Evolving wavelet neural networks for breast cancer classification

Digital Mammograms are x-ray images of the breast and one of the preferred early detection methods for breast cancer. However, mammograms are still difficult to interpret, and associated with this problem is a high percentage of unnecessary biopsies, misdiagnoses and late detections. The focus of this research is to use neuroevolutionary mechanisms for detecting breast cancer from mammographic images. The aim is to design a sophisticated classification tool that detects breast cancer at its early stages, so that treatment has a better chance of success. Wavelet neural networks have the ability to capture and extract information at various frequency levels by using different dilation and scaling values of the wavelet function. In this work, the wavelet neural network parameters are evolved using on the concept of Cartesian Genetic Programming, resulting in an evolved neural network which is trained for mass diagnosis. In the reported study the proposed algorithm achieves a classification accuracy of 89.57% on a real dataset composed of 200 images. Such a computer-based classification system has the potential to provide a second opinion to the radiologists, thus assisting them to diagnose the malignancy of breast cancer more precisely.