A Neural Network Decision-Support Tool for the Diagnosis of Breast Cancer
Joseph Downs, Robert F. Harrison, Simon S. Cross · White Rose Research Online (University of Leeds, The University of Sheffield, University of York) · 1994
An application of the ARTMAP neural network to the diagnosis of breast cancer is described. Performance results are given for 10 individual ARTMAP networks and the five most accurate such networks using "pooled" decision making (the so-called voting strategy). The results are compared with those of expert and neophyte human pathologists. These show that ARTMAP diagnoses are at least as accurate as those of the expert and can approach the optimum for the domain. However, human pathologists bias their predictions in order to minimise false positive predictions at the expense of increased false negatives. The same effect is achieved in ARTMAP by pruning category cluster nodes which make positive predictions.........