Interobserver variability in the measurement of internal carotid stenosis

To assess interobserver variability in the measurement of carotid stenoses from digital subtraction angiograms displayed in different ways (nonmagnified or magnified, white or black arteries); and to compare human readers with computer-generated densitometric measurements of vessel stenosis.

Digital subtraction angiograms of 20 proximal internal carotid artery stenoses were laser printed in the following ways: (1) Nonmagnified white artery on a black background; (2) Magnified white artery on a black background; (3) Nonmagnified black artery on a white background; (4) Magnified black artery on a white background. This resulted in 80 images of internal carotid artery stenoses. These stenoses were independently measured by 4 radiologists using the North American Symptomatic Carotid Endarterectomy Trial method. A computer-generated densitometric measurement of the black nonmagnified images was also obtained.

The most reliable stenosis measurements were obtained from the nonmagnified black and white artery images. The interobserver variability in the measurement of internal carotid stenoses using these images was quite small. Variability increased with the use of magnification. The computer-generated stenosis measurements were consistently much higher than those of the radiologists.

There was significant variability in measurements made from magnified images and between human readers and computer-generated measurements. This has great clinical significance. Readers of digital angiographic images must determine the most reliable, reproducible images generated by their equipment, as these measurements significantly affect treatment of patients with symptomatic internal carotid artery stenosis.