TABLE OF CONTENTS Ulrich Scheipers
Resonant Medical Inc. 2050 Bleury, Montreal, Quebec, H3A 2 J5, Canada ulrlch@schelpers.org
The application of receiver operating characteristic (ROC) methods for computer-aided detection systems is reviewed. A statistical framework for ROC analysis is presented and different methods of evaluating the performance of computer-aided detection systems are derived.
Most detection systems that are used today are dependent on a separation threshold, which, in many cases, can be chosen by the operator without restrictions. The separation threshold separates the range of output values of the detection system into different target groups, thus, conducting the actual detection process.
In the first part of this chapter, threshold specific performance measures, e.g. sensitivity and specificity, are presented. In the second part, a threshold-independent performance measure, the area under the ROC curve, is reviewed. Only the use of separation threshold-independent performance measures provides detection results which are overall representative for computer-aided detection systems.
This chapter provides the interested scientist with all information needed to conduct ROC analysis and to integrate algorithms performing ROC analysis into detection systems while understanding the basic principles of detection.
Many works on computer-aided detection have been published during the last years. However, no single measure is used to express the detection performances of the systems proposed but many different measures are used instead. This multitude of different measures makes comparing different systems with each other rather complicated and sometimes impossible.
Furthermore, the lack of a representative performance measure is striking in many works. As...
