Image compression technology holds the promise of reducing data archiving needs, thereby decreasing storage costs and allowing for quicker transmission of image data sets and more rapid interpretation. The ratio of allowable compression hinges on the quality of the decompressed image and higher compression ratios may lead to artifacts, such as blurring, which render an image non-diagnostic quality.
A multinational research team from the departments of radiology at the Medical University of Vienna in Austria and Addenbrooke’s Hospital NHS Trust and the University of Cambridge in England, recently conducted a retrospective study comparing a set of compression ratios utilizing 3D JPEG2000 algorithms for emergency department abdominal CT studies. The results of their research were published in this month’s Radiology.
“In the present study, a sophisticated version of this algorithm is used, which extends the compression algorithm from two-dimensional (using only redundancies of one image) to the third dimension and also uses the correlations between adjacent sections for compression,” the authors wrote.
The team evaluated the diagnostic accuracy of their compression ratios on the entire abdominal CT exam, rather than a specific organ system, in terms of diagnostic confidence and diagnostic accuracy. For the study, the researchers reviewed a set of 104 examinations in emergency room patients with acute abdominal complaints, who underwent abdominal CT studies on a Siemens Medical Solutions’ Sensation Cardiac 64 system during a four-month period.
Images were compressed and saved with 3D JPEG2000 at the following compression ratios: 10:1, 12.5:1, and 15:1. After decompression, the images were transferred to a Siemens Leonardo workstation for viewing and interpretation by three board-certified radiologists with six to nine years experience in reading abdominal CT studies.
Images were available uncompressed, as well as in the three compression ratios of the study, and interpretation time was restricted to five minutes per case. The readers scored both their confidence in the diagnoses and the presence of any blurring artifacts on a five-point scale as part of their interpretation. A final reference diagnosis was assigned by a radiologist who did not participate as a reader using the original images, all clinical data, follow-up data and surgical reports.
“The clinically relevant findings (i.e., primary diagnoses) were diagnosed correctly in 91.3 percent (463 of 507), 90.5 percent (459 of 507), 89 percent (451 of 507), and 90.1 percent (457 of 507) of the total number of primary diagnoses at 1:1, 1:10, 1:12.5, and 15:1, respectively,” the researchers reported.
The mean diagnostic confidence scores for the three readers were 4.83 for uncompressed images, 4.87 for 10:1 compression, 4.77 for 12.5:1 compression, and 4.84 for 15:1 compression. The researchers noted that compression artifacts were found almost exclusively in large homogeneous areas of the abdomen, specifically in the liver, the spleen, the retroperitoneal and mesenteric fat.
“The results of our study show that 3D JPEG2000 compression, up to ratios of 15:1, did not impair the diagnostic accuracy of either the primary or ancillary diagnoses for our patient sample,” the researchers wrote. “However, the increasing number of image artifacts also leads to a slightly reduced confidence about negative findings at 10:1 and a moderately reduced confidence about negative findings at 15:1.”
Therefore, the authors recommend “compression ratios between 10:1 and 12.5:1 for CT examinations of the abdomen.”