Breast cancer diagnosis using artificial neural networks

Breast Cancer is one of the most dangerous diseases for women. Mammography is an effective method in early detection. However, there are difficulties in accurate analysis of some mammogram images. Therefore, a method of data analysis using artificial neural networks (ANNs) has been developed. In this thesis, the performances on the Wisconsin breast cancer data (WBCD) of three different neural network models: Multi-layer neural networks (MLPs), Trigonometric Neural Networks (TNNs), and Exponential Neural Networks (ENNs) are examined. These models are based on a back propagation algorithm, with different activation functions. The activation function is one of most factors to influence the performance of ANNs. The purpose of thesis is to test the hypothesis that the performance of TNNs and TNNs on breast cancer dataset is better than MLPs. The strategic experiments are implemented. The overall performances of three models are evaluated and discussed through an analysis of four aspects of testing results: correctness rate, root mean squared error, training speed and misclassification cost. Moreover, from the testing results, the basis of further work is formed.