Observer performance in detecting multiple radiographic signals. Prediction and analysis using a generalized ROC approach.

Abstract

The theories of decision processes and signal detection provide a framework for evaluation of observer performance. Some radiological procedures involve a search for multiple similar lesions, e.g., plain radiographic examinations for gallstones or pneumoconiosis. Presuming knowledge of the conventional ROC curve for detection of a single radiographic signal, a model is presented which is used to predict observer performance in an experiment requiring detection of more than one such signal. An experiment tests the validity of this model for detecting radiographically the presence of zero, one, or two low-contrast, 2-mm diameter, Lucite beads. Results confirm the validity of the model and suggest that observer performance in relatively complex detection tasks can be predicted from simpler experiments.