Differential diagnosis in Dermatology is based on the interpretation of visual information in the form of clinical and histopathological images. Traditionally, reference images needed to be retrieved from textbooks and/or appropriate journals. To overcome the inherent limitations of those storage media with respect to the number of images stored, display, and search parameters available, we developed an online database of electronically stored dermatologic images. In order to determine the optimal relation between image quality, file size and cost-effectiveness we investigated various compression and storage methods. In order to provide convenient access we provided for multiparameter search capability and world wide web access.
The internet is becoming the second most important source for professional information, rivaling textbooks and journals.(1)(2) The advantages of digital over printed information include:
Dermatology will particularly benefit from online-information distribution, since images, more than text, are expensive to
print, and atlas collections of clinical and histopathological images are usually confined to medical libraries. A combined
display of clinical and histological images for each disease , as planned for the next stage of our data base, should bring together both "worlds" in a unique and educational matter.
To allow international access of the databases we chose the WWW as interface because it is widely available, and its browsers will soon become a standard application.(3)
Clinical images were taken with a 35 mm camera and stored as 24*36 mm photographic slides. Images were digitized via two methods:
We used 486-66DX2 PCs with 8 or 16MB RAM equipped with a ATI-Mach32 Ultra Pro 32bit graphics card with 2MB VRAM.
The lossless compression technique, Lempel-Ziv-Welsch algorithm (LZW), (4) and the lossy compression algorithm of the Joint Photographic Experts Group (5) (JPEG) were examined with different resolutions, quality factors and images.
The image database was designed using the Entity-Relationship approach. The Prototype was implemented with operating system MS-Windows, programming language Visual Basic and the database access was facilitated via the Open Database Connectivity (ODBC)-Interface.
A static access structure consisting of interconnected Hyper-Text-Mark-Up-Language (HTML) files was generated by a Visual Basic script. For remote access these files were stored on the WWW-Server of the regional computing center (Regionales Rechenzentrum Erlangen; RRZE). In addition a query interface and dynamic lecture scripts were implemented using HTML-Forms together with the Common-Gateway-Interface (CGI) and a PERL script. WWW browsers Mosaic (6) and Netscape (7) were used.
The images on a Photo Compact Disk (PCD) proved to be superior to those obtained via slide scanner with respect to brightness, and color accuracy. This is presumably due to the 4096 times higher colour resolution of the commercial scanner which allows accurate colors even in very dark and very bright areas.
Different lossless compression techniques resulted in a compression factor of 1.1 up to 3.5 with dermatologic images. With the most common technique, the Lempel-Ziv-Welsch algorithm (LZW) (4) is implemented by the Aldus Tagged Image File Format (TIFF) achieving a compression factor less than 1.5.
We examined the "lossy" compression JPEG (5) (8) attempting to correlate necessary image quality with the compression factor. At quality factor 60 the compression factor ranged from 18 to 75 depending on the content of the image. We decided to use a quality factor of 60 corresponding to a compression of about 1:35 for clinical images (n=200) and 1:40 for histological images (n=33) at a resolution of 768*512*24. Some images were cropped so that an average file size of 30KB resulted. With this approach hardly any differences between the original and the compressed image could be detected in > 95% of the images
Resolution had a strong impact on the compression factor achieved at a given quality level. At a given quality factor 60 we achieved about 1:41 at 1536*1024*24 (n=25) and about 1:35 at resolution 768*512*24.
Decompression of a JPEG image on a 486DX266 took about 4 seconds for a 768*512*24 image and approximately 16 seconds for a 1536*1024*24 image both compressed at a quality factor of 60.
A prototype of the database interface was programmed in Visual Basic [10,11]. Two masks were developed: an input mask for
the database-operator and an query-masks for the user
.
For input of descriptive text scroll boxes, combo boxes, radio buttons, and push buttons were used for user convenience. The combo boxes were filled with all acceptable values, so that only the defined keys could be entered in the database.
The query interface was constructed similar to the input interface. Only the available values of the database are offered in the combo boxes. This helps the user to get a first impression the image resources.
We generated Hypertext-Mark-Up-Language (HTML) documents with Visual Basic using the data provided by the database.(10) WWW-access
of the database grew exponentially from 9/94 until 3/95 and seems to plateau at approximately 35,000 to 40,000 accesses per
month. (Up to date access statistics.)
The PCD proved to be an inexpensive method to archive images. Neither colors nor details will change over the life span of the disc, which is currently assumed to be longer than thirty years. The maximum available resolution of 2048*3072*24 will be satisfactory also for future applications.
While the PCD is superior in image quality, a slide scanner should be the digitizing method of choice, if only occasionally images need to be digitized.
A third alternative is digital photography, which is developing very quickly. Only two years ago hardly any applications for digital photography existed. Prices have dropped and the quality has improved. For colored images of non-still subjects (e.g. clinical), one can obtain a resolution up to 2000*3000*36 (Kodak DCS 460c). For still (e.g. histological) images resolutions up to 6250*8450*36 (Arca Swiss SC 1) are offered. The quality of images taken with digital cameras sometimes exceeds conventional photographs.(9) With manipulation techniques, e.g. changes in one color band to equalize the effect of different lighting conditions, the display of the image with original colors is possible. We predict that digital photography will be very important in dermatologic imaging in a few years.
In order to appreciate the effect of compression it is useful to consider the following numbers: An image stored at 768*512*24 needs 1,2 MB of storage space; compression at a factor of 25 results in 48 KB per image. One PC hard-disc can store up to 2 GB today, that is more than 40,000 images, the size of the whole Erlangen archive. The compression factor which is achieved depends not only on the quality factor but also on the resolution of the original image. With a higher resolution the same quality leads to a higher compression, e.g. at a quality factor of 40, compression of a image with 768*512*24 leads to a factor of 25, with resolution of 1536*1024*24 one gets 29. This is because additional pixels at a higher resolution are similar in colour to the pixels nearby.
The compression factor of an image with a given quality factor depends also on the content. With clinical pictures we had a variation of about 3 - the image which was compressed the best used only one third of the memory space when compared to the worst image.(11)
For wide area networks, compression at a high factor is a necessary prerequisite for the on-line usage of images. Response times for a picture of more than 10 seconds would only be acceptable in special cases.
In order to get acquainted with the Erlangen Online Image Atlas we have set up a guided tour. It demonstrates how a user can find the images he is looking for.
(1) Estrada S: Connecting to the Internet. O'Reilly & Assoc., Sebastopl, 1993
(2) Krol E: The Whole Internet. O'Reilly & Assoc., Sebastopol, 1994
(3) FAQ-World Wide Web:http://sunsite.unc.edu/boutell/faq/www_faq.html. (1995)
(4) FAQ-compression:http://www.cis.ohiostate.edu/hypertext/faq/usenet/compressio n -faq/top.html. (1995)
(5) ISO 92:ISO/IEC, Information Technology - Coded Representation of Picture and Audio Information - Digital Compression and Coding of Continuous-tone Still Images, International Standard 10918 (1992)
(6) Mosaic 2.0b4 [computer program]:ftp.ncsa.uiuc.edu in the subdirectories of /Web/Mosaic, depending on hardware and operating system. (1995)
(7) Netscape 1.1N [computer program]:all major systems available at ftp.mcom.com (1995)
(8) Sneiderman CA, Schosser R, Pearson TG: A Comparison of JPEG and FIF Compression of Color Medical Images for Dermatology. Computerized Medical Imaging and Graphics 18: (5) 339-42, 1994
(9) Knapp M: Digitale Photographie. Vogel, Würzburg, 1994
(10) A. Bittorf, N.C. Krejci-Papa, T.L. Diepgen: Development of a dermatological image atlas with worldwide access for the continuing education of physicians. Jornal of Telemedicine and Telecare 1:(1), 45-53, 1995
(11) A. Bittorf, N.C. Krejci-Papa, T.L. Diepgen: Storage and Retrieval of Digital Images in Dermatology. Skin Research and Technology 1, 1995 (in press)
All contents copyright (C), 1995. Dermatology Online Journal University of California Davis