Abstract

This experiment measured human observers' ability to classify vessels with four degrees of stenosis that were rendered on film. The noisy image data simulated a radiologic image. One rendering linearly mapped the image data into optical density. With a second rendering, a threshold of 25% of maximum vessel contrast was applied before the mapping. The observers' classification responses were analyzed by an extension of the standard receiver-operating-characteristic procedure.¹ The statistical efficiency of the observers was 0.55 (0.05 standard error) relative to a computer classification procedure on the same image data and 0.059 (0.023) relative to an optimum calculation. The observers' efficiency was higher with the threshold-rendered images and decreased only slightly as the displayed vessel contrast decreased by a factor of 64. Observers' classification of stenosis may be based primarily on the vessel's edge location, because the display manipulation that degraded gray-scale information (while sharpening edges) increased the observers' efficiency.

© 1990 Optical Society of America