1 - The Accuracy of Diagnostic Radiology

Abstract

This chapter describes commonly used measures of accuracy, including their inherent advantages and shortcomings, and highlights possible interactions between research design and measures of accuracy within the context of screening mammography and computer-aided detection (CAD) in screening mammography. Sensitivity and specificity are basic measures of the accuracy of diagnostic tests. They also appear to be the most popular indices of diagnostic accuracy in the radiological literature. They are diagnostic accuracy descriptors that do not vary greatly among patient populations, but they can vary greatly among radiologists. Overall accuracy is also a popular index of diagnostic accuracy in the radiological literature, though perhaps somewhat less popular than sensitivity and specificity. The overall accuracy of a diagnostic test is calculated by adding hits and correct rejections and dividing by the total number of patients tested. Signal detection theory estimates diagnostic accuracy by analyzing the receiver (or relative) operating characteristic. Receiver operating characteristic (ROC) curves are obtained by plotting hit rate [hits/(hits + misses)] on the y-axis and false alarm rate [false alarms/(correct rejections + false alarms)] on the x-axis of a two-dimensional graph. Each point on the ROC curve corresponds to a pair of hit and false alarm rates that result from use of a specific cut-off value. The ROC curve enables a direct visual comparison of two or more diagnostic tests at all possible cut-points, and hit and false alarm rates can be easily read from the graph. Currently, the area under the ROC curve (AUC) is the preferred and most popular signal detection theory index of accuracy and the values for AUC can range from 0.0 to 1.0.