Lossless and Lossy: Study Supports Compression of Mammograms
Compression of images is not an alien concept in radiology. From radiography to CT, large image files transmitted between physicians’ offices or stored over the long term often undergo some kind of compression. Mammography—one of the last modalities to go digital—remains behind the curve, however. Can mammographic images be compressed while maintaining enough fidelity to be clinically useful?
Stamatia Destounis, MDThat’s the question that Stamatia Destounis, MD, a radiologist at Elizabeth Wende Breast Care LLC, Rochester, NY, hoped to answer with a recent study assessing the readability of mammograms compressed using an access-over-network (AON) wavelet compression engine developed by FUJIFILM Medical Systems USA Inc, Stamford, Conn. The algorithm is already used as part of the company’s Synapse PACS in order to compress other types of radiologic images, including musculoskeletal and CT images, for transmission between computers and for both short- and long-term storage.
Stamatia Destounis, MD
“Compression has been slow to make it to mammography because of the resolution of the images and the fear that once you compress them, you’ll lose some vital information.” —Stamatia Destounis, MD, Elizabeth Wende Breast CareDestounis continues, “If you’re looking at a bone film, you’re looking specifically at a fracture or some other irregularity in the bone. You’re not looking for a small abnormality like cancer. Everything is so subtle with mammography. If you compress these subtle changes in glandular tissue, that could interfere with your reading.” While lossless compression has become acceptable in mammography, lossy compression that in any way sacrifices data has remained the purview of other modalities. “Perfection is a big thing in breast imaging,” Destounis notes. “There’s a lot of governance that goes along with mammography.” Destounis adds that because mammography was one of the last modalities to become digital, most facilities haven’t yet had to grapple with long-term storage of breast images. “We’ve been living in a screen/film world,” she says, “but in the past few years, as more and more facilities move into the digital sphere, we’re beginning to see these issues.” With file sizes routinely exceeding 60MB per study, the cost associated with storing mammographic images could be yet another barrier to widespread adoption of full-field digital mammography. “These files can be quite large, and it’s impossible to store them forever in their original form,” Destounis says. “Some compression has to occur, or it slows down your PACS, and you have to invest so much money in your archiving and disaster recovery.” Large file sizes also lead to difficulties in image transmission that are not typically encountered when working with other modalities. “If you had a physician who wanted to look up a patient’s mammograms, but had only a laptop, he or she wouldn’t be able to view the images,” Destounis. “They’re so large that if you tried to transmit them, it could crash the system.” A Blinded Study To ascertain whether applying lossy compression to breast images compromised their diagnostic usefulness, Destounis and three colleagues at Elizabeth Wende—Posy Seifert, DO; Patricia Somerville, MD; and Philip Murphy, MD—set up a blinded study in which 16 experienced radiologists compared 100 digital mammograms in three compression modes. “We looked at the original images and then looked at images that were compressed at 20:1 and 40:1,” Destounis explains. “All images were in pairs, so you could be comparing lossless with 20:1, lossless with 40:1, or 20:1 with 40:1.” The radiologists were instructed to compare each pair of images and decide which was more compressed, with the expectation that at a certain compression level, there would be a sufficient loss of fidelity for the image to be harder to read. “When you take a phantom in a lab, or any image that may have calcifications or masses, and you compress and compress, there’s a point where things start to lose their crispness,” Destounis says. “It was thought that when the images were irreversibly (lossily) compressed, you really couldn’t see as well.” Readers recorded whether the quality of the images was sufficient for comparison with next-year mammograms and which image was more compressed. If the radiologist decided that both images were sufficient for comparison with next-year mammograms, he or she was asked to record that preference on a 7-point Likert scale ranging from +3 (meaning that the image on the right was much better) to –3 (the image on the left was much better). What the radiologists found was surprising. Time after time, they selected the wrong image as the more compressed of the two. As for the question of perceived equivalence for comparison with next-year studies, it was achieved with greater than 95% confidence. “We literally could not tell the difference between any of the images,” Destounis says. “Even when we thought we were right, we were wrong.” Readers were also asked to evaluate the quality of the images according to a variety of criteria. They indicated that the area at or near the skin line was displayed at least as well by the more compressed image as by the less compressed image in greater than 90% of the image pairings. Results are still being compiled and will be released shortly. Compression Options FUJIFILM’s AON engine works by compressing the image at multiple ratios, depending on its intended use, creating up to three versions of the mammogram. Compression ratios are configurable by image type, modality manufacturer, body part, and facility preferences. Next, the PACS user selects the desired version of the image, based on clinical requirements and the network’s bandwidth. If the user requests a compressed image at a PACS workstation, the file is delivered in compressed form and decompressed for display by the workstation’s software. Images can undergo lossless or lossy compression at ratios of up to 100:1, but the engine always writes a lossless DICOM image version as the original, in compliance with the Mammography Quality Standards Act (MQSA). While a lossily compressed file can differ significantly from the original at the bit level, the differences are considered indistinguishable to the human eye. Flaws caused by lossy compression that can be perceived are compression artifacts, and in building the AON compression engine, Fuji’s goal was to have no artifacts. “With three modes of storage, you can store the lossless image for a short period of time so that it’s around if you need to refer back to it,” Destounis explains. “For long-term storage, you have the highly compressed (lossy) image that doesn’t take up as much space.” Destounis’ study makes reference to a 2005 Radiology study1 that garnered equally convincing results, but used digitized images, as opposed to digital ones. In that study, five experienced radiologists read irreversibly compressed mammograms at 40:1 and 80:1 ratios. Readers were asked to identify calcifications and masses on 112 original and compressed images. Like Destounis’ research, the study showed that irreversible compression ratios—even at 80:1—had no impact on the rate of detection. “Compression has been fought for in other areas of radiology, but it seems to be a big problem for breast imaging,” Destounis notes. “It’s a highly litigious area. We don’t want to lose any information or data. This study was undertaken to make sure we’re not doing a disservice to the patient.” Aside from the obvious advantages to be gained from compression, such as cheaper storage, faster processing speeds, and more reliable transmission, Destounis observes that mammography is uniquely suited to the technique because it is the only modality for which comparison with prior studies is legally required. “You want your findings to be easily accessible,” she says. “You want all the clinicians involved to be able to access your images. You want it to be fast, seamless, and cheap, and you have to be able to store and retrieve these images when the patient comes back. You have to be able to prove that you did the study.” Ultimately, Destounis says, compression will have to be accepted for full-field mammography to be economically viable. “It’s very important that the word gets out,” she says. “We need to go back to the FDA with this study and show that these experienced radiologists couldn’t tell the difference, but this change is inevitable, whether it happens now or six months from now. Everyone is eventually going to have to fall in line.”