Using pattern recognition approach for providing second opinion of breast cancer diagnosis

The objective of study is to develop intelligent decision support system to aid radiologist in diagnosis using pattern recognition techniques to estimate diagnostic function. In this study 3 approaches investigated namely statistical, neural networks and optimization techniques which were applied on the Wisconsin dataset. Trained neural networks, with the data set used as input, improve on the independent variables LDF and LR for discriminating between true and false cases. The performance of Multilayer Perceptrons, Delta-Bar-Delta neural networks, LDF and LR can be improved with optimization of the features in the input. Neural network analyses show promise for increasing diagnostic accuracy of classifying the cases. The areas under the ROC curves for MLP, and DBD were 0.929, and 0.927 respectively. For the full models of LDF and LR were 0.887 and 0.917 respectively. With the use of forward selection (fs) and backward elimination (be) optimization techniques, the areas under the ROC curves for MLP and the LR were increased to approximately 0.93.