Model selection for a medical diagnostic decision support system: a breastcancer detection case

There are a number of different quantitative models that can be used in a m edical diagnostic decision support system (MDSS) including parametric metho ds (linear discriminant analysis or logistic regression), non-parametric mo dels (K nearest neighbor, or kernel density) and several neural network mod els. The complexity of the diagnostic task is thought to be one of the prim e determinants of model selection. Unfortunately, there is no theory availa ble to guide model selection. Practitioners are left to either choose a fav orite model or to test a small subset using cross validation methods. This paper illustrates the use of a self-organizing map (SOM) to guide model sel ection for a breast cancer MDSS. The topological ordering properties of the SOM are used to define targets for an ideal accuracy level similar to a Ba yes optimal level. These targets can then be used in model selection, varia ble reduction, parameter determination, and to assess the adequacy of the c linical measurement system. These ideas are applied to a successful model s election for a real-world breast cancer database. Diagnostic accuracy resul ts are reported for individual models, for ensembles of neural networks, an d for stacked predictors. (C) 2000 Published by Elsevier Science B.V.