Study: Image compression ratios of 4:1, 8:1 indistinguishable from original
A retrospective study of CT abdominal studies conducted in the department of radiology at University Hospital Vienna in Austria and Amsterdam Academic Medical Center in the Netherlands found that images compressed using 3D JPEG 2000 at 4:1 and 8:1 ratios were visually indistinguishable and essentially indistinguishable, respectively, from the original images.

Clinicians at the institutions sought to determine the maximum ratio at which compressed images are indistinguishable from the original using a 3D wavelet compression algorithm, 3D JPEG 2000. The impetus for the study was to reduce the size of image study storage (decreasing archive costs) and to speed the transmission time of diagnostic-quality images (potentially providing better patient care for teleradiology of emergency department CT exams). Results of the analysis were published in the November issue of the journal Radiology.

A set of 60 abdominal CT scans were compiled from different patients with acute abdominal complaints over an eight -month period (June 2005-January 2006).

“By using the most common diagnoses of acute abdominal pain in the literature and the prevalence of the most common emergency diagnoses at our institution, a group of five patients for each of the 12 diagnoses with a different manifestation or stage was selected: acute appendicitis, acute cholecystitis, small bowel obstruction, acute pancreatitis, renal colic, cancer, diverticulitis, abscess, aortic aneurysm, intestinal ischemia, pyelonephritis, and nonspecified or undetermined causes without any abnormal findings at CT,” the authors wrote.

All images were independently read by three board-certified radiologists who had different levels of experience, with a minimum of 6 years to a maximum of 18 years of experience in reading abdominal CT images. An evaluation was conducted in pairs, consisting of compressed images against original images.

“Four pairs consisted of one compressed image (with a ratio of 4:1, 8:1, 12:1, or 16:1) and one original image; the fifth pair consisted of two identical original images (1:1) and served as the control pair,” the authors wrote. “A total of 600 image pairs were analyzed, 300 pairs with a 3-mm section thickness and 300 pairs with a 6-mm section thickness.”

(a) CT scans of 64-year-old patient with liver metastases. Upper row derived from 3-mm-thick; lower row, from 6-mm-thick sections. Visually perceivable blurring artifacts (arrows) appear at 12:1 and 16:1 ratios. (b) CT scans of 29-year-old patient with acute appendicitis. Periappendiceal fat stranding (arrows) appears indistinct at 16:1 ratio. (c) CT scans of 55-year-old patient with acute pancreatitis. Note increased blurring of mesenteric fat (arrows) and loss of sharpness of septation (arrowheads) in subcutaneous fat tissue at 12:1 and 16:1 ratios. Image and caption courtesy of the Radiologic Society of North America (RSNA).  
The researchers found that the three radiologists rated 4:1 compressed images as having original image quality (a 1:1 compression ratio) 100, 80, and 98 percent of the time for 3-mm slice thickness studies and 98, 95, and 100 percent of the time for 6-mm slice studies.

At an 8:1 compression ratio, the readers found that either the images were almost identical in appearance to the original or that there were minimal artifacts between the two images, allowing recognition of the original. Only 3 percent of the 3-mm cases, and 2 percent of the 6-mm exams, were rated as having considerable difference in image quality between the two images.

However, 12:1 compression ratios were rated as having considerable differences from original image quality as much as two-thirds of the time by one radiologist; while 16:1 compression ratios were rated by all three radiologists as having considerable differences from original images in almost all instances.

“On the basis of the hypothesis that visually identical images will lead to identical diagnoses, we concluded that it is acceptable to use a compression ratio of 4:1 for primary diagnosis in routine radiologic practice,” the authors wrote.

“We further infer that images compressed at a ratio of 8:1 are very close to being visually lossless,” they noted. “As discrimination between original and compressed images was not possible and differences in image quality were rated as minimal, we also considered this compression ratio as acceptable for primary diagnosis.”