Enhanced Neural Networks and Medical Imaging

摘要

This paper shows that the application of Enhanced Neural Networks when dealing with classification problems is more powerfull than classical Multilayer Perceptrons. These enhanced networks are able to approximate any function f(x) using a n-degree polinomial defined by the weights in the connections. Also, the addition of hidden layers in the neural architecture, increases the degree of the output equation associated to output units. So, surfaces generated by these networks are really complex and theoretically they could classify any pattern set with a number n of hidden layers. Results concerning medical imaging, breast cancer diagnosis, are studied along the paper. The proposed architecture improves obtained results using classical networks, due to the implicit data transformation computed as part of the neural architecture.