search railabs.duhs.duke.edu.

Papers Published

1. Markey, M.K. and Lo, J.Y. and Vargas-Voracek, R. and Tourassi, G.D. and Floyd, C.E., Jr., Perceptron error surface analysis: a case study in breast cancer diagnosis, Comput. Biol. Med. (UK), vol. 32 no. 2 (2002), pp. 99 - 109 [S0010-4825(01)00035-X] .
   (last updated on 2007/04/15)

   Abstract:
   Perceptrons are typically trained to minimize the mean square error (MSE). In computer-aided diagnosis (CAD), model performance is usually evaluated according to other, more clinically relevant measures. The purpose of this study was to investigate the relationship between MSE and the area (A_z) under the receiver operating characteristic (ROC) curve and the high-sensitivity partial ROC area (_0.90A_z'). A perceptron was used to predict lesion malignancy based on two mammographic findings and the patient's age. For each performance measure, the error surface in weight space was visualized. A comparison of the surfaces indicated that minimizing the MSE tended to maximize A_z, but not _0.90A_z'

   Keywords:
   cancer;error analysis;mammography;medical diagnostic computing;minimisation;perceptrons;performance index;sensitivity analysis;