Lossy 3D JPEG2000 Compression of Abdominal CT Images in Patients with Acute Abdominal Complaints: Effect of Compression Ratio on Diagnostic Confidence and Accuracy

Abstract

To retrospectively assess the effect of lossy three-dimensional (3D) Joint Photographic Experts Group 2000 (JPEG2000) compression on diagnostic confidence and diagnostic accuracy at emergency abdominal computed tomography (CT). In this institutional review board-approved study, transverse images from 104 consecutive multidetector CT examinations (section thickness, 3 mm; reconstruction interval, 2 mm) in patients with acute abdominal complaints were subjected to lossy 3D JPEG2000 compression by using three compression ratios (10:1, 12.5:1, and 15:1, with reference to 384 kB [12 bits] as original image size). Three radiologists independently read the original and compressed CT studies. Patient order and compression ratios were randomized, and readers were blinded to that information. For each organ, the presence of compression artifacts, the diagnosis, the confidence in the diagnosis according to a five-point scale, and the confidence about negative findings were noted. All diagnoses were compared with a standard of reference constructed by an abdominal CT expert by using the original images, surgical reports, and patient follow-up data. Logistic regressions, the Friedman test, and analysis of variance were used for statistical analysis. Primary diagnoses were correct in 91.3% (463 of 507), 90.5% (459 of 507), 89.0% (451 of 507) and 90.1% (457 of 507) of the total number of primary diagnoses at 1:1, 1:10, 1:12.5 and 15:1, respectively. These values did not vary significantly (P = .456) with compression ratios. The radiologists' mean confidence about the primary diagnoses was also almost identical at different compression ratios (4.83, 4.87, 4.77, and 4.84 at 1:1, 1:10, 1:12.5 and 15:1, respectively). However, the radiologists' mean confidence about negative findings in the liver was reduced in 50.3% (157 of 312) of studies at 15:1. Diagnostic accuracy was not impaired at compression ratios up to 15:1. However, because of the significant reduction of the confidence about negative findings at 15:1, compression ratios no higher than 12.5:1 are recommended.