Differentially Evolved RBFNN for FNAB-Based Detection of Breast Cancer

In this work, computational intelligence-based detection of malignant and benign breast cancer classification has been obtained through the evolved radial basis function neural network. The cancer category has defined over observed parameters values of breast lesion extracted through fine needle aspiration. Generally, the final performance of a neural network is decided by the quality of the learning algorithm and transfer function characteristics selected over the active nodes. Gradient-based learning is very popular and useful also but suffered from trapping in the local minima and accuracy deficiency in the outcomes. Hence, the performance quality of the radial basis function neural network has been improved by evolving the basis function parameters and connection weights through a new mutation strategy in the differential evolution. The proposed approach maintains a better balance between exploration and exploitation by providing the probabilistic approach of mutation strategy selection among the defining the differential vector through an available best member or randomly selected member. The mutation strategy guided by the random member-based approach helps in exploration at a large level while the best member-based differential vector helps in faster convergence. The proposed work also provided the provision to have the neural network parameters in the integer value domain so that there will saving in the memory and easiness over hardware realization. The proposed algorithm's performances have been compared against the different commonly used strategies of available mutation strategies in the differential evolution and gradient-based strategy and have shown significant benefits in efficiency and precision. The proposed solution model can be used to assist the cytologist in making the final decision robust and accurate.