Effect of image compression on telepathology: A randomized clinical trial

Archives of Pathology & Laboratory Medicine, Nov 2000 by Marcelo, Alvin, Fontelo, Paul, Farolan, Miguel, Cualing, Hernani

* Context.-For practitioners deploying store-and-forward telepathology systems, optimization methods such as image compression need to be studied.

Objective.-To determine if Joint Photographic Expert Group (JPG or JPEG) compression, a lossy image compression algorithm, negatively affects the accuracy of diagnosis in telepathology.

Design.-Double-blind, randomized, controlled trial.

Setting.-University-based pathology departments.

Participants.-Resident and staff pathologists at the University of Illinois, Chicago, and University of Cincinnati, Cincinnati, Ohio.

Intervention.-Compression of raw images using the JPEG algorithm.

Main Outcome Measures.-Image acceptability, accuracy of diagnosis, confidence level of pathologist, image quality.

Results.-There was no statistically significant difference in the diagnostic accuracy between noncompressed (bit map) and compressed OPG) images. There were also no differences in the acceptability, confidence level, and perception of image quality. Additionally, rater experience did not significantly correlate with degree of accuracy.

Conclusions.-For providers practicing telepathology, JPG image compression does not negatively affect the accuracy and confidence level of diagnosis. The acceptability and quality of images were also not affected.

(Arch Pathol Lab Med. 2000;124:1653-1656)

Medical specialties that rely on images to formulate a diagnosis lend themselves to the store-and-forward method of telemedicine. With this method, images are captured and then forwarded, often through the Internet, to a remote expert for asynchronous review at a later time. Pathology (telepathology), radiology (teleradiology), and dermatology (teledermatology) are among the most advanced areas of telemedicine because of their image-intensive nature and minimal requirement for patient interaction.1-3

Telepathology in particular requires the evaluation of microscopic images to formulate a diagnosis. Store-and-- forward telepathology allows a pathologist in a remote location to digitize images of a challenging case for second opinion consultation. These images are usually transmitted through the Internet, which is now widely used as an exchange medium for scientific data. Efficient methods of image transmission, especially by compression, are of great interest to the scientific community, since high-quality digital images may attain sizes of 1 megabyte or greater. While larger files generally produce better images, they also take longer to send through the Internet. Depending on the type of Internet connection and the amount of "traffic," a 1-megabyte file may take 5 minutes or more to download. Its compressed version, on the other hand, may be received in less than 30 seconds.

Joint Photographic Experts Group (JPG) is a widely accepted image compression algorithm. Any Internet-connected computer today will have a web browser, such as Netscape or Internet Explorer, which is capable of displaying JPG files.4-7 This algorithm accomplishes compression by exploiting known limitations of the human eye, particularly its inability to detect minute color and shades-of-gray details.8 During the compression process, these small details are removed without noticeable difference if viewed with the naked eye. The final compressed image, therefore, will contain less data than the original. It is this loss of data, its possible effect on the quality of the image, and ultimately its effect on diagnosis that generates a cause for concern.

Many studies have been published on telepathology,9-14 but only a few researchers15,16 have actually compared noncompressed with compressed images using a structured study design. The aim of this study was not to compare the diagnostic accuracy of telepathology with that of glass slide diagnosis, but rather to determine whether the loss of data in JPG compression adversely affects the quality of images and the accuracy of diagnosis.

METHODS

Ten previously diagnosed cases were chosen from the teaching files of the Department of Pathology, University of Illinois University Hospital Chicago (UIC). Six representative snapshots from each case (magnifications X2.5, X10, and X40) were captured using a Polaroid DMC 1 digital camera (Polaroid Corp, Cambridge, Mass) mounted on a Leica DMLB (Leica Microsystems, Heidelberg, Germany). All initial images were saved in the Windows bit-map (BMP), noncompressed format (average size ~1000 kilobytes, at 667 x 500 pixel resolution). Bit-map images were then compressed using Adobe PhotoShop 5.0 (Adobe Systems, San Jose, Calif). The resulting compressed files were about 100 kilobytes at 90% compression. Images were archived on Macintosh 8100 computer (Apple Computers, Cupertino, Calif) and served on the World Wide Web using Quid Pro Quo 2.1.2 (Social Engineering, Berkeley, Calif) on a T-3 (45 megabits/s) line. The Web site was password protected and was accessible only after proper authentication. File extensions were masked server-side using randomly selected letters.