Feature selection in cell image analysis: use of the ROC curve
Sherwood, E.M.; Bartels, P.H.; Wied, G.L.
Acta Cytologica 20(3): 255-261
1976
ISSN/ISBN: 0001-5547 PMID: 775870 Document Number: 105673
The receiver operating characteristic (ROC curve) is widely used in signal detection theory to measure how well a system is capable of distinguishing 2 different input types, for example signal or noise. This paper describes the use of ROC curves for the determination of which features from a large number of different computed features render a minimum error classification for cell images from 2 types of cells. A brief introduction to the concepts of signal detection is given first. This is followed by a discussion of the application to cell image data, and practical examples.