Vessel analysis programs fall short in carotid measurements

Semi-automated vessel analysis software offered by five vendors tends to underestimate the degree of carotid stenosis, particularly at the higher levels that indicate a need for corrective action, according to a study presented at the RSNA meeting Sunday.

Among the findings of a research team led by Dr. William Boonn: Underestimation of stenosis greater than 70% ranged from 12% to 50%.

Boonn, an imaging informatics fellow at VA Maryland Health Care System when the study was conducted, is now at the University of Pennsylvania.

Using semi-automated vessel analysis software led to significant variability and underestimation of carotid stenosis, the study concluded. These results, along with analysis failures, worsen as the percentage of stenosis increases.

Precise data for individual workstations and software packages were not disclosed during the presentation.

The study analyzed data from 23 carotid arteries from 12 patients undergoing carotid CT angiography using multislice CT. One artery was excluded as a result of postsurgical changes. The software measurements were compared with manual measurement of all 23 carotid arteries performed in consensus with a standard procedure described in recent literature and based on criteria in the North American Symptomatic Carotid Endarterectomy Trial (NASCET).

Users of semi-automated vessel analysis systems must select the vessel to be segmented and the segment to be analyzed, Boonn said. Automated results from six commercially available systems and packages from five vendors were compared with the manual measurements.

No variation in the manual and automated measurements appeared among the eight normals in the study group, Boonn said. But as the degree of stenosis increased, the variability increased as well, mostly in the form of underestimation.

This finding is critical, because the NASCET has shown that patients with high-grade stenosis can benefit from carotid endarterectomy.

Even with the underestimation of stenosis, vessel analysis systems offer some benefits, Boonn said. Segmentation can improve visualization, and curved multiplanar reformatting can help visualize tortuous vessels.

But the study's results suggest that vendors must improve the consistency and accuracy of automated stenosis measurements, according to Boonn.

"We advocate the use of publicly available standardized data sets that can be used by vendors to validate their measurements," he said.