A comparison of maximum entropy estimation and multivariate logistic regression in the prediction of axillary lymph node metastasis in early breast cancer patients
Poh Lian Choong, C.J.S. de Silva · 2002
This paper is concerned with the use of artificial neural networks (ANN) to construct distributions to carry out plausible reasoning in the field of medicine. It describes a comparison between multivariate logistic regression (MLR) and the entropy maximization network (EMN) in terms of explicit assessment of their predictive capabilities. The EMN and MLR have been used to determine the probability of harboring lymph node metastases at the time of initial surgery by assessment of tumor based parameters. Both predictors were trained on a set of 84 early breast cancer patient records and evaluated on a separate set of 92 patient records. Differences in performance were evaluated by comparing the areas under the receiver operating characteristic curve, A/sub z/. The EMN model performed more accurately with A/sub z/=0.839, compared to the MLR model with A/sub z/=0.809.