Use of a computer-assisted detection (CAD) algorithm to identify coronary artery stenosis on coronary CT angiography (CTA) resulted in relatively high sensitivity and negative predictive value (NPV) in a recent study from South Korea.
However, the algorithm did tend to overestimate the degree of stenosis and misinterpret nonobstructive lesions as significant coronary artery disease, making the tool better suited as a “second reader” to help exclude significant stenosis in patients with acute chest pain presenting to an emergency department (ED).
“A high NPV is especially useful in the ED setting,” wrote Ji Hye Min, MD, of Sungkyunkwan University School of Medicine in Seoul, and colleagues. “As shown in the previous studies, the capability of the CAD algorithm to exclude significant coronary artery stenosis in patients with acute chest pain would be most useful in populations with low prevalence of coronary artery disease. Accordingly, use of the CAD algorithm may facilitate prompt discharge of patients who have normal or nearly normal coronary CTA status, particularly when an expert opinion is not available and ED physicians are pressed for time.”
The study was published in the April issue of the American Journal of Roentgenology.
Findings were based on an investigation of 128 consecutive patients with acute chest pain who underwent 128-slice dual-source coronary CTA and invasive coronary angiography in the ED. Coronary CTA data were analyzed with a customized CAD algorithm which automatically segments the coronary artery tree, labels the major coronary arteries and detects stenotic lesions without human interaction.
After excluding 25 patients for data errors or a history of stents and bypass surgery, Min and colleagues found that the CAD algorithm yielded 100 percent sensitivity and 100 percent NPV in per-patient analysis, along with 90 percent sensitivity and 95.7 percent NPV in per-vessel analysis. A relatively high frequency of false-positives, however, resulted in a specificity of 23.1 percent and 68.1 percent positive predictive value in per-patient analysis using the CAD algorithm.
For comparison, human interpretation of the patients’ coronary CTA data yielded 98.4 percent sensitivity, 79.5 percent specificity and 99 percent NPV for diagnosing significant stenosis on a per-patient basis.
Min and colleagues noted that the high number of false-positive results “serves as a reminder that interpretation by an expert human radiologist will still be necessary to validate the CAD algorithm for clinical application.”