Lossy image compression affects CAD performance

CHICAGO - Although it doesn't influence 3D volume rendering, lossy image compression can negatively affect computer-aided detection (CAD) performance, according to research presented at the RSNA 2011 meeting.

In a study evaluating the impact of several irreversible compression ratios on 3D-volume rendered images and CAD algorithms, a research team from the University of Mainz in Germany found no significant difference in perceptible image quality for 3D images. Increasing compression ratios did lead to deterioration in CAD results, however.

"With increasing compression ratios, CAD shows less true-positive and more false-positive results," said Dr. Daniel Pinto dos Santos. He presented the study's findings during a Thursday scientific session.

With the size and number of radiology images on the rise, the use of irreversible compression could optimize image storage space and transmission time, Pinto dos Santos said. Numerous studies regarding the reporting of irreversibly compressed image data have been published, and several radiology societies have issued recommendations for compression ratios that could be employed without loss of diagnostic quality.

However, no data have been published analyzing the effect of irreversible compression on image postprocessing algorithms, he said. As a result, the research team sought to tackle this task by evaluating its effect on 3D volume rendering and CAD algorithms.

The study included 30 CT angiography (CTA) datasets and 30 MDCT datasets of lung exams. The CTA datasets were acquired at 1-mm slice thickness with standard acquisition parameters and received irreversible JPEG 2000 compression at ratios of 8:1, 10:1, and 15:1. An AquariusNet Viewer (TeraRecon) was used for 3D-volume rendering.

The 3D images were compared using both objective and subjective parameters, including visual rating by two experienced readers in a blinded fashion, as well as calculation of peak signal-to-noise (S/N) ratio and high dynamic range-visual difference predictor (HDR-VDP).

In the 30 chest CT exams, acquisition was performed at 1-mm slice thickness using standard parameters. These images also received irreversible JPEG 2000 compression at ratios of 8:1, 10:1, and 15:1, and CAD was performed using an AquariusNet APS Server (TeraRecon). Two experienced readers then analyzed the CAD results in consensus to determine false-positive and false-negative marks.

Although the researchers did find that peak S/N ratio decreased with higher compression ratios, no significant difference was noted for HDR-VDP and visual ratings of the 3D CTA images. However, increasing compression ratio was associated with more false-positive and false-negative CAD results, ranging from 7% more at 8:1 compression to almost 20% more for 15:1.

"Prior to widespread use of irreversible image compression, there should be further studies regarding these algorithms and others as well," he said. "Vendors should [also] test their algorithms with compression."

Page 1 of 775
Next Page