Resolution requirements for digital images in dermatology Andreas Bittorf, PhD, Manige Fartasch, MD, Gerold Schuler, MD, and Thomas L. Diepgen, MD Erlangen, Germany
Background: The digital image has become an important tool in dermatology because of the rapid development of computer hardware, networks, and the World Wide Web. Objective: Our purpose was to examine the resolution requirements for digital images. Methods: Slides of eight selected images showing characteristic lesions were produced in five different resolutions each, ranging from 192 x 128 x 24 (192 pixels x, 128 pixels y, and 24-bit color depth) up to 3072 x 2048 x 24. They were compared side by side by a group of six experienced dermatologists using a standardized questionnaire. , Results: Images at the resolution of 768 × 512 x 24 were perceived as equivalent to higher resolutions, whereas a clear difference was visible between 768 x 512 x 24 and 384 × 256 x 24. The slide projector had a significant influence on the rating of the images. Conclusion: For digital images in dermatology a resolution of 768 x 512 x 24 is suitable to recognize the relevant details of the source image. (J Am Acad Dermatol 1997;37:195- 8.)
With the development of computer hardware and networks the digital image has become an accessible and important tool in dermatology. The World Wide Web, videoconferencing, multimedia mail, and image databases facilitate applications of digital images in a visually oriented discipline. For example, an image database is available through the World Wide Web and used for student education as well as a reference image atlas for physicians. 1'2 "Teledermatology" applications that use digital images 3 have become popular. Finally, an increasing number of CD-ROMs containing dermatologic images have been developed, starting with the CD-Derma Series in 1992. 4 Resolution, compression, and color depth of the images have to be optimized to enable fast access and to offer acceptable image quality. The present study focuses on the resolution aspect.
From the Department of Dermatology, University of Erlangen. Supported by the German Research Net Association (DFN-Verein) and the German Research Ministry (BMBF) under Grant No. TK558-RTB 03.1: 3.5. Accepted for publication Feb. 18, 1997. Reprint requests: Andreas Bittorf, PhD, or T. L. Diepgen, MD, Dermatologische Universit~itsklinik Erlangen, Abteilung Dokumentation und Biostatistik, Hartmannstr. 14, 91052 Erlangen, Germany. Copyright © 1997 by the American Academy of Dermatology, Inc. 0190-9622/97/$5.00 + 0 16/1/81281
When a high resolution is chosen, problems with storage space and required computing power may arise; this implies that access times may be too long. The use of a resolution that is too low may result in a loss of diagnostic information. Thus the critical issue is to choose the right resolution for the intended application. The aim of the study is to address the following questions specifically: (1) What is the relevant difference between various resolutions? (2) What is the optimal compromise between the image quality and its pixel resolution (i.e., storage space)? MATERIAL AND METHODS Material
Eight different types of dermatologic images were selected showing various types of skin lesions: bullae, pigmented lesions (lentigo maligna), papulosquamous lesions (psoriasis vulgaris), pustule (pustulosis palmaris et plantaris), papulopustular lesions (perioral dermatitis), eczematous lesions (allergic contact dermatitis), ulcer, and papule (lichen planus) (Fig. 1). Preparation of the slides The 35 mm photographic transparencies were taken on Fujichrome Sensia 100 slide film, scanned by a commercial provider, and stored on a Kodak Photo compact disk (PCD). The images were read from the PCD in the resolutions of 192 x 128 x 24 (192 pixel x, 128 pixel y, 24-bit color depth [16.8 x 10 6 colors]) 195
196
Journal of the American Academy of Dermatology August 1997
Bittorfet al. . .................. :
~
:::::~','":,"~:::r'"
~i%,
Fig. 1. Images used in the study. From upper left: Bullae, pigmented lesion, papulosquamous lesion, pustules, papulopustules, eczematous lesion, ulcer, and papule.
Ref. Image better
7
1
'5
II
No difference 4
3 2 Ref. Image
worse
1 t92
384
1536
3072
Fig. 2. Results of the evaluation. For high resolutions (1536 and 3072) no difference between them and the reference image was visible, whereas there is a clear quality difference with the low-resolution images. Ratings for the observers are depicted as average over the different images.
(0.07 MB), 384 x 256 × 24 (0.28 MB), 768 x 5 1 2 x 24 (1.13 MB), 1536 x 1024 x 24 (4.5 MB), and 3072 × 2048 x 24 (18 MB) and stored in the Tagged Image File Format (TIFF). Slides of these files were produced with an exposer (Lasergraphics LFR Plus with a Lasergraphics SmartBack camera) at its maximum resolution of 4000 lines per inch. Gamma correction was applied to ensure correct colors. All films were processed at the same time to avoid variations in the chemical development process. As a control for the capabilities of the exposer, we used line drawings containing diagonals and ovals with different line widths at the same resolutions as the clinical images.
Projection As a projection area a matte surface screen (1.4 × 2.1 m) was used for each slide. With the use of the dot
resolution of the human eye at a maximum angle of approximately 50 seconds, 5 the maximum distance at which dots can be resolved for the high-resolution images is 2.82 m. This fact was taken into account when the images were viewed.
Study design: Image quality evaluation Because photographic slides are the standard for dermatologic documentation, we showed the images using slides instead of computer screens. This procedure allowed us to avoid an impact on the rating because physicians are not accustomed to view digital images on a computer screen. Furthermore, there are no computer screens available that can display a 3072 × 2048 pixel image. A resolution of 768 × 512 x 24 can be handled by standard personal computers. Our preliminary tests
Journal of the American Academy of Dermatology Volume 37, Number 2, Part 1
Bittorfet al. 197
6,6] 5,61 5,41 52" 5,0'
4,8' 4,6'
4,4' 4,2 Reference image left
Reference image right
Fig. 3. Comparison of slide projectors. Images displayed with left-hand projector were ranked as better by each physician when compared with that displayed with right-hand projector. Ratings for the observers are depicted as average over all images.
have shown that this resolution is acceptable to depict all necessary details. To evaluate the ability to differentiate between the chosen resolution and higher as well as lower ones, we always showed a reference image with a resolution of 768 x 512 × 24 side by side with another resolution. The sequence of the images and the side at which the reference image was shown were chosen randomly. Each image pair was shown for approximately 20 seconds (longer on request); after each set of eight images, there was a break of 1 minute. Six physicians with different levels of experience were selected and participated in the study: the director of a department of dermatology, three senior dermatologists, and two residents. They were not aware that there was always a probable invisible difference between the pictures and that there was a reference image.
Standardized questionnaire The images were rated by means of a seven-step scale. 1= 2= 3= 4= 5= 6= 7=
Left-hand image much better Left-hand image clearly better Left-hand image slightly better No difference visible Right-hand image slightly better Right-hand image clearly better Right-hand image much better
In addition, it was possible to indicate whether a difference was detected but the better image could not be appreciated. Furthermore, there was a yes/no answer to the question if the quality of the worst image (192 x 128 x 24, 384 x 256 x 24, 768 x 512 x 24) still allowed a clinical diagnosis.
For analysis of the results all ratings were normalized as if the reference image were always on the left side. Thus values higher than 4 mean that the reference image is better and values lower than 4 mean that the comparison image is better. RESULTS Images at the resolution o f 768 × 512 × 24 are superior (average, 5.08 = slightly better) to those with a resolution o f 384 × 256 × 24 and m u c h better (average, 6.71 -- m u c h better) than those with 192 × 128 × 24. Images at a resolution of 1536 × 1024 × 24 and 3072 × 2048 × 24 are comparable to those at 768 × 512 × 24 (average, 4.10, 4.17 = no difference visible (Fig. 2). Use o f the median rather than the average resulted in almost the same values (6.88, 5.13, 4.19, 4.25, respectively). At 192 × 128 x 24 the quality of the image rarely allowed (2.1%) the definition o f a diagnosis, whereas 384 × 256 × 24 was good in more than 50% o f the cases. With 768 × 512 × 24, the quality was acceptable for diagnosis in seven of eight cases. The most problematic image was the papulosquamous one, which was judged acceptable in only 50% of cases. There was no correlation between the diagnostic rating and the detection of a difference between the high-resolution images (1536 × 1024 × 24 and 3072 × 2048 x 24) and the reference image. Depending on the observer the average deviation from the value 4 (no difference visible) ranged from 0.94 to 1.94 and increased with the educational level o f the observer. Deviations from the
198 Bittorfet al. value 4 for the high-resolution images were distributed randomly; sometimes the 3072 x 2048 x 24 image was ranked as worse than the reference image and simultaneously the 1536 × 1024 × 24 image was ranked to be equal or better by the same observer or vice versa. There was no image in which a physician noted that a difference was visible but could not decide which image was the better one. There was no direct correlation between the "deviation of 4" and the "diagnosis OK" rating of the observers. In line drawings there was a clear difference visible even when the 1536 x 1024 and the 3072 × 2048 images were compared when viewed from the distance calculated as described in the "Methods" section. Thus we did not use the sevenstep scale to rate the control images. At 768 x 512, steps were clearly visible in the diagonals, whereas at 1536 x 1024 the image appeared almost like a smooth line; at 3072 × 2048, no steps at all could be detected. Images displayed with the right-hand slide projector were ranked better than the left-hand projector by each physician (Fig. 3). DISCUSSION
Photographic film has the capacity to record anatomic detail at high clarity.6 Thirty-five millimeter photographic film has a resolution of approximately 6000 × 4000 dots, which is much higher than that available on computer screens. To our knowledge, no study in dermatology has previously compared the quality of digital images with the use of photographic slides. Perednia, Gaines, and Butruille 7 have compared digital images with slides, stating that there was no significant difference in the informativeness of digital images (574 x 489 × 24) viewed on a computer screen than projected original slides. However, the comparison of images side by side has not been performed. Furthermore, different media (slide projection and computer monitor) were used for display in this and other studies. 6,8 With this study design the question can be addressed whether the quality of the displayed image is acceptable for diagnosis under the given conditions. However, the result is influenced by many factors of which resolution is only one (e.g., the quality and calibration of the, computer screen and the type of illumination). By projecting the images side by side using the same medium (slides), we could focus on the resolution aspect.
Journal of the American Academy of Dermatology August 1997
The effect of different projection quality was eliminated by the random distribution of the slides. We chose a small number of images to avoid an overly long duration of the trial and thus a loss of concentration by the observers; this was also the reason to take a break after each sequence of eight images. We decided to select characteristic primary lesions to study the influence of different resolutions. We used the seven-step scale instead of a three-step scale to obtain better differentiation among the lower quality images. Analyses confirmed our previous assumption that the resolution of our choice (768 x 512 x 24) appears to be acceptable for good images. The quality of the photographic slides has not been directly addressed by any study because all relied on high-quality images taken by a professional photographer. The dot or pixel resolution is only one factor influencing the diagnostic rating of an image. Other important ones are the spatial resolution (i.e., the number of dots per centimeter shown), the distance of the observer from the image, and the color accuracy. The size of the original lesion should be taken into account because there is a great difference if a single pustule is shown with 768 × 512 dots or if it is a whole-body view. Standardized color display is a complicated problem. However, in digital format it can be handled better than with slides. REFERENCES
1. BittorfA, Krejci-Papa NC, Diepgen TL. Development of a dermatological image arias with worldwide access for the continuing education of physicians. J Telemed Telecare 1995;1:45-53. 2. Bittorf A, Krejci-Papa NC, Diepgen TL. Storage and retrieval of digital images in dermatology. Skin Res Tech 1995;1:192-9. 3. Perednia DA, Brown NA. Teledermatology: one application of telemedicine. Bull Med Library Assoc 1995;83:42-7. 4. Dockx PFE. General dermatology, CD-Derma Series. 1992. 5. Krestel E. Bildgebende Systeme ftir die medizinische Diagnostik. Berlin: Siemens; 1988. 6. Schosser RH, Sneiderman CA, Pearson TG. How dermatologists perceive CRT displays and silver halide prints of transparency-based images: a comparison study. J Biol Photogr 1994;62:135-7. 7. Perednia DA, Gaines JA, Butruille TW. Comparison of the clinical informativeness of photographs and digital imaging media with multiple-choice receiver operating characteristic analysis. Arch Dermatol 1995; 131:292-7. 8. Sneiderman CA, Cookson JP, Hood AF. Usage of computer graphic images in teaching dermatology. Comput Med Imaging Graphics 1992;16:151-2.