American College of Cardiology/ European Society of Cardiology international study of angiographic data compression phase I
The effects of lossy data compression on recognition of diagnostic features in digital coronary angiography
Richard A. Kerensky, MD, FACC ,
Jack T. Cusma, PhD ,
Paul Kubilis, MS,
R.üdiger Simon, MD, FACC||,
Thomas M. Bashore, MD ,
John W. Hirshfeld, Jr., MD, FACC¶,
David R. Holmes, Jr., MD, FACC,
Carl J. Pepine, MD, FACC and
Steven E. Nissen, MD, FACC*
* Cleveland Clinic Foundation, Cleveland, Ohio, USA
Duke University, Durham, North Carolina, USA
Mayo Clinic, Rochester, Minnesota, USA
University of Florida, Gainesville, Florida, USA
|| University of Kiel, Kiel, Germany
¶ University of Pennsylvania, Philadelphia, Pennsylvania, USA
View larger version (39K):
[in a new window]
|
Figure 1 Representative data form completed for each angiographic sequence. In this example a stent was seen in the circumflex artery and calcification in the left anterior descending (LAD) artery. The Coronary Artery Surgery Study (CASS) location for stenosis estimate was the proximal LAD, which was estimated at 70% diameter stenosis. Image quality was acceptable, and confidence in interpretation was 8 on a 0–10 scale.
|
|
View larger version (13K):
[in a new window]
|
Figure 2 Mean sensitivity and 95% confidence bounds for detection of any of the four diagnostic features at the various CRs. There was no decrease in sensitivity at ratios of 6:1 and 10:1. Sensitivity decreased at a CR of 16:1 compared with no compression (*p = 0.004).
|
|
View larger version (16K):
[in a new window]
|
Figure 3 Mean sensitivity for detecting each of the individual diagnostic features at the various CRs. Sensitivity for detecting each of the features varied widely in uncompressed images. The sensitivity for the detection of calcium decreased significantly at a CR of 16:1 compared with uncompressed images. There were trends toward decreased sensitivity with increasing CR for dissection and stent, but these trends did not reach statistical significance.
|
|
View larger version (20K):
[in a new window]
|
Figure 4 Mean sensitivity and 95% confidence bounds for each of the diagnostic features in uncompressed images and at each of the three CRs. a) complex stenosis or filling defect (*p = 0.012); b) dissection; c) stent; d) calcification (*p < 0.001).
|
|
View larger version (12K):
[in a new window]
|
Figure 5 Specificity for detection of dissection in uncompressed images and at each of the CRs. There was no significant decrease in specificity with image compression at these ratios.
|
|
View larger version (26K):
[in a new window]
|
Figure 6 Box plot showing the median stenosis severity for the consensus panel viewing uncompressed images (left), and observers viewing the uncompressed and compressed images. There was no effect of compression on stenosis severity estimation, comparing reviewers with the consensus panel.
|
|
View larger version (29K):
[in a new window]
|
Figure 7 Variability in observer estimates of percentage diameter stenosis depending on the severity of the stenosis as determined by the consensus panel. Greater variability was seen in the less severe stenoses, but increasing CRs had no effect on the pattern or degree of variability.
|
|
View larger version (13K):
[in a new window]
|
Figure 8 Percent of reviewers finding image quality acceptable with no compression and at each of the compression ratios. At CRs of 10:1 and 16:1, there was a statistically significant decrease in the percent of reviewers who rated images as acceptable (p < 0.001).
|
|
View larger version (12K):
[in a new window]
|
Figure 9 Mean confidence score and 95% confidence bounds for no compression and compressed images. Confidence score decreased significantly at CRs of 10:1 and 16:1 relative to no compression (p = 0.017 and p < 0.001, respectively).
|
|
|
