not be the most suitable parameter to express DIL. CL is the second most frequently used parameter to express the DIL. All studies except one13 reported the CL that was applied (Table 1). Expressing the information loss by the CL seems ideal as it is image content independent. However, a closer look at the JPEG standard reveals the lack of standardization. CL is used to set the factors in the quantization table which are determining the DIL, but compression scale and factors in the quantization table are not standardized. Different compression scales, scales with opposing directions or even flaws in compression scales17 can be found in different software. The only parameter that expresses the DIL in JPEG compression is a quantization table. A quantization table, as presented in the JPEG standard,35 is used most often, although its use is not required by the standard. The use of the quantization table has two practical limitations in that it is specific for JPEG, and it contains 64 numbers that are not practical to compare. It has to be emphasised that images scanned at lower spatial resolutions must be compressed with lower DIL than images scanned at higher spatial resolution in order to achieve comparable quality.11 Therefore, DIL expressed with a quantization table is spatial resolution specific and may not be generalized to other resolutions. An ideal parameter for expressing the DIL should: (1) express the degree of medically important information loss, (2) be independent of compression method, (3) be independent of image content and (4) be simple to express. To express DIL, numerical analyses of pixel values remain very attractive, as they are less time consuming than human observer based evaluation. Recently, new approaches in numerical analysis have been proposed that report better correlation with subjective observer ratings,36 – 39 but they have not been used for evaluation of image compression in dental radiology yet. Another approach that has been proposed is image quality assessment based on medical image quality concepts.40 It considers image compression as one of the modules in a digital imaging system and should, therefore, be evaluated by the same methods. To express the DIL more objectively, a set of quantitative measurements of medical image quality (input/output response curve, high contrast resolution, low contrast discrimination and noise) that are used for the evaluation of different modules of digital imaging system (digitizer, display, image processing, printer) are proposed.40 This type of assessment provides information regarding the type of loss and allows objective comparison between different compression methods. File size reduction File size reduction is usually expressed by CR. As the aim of image compression is efficient storage of images, file size of the compressed image would be a more appropriate parameter than CR. As already discussed, the original file Dentomaxillofacial Radiology

size depends on spatial resolution. Therefore, to reduce file size efficiently, an image should be acquired with the lowest spatial resolution adequate for diagnosis and then further compressed. Image content is another factor determining file size of compressed image. Except for consideration of noise, there is a lack of studies addressing this issue. However, there are two possibilities to demonstrate influence of image content: (1) within the studies in which CRs were reported for individual images, and (2) between the studies that compressed different images at the same DIL. For the first possibility, a study evaluating the limit of image compression for endodontic diagnosis is appropriate.17 Reported CR of 14 clinical periapical radiographs ranged between 40.9% and 46.7% at lowest DIL. For the second possibility, where software with comparable compression scales must be utilized to compress different images, two comparisons can be made. The first comparison can be made between the two studies by Janhom et al.10,12 In the earlier study, images of clinical bitewings were compressed to 7% at CL 27,10 while in the latter, images of extracted teeth, embedded in plaster, were compressed to 11.1% at the same CL.16 The second comparison can be made between the two studies using two versions of Adobe Photoshop with the same compression scale (Table 1). Higher CR values at all CLs were reported in the earlier study11 than in the later one.17 Therefore, it is possible that CR achieved with in vitro images containing large uniform areas (blank areas or plaster instead of bone with trabecular structure) achieve a higher CR than real clinical images at the same DIL. According to the authors’ knowledge, there are no published papers on the influence of image content on CR; for example, the number of teeth, the height of bone or the amount of blank areas. Conclusion In conclusion, lossy image compression can be used in clinical radiology, but acceptable DIL is difficult to assess and to standardize. CR is probably not suitable to express DIL because it is image content and spatial resolution dependent. CL is also probably not suitable to express DIL due to lack of compression scale standardization. Consequently, the use of these parameters for comparison between studies and for PACS implementation is therefore questionable. It is common in dental radiology for a single image to be used for several diagnostic tasks. For example, diagnosis of caries, periapical periodontitis or periodontal disease. Therefore, DIL that is acceptable for all potential diagnostic tasks should be used for clinical images. Research aimed at ensuring efficient file size reduction should be focused on finding the lowest acceptable image resolution and on finding the most efficient compression method, while preserving diagnostically important information.

Lossy JPEG compression: a review A Fidler et al

73

References 1. Wenzel A. Digital radiography and caries diagnosis. Dentomaxillofac Radiol 1998; 27: 3 – 11. 2. Kang KS, Park HW. Lossless medical image compression by multilevel decomposition. J Digit Imaging 1996; 9: 11 – 20. 3. Erickson BJ, Manduca A, Persons KR, Earnest 4th F, Hartman TE, Harms GF, et al. Evaluation of irreversible compression of digitized posterior-anterior chest radiographs. J Digit Imaging 1997; 10: 97 – 102. 4. Erickson BJ. Irreversible compression of medical images. J Digit Imaging 2002; 15: 5– 14. 5. Versteeg CH, Sanderink GC, van der Stelt PF. Efficacy of digital intra-oral radiography in clinical dentistry. J Dent 1997; 25: 215 –224. 6. Gotfredsen E, Wenzel A. Image compression in storing and communicating digital radiographs. Int J Comput Dent 2001; 4: 273 –279. 7. FDA, Center for Devices and Radiological Health. Guidance for the Submission of Premarket Notifications for Medical Image Management Devices. 2000; http://www.fda.gov/cdrhode/guidance/ 416.pdf 8. Wenzel A, Borg E, Hintze H, Grondahl HG. Accuracy of caries diagnosis in digital images from charge-coupled device and storage phosphor systems: an in vitro study. Dentomaxillofac Radiol 1995; 24: 250 – 254. 9. Wenzel A, Gotfredsen E, Borg E, Grondahl HG. Impact of lossy image compression on accuracy of caries detection in digital images taken with a storage phosphor system. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1996; 81: 351 – 355. 10. Janhom A, van der Stelt PF, van Ginkel FC, Geraets WG. Effect of noise on the compressibility and diagnostic accuracy for caries detection of digital bitewing radiographs. Dentomaxillofac Radiol 1999; 28: 6– 12. 11. Yuasa H, Ariji Y, Ohki M, Naitoh M, Shiojima M, Ushida M, et al. Joint Photographic Experts Group compression of intraoral radiographs for image transmission on the World Wide Web. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1999; 88: 93 –99. 12. Janhom A, van der Stelt PF, van Ginkel FC. Interaction between noise and file compression and its effect on the recognition of caries in digital imaging. Dentomaxillofac Radiol 2000; 29: 20 –27. 13. Eraso FE, Analoui M, Watson AB, Rebeschini R. Impact of lossy compression on diagnostic accuracy of radiographs for periapical lesions. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2002; 93: 621 – 625. 14. Fidler A, Likar B, Pernus F, Skaleric U. Impact of JPEG lossy image compression on quantitative digital subtraction radiography. Dentomaxillofac Radiol 2002; 31: 106 – 112. 15. Fidler A, Likar B, Pernus F, Skaleric U. Comparative evaluation of JPEG and JPEG2000 compression in quantitative digital subtraction radiography. Dentomaxillofac Radiol 2002; 31: 379 –384. 16. Janhom A, van der Stelt PF, Sanderink GC. A comparison of two compression algorithms and the detection of caries. Dentomaxillofac Radiol 2002; 31: 257 – 263. 17. Siragusa M, McDonnell DJ. Indirect digital images: limit of image compression for diagnosis in endodontics. Int Endodont J 2002; 35: 991 – 995. 18. Pabla T, Ludlow JB, Tyndall DA, Platin E, Abreu Jr M. Effect of data compression on proximal caries detection: observer performance with DenOptix photostimulable phosphor images. Dentomaxillofac Radiol 2003; 32: 45 – 49. 19. Koenig L, Parks E, Analoui M, Eckert G. The impact of image compression on diagnostic quality of digital images for detection of chemically-induced periapical lesions. Dentomaxillofac Radiol 2004; 33: 37 –43.

20. Gotfredsen E, Wenzel A, Grondahl HG. Observers’ use of image enhancement in assessing caries in radiographs taken by four intraoral digital systems. Dentomaxillofac Radiol 1996; 25: 34 – 38. 21. Gurdal P, Hildebolt CF, Akdeniz BG. The effects of different image file formats and image-analysis software programs on dental radiometric digital evaluations. Dentomaxillofac Radiol 2001; 30: 50 –55. 22. Wenzel A, Hintze H. Comparison of microscopy and radiography as gold standards in radiographic caries diagnosis. Dentomaxillofac Radiol 1999; 28: 182 –185. 23. Hildebolt CF, Bartlett TQ, Brunsden BS, Hente NL, Gravier MJ, Walkup RK, et al. Bitewing-based alveolar bone densitometry: digital imaging resolution requirements. Dentomaxillofac Radiol 1994; 23: 129 – 134. 24. Janhom A, van Ginkel FC, van Amerongen JP, van der Stelt PF. Scanning resolution and the detection of approximal caries. Dentomaxillofac Radiol 2001; 30: 166 –171. 25. Attaelmanan A, Borg E, Grondahl HG. Digitisation and display of intra-oral films. Dentomaxillofac Radiol 2000; 29: 97 – 102. 26. Watson AB. DCTune: A technique for visual optimization of DCT quantization matrices for individual images. Society for Information Display Digest of Technical Papers 1993; XXIV: 946 –949. 27. Wallace GK. The JPEG still picture compression picture standard. Commun Assoc Computing Machinery 1991; 34: 30 –44. 28. Erickson BJ, Manduca A, Palisson P, Persons KR, Earnest 4th F, Savcenko V, et al. Wavelet compression of medical images. Radiology 1998; 206: 599 – 607. 29. Cederberg RA, Frederiksen NL, Benson BW, Shulman JD. Influence of the digital image display monitor on observer performance. Dentomaxillofac Radiol 1999; 28: 203 –207. 30. Ludlow JB, Abreu Jr M. Performance of film, desktop monitor and laptop displays in caries detection. Dentomaxillofac Radiol 1999; 28: 26 –30. 31. Versteeg CH, Sanderink GC, Lobach SR, van der Stelt PF. Reduction in size of digital images: does it lead to less detectability or loss of diagnostic information? Dentomaxillofac Radiol 1998; 27: 93 – 96. 32. Svanaes DB, Moystad A, Risnes S, Larheim TA, Grondahl HG. Intraoral storage phosphor radiography for approximal caries detection and effect of image magnification: comparison with conventional radiography. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1996; 82: 94 –100. 33. Moystad A, Svanaes DB, Larheim TA, Grondahl HG. Effect of image magnification of digitized bitewing radiographs on approximal caries detection: an in vitro study. Dentomaxillofac Radiol 1995; 24: 255 – 259. 34. Wenzel A, Verdonschot EH. Some considerations in the evaluation of diagnostic tests in dentistry. Dentomaxillofac Radiol 1994; 23: 179 – 182. 35. ISO/IEC JTC 1/SC 29/WG 1 IIF-: Information technology-JPEG 2000 image coding system: Core coding system [WG 1 N 1890], September 2000. 36. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing 2004; 13: 600 – 612. 37. Przelaskowski A. Vector quality measure of lossy compressed medical images. Computers Biol Med 2004; 34: 193 – 207. 38. Chen TJ, Chuang KS, Wu J, Chen SC, Hwang IM, Jan ML. A novel image quality index using Moran I statistics. Phys Med Biol 2003; 48: N131 – N137. 39. Wang Z, Bovik AC. A universal image quality index. IEEE Signal Processing Letters 2002; 9: 81 – 84. 40. Kocsis O, Costaridou L, Mandellos G, Lymberopoulos D, Panayiotakis G. Compression assessment based on medical image quality concepts using computer-generated test images. Comput Methods Programs Biomed 2003; 71: 105 – 115.

Dentomaxillofacial Radiology