Improving diagnostic accuracy using a hierarchical neural network to modeldecision subtasks

A number of quantitative models including linear discriminant analysis, log istic regression, k nearest neighbor, kernel density, recursive partitionin g, and neural networks are being used in medical diagnostic support systems to assist human decision-makers in disease diagnosis. This research invest igates the decision accuracy of neural network models for the differential diagnosis of six erythamatous-squamous diseases. Conditions where a hierarc hical neural network model can increase diagnostic accuracy by partitioning the decision domain into subtasks that are easier to learn are specificall y addressed. Self-organizing maps (SOM) are used to portray the 34 feature variables in a two dimensional plot that maintains topological ordering. Th e SOM identifies five inconsistent cases that are likely sources of error f or the quantitative decision models; the lower bound for the diagnostic dec ision error based on five errors is 0.0140. The traditional application of the quantitative models cited above results in diagnostic error levels subs tantially greater than this target level. A two-stage hierarchical neural n etwork is designed by combining a multilayer perceptron first stage and a m ixture-of-experts second stage. The second stage mixture-of-experts neural network learns a subtask of the diagnostic decision, the discrimination bet ween seborrheic dermatitis and pityriasis rosea. The diagnostic accuracy of the two stage neural network approaches the target performance established from the SOM with an error rate of 0.0159. (C) 2000 Elsevier Science Irela nd Ltd. All rights reserved.