Breast cancer screening using evolved neural networks
Walker H. Land, L.E. Albertelli · 2002
This paper is based on a modified form of Fogel's evolutionary programming approach (1994) for evolving neural networks for the detection of breast cancer using fine needle aspirate (FNA) data. The evolved architectures routinely achieved a classification accuracy of greater than 96% for both the validation and test sets. Statistical analysis of fifty-two experiments demonstrated that the type II error distributions are both smaller and "tighter" than the type I error distributions for both the validation and test sets. Specifically, the mean value of the type II errors is less than one-half the mean value of the type I errors for the validation set while the mean value of the test set type II errors is less than one-fourth the mean value of the type I errors. Finally, the evolved architectures are generally simple structures with the most complex structure containing 8 nodes in the input layer, 5 nodes in the hidden layer, and one node in the output layer {8,5,1}. The simplest architecture was a {4,2,1}.