Neural Pattern Recognition Model for Breast Cancer Diagnosis

—In this article, we introduce a neural pattern recognition model for breast cancer diagnosis. This proposed model uses a two-stage back-propagation approach including both linear and nonlinear components of calculations along with iterative training processes and a learning shift controller. The iterative training processes allow the model to gradually increase the number of hidden neurons and input data size, reusing each of the iteratively final weights as new initial weights for the next iterative training stage. A learning rate is accordingly adjusted by the learning shift controller. This training approach ensures that even the local minima of the model have low enough sum-squared errors. The average testing diagnosis accuracy of our model is 98% for benign and malignant breast cancer. Therefore, our research results indicate that the proposed model can provide consistently high accuracy in the diagnosis and classification of benign and malignant breast cancer.