Breast cancer identification using feature level fusion and hybrid GA-PSO optimized neural network

Breast cancer, being a critical disorder in urban women, needs to be screened as early as possible for an effective treatment. In this study, the mammographic image analysis society (MIAS) database mammogram images are utilized. The images are first passed to a median filter before thresholding. The pectoral muscles are also removed with image background and labels. Contrast limited adaptive histogram equalization (CLAHE), wavelet denoising and guided filtration are applied for a region of interest (RoI) extraction. The resultant image features are extracted using gray level co-occurrence matrix GLCM, improved local binary pattern (iLBP) and discrete wavelet transform (DWT). These features are used to classify the image data using an artificial neural network (ANN), whose weights and bias are optimized with hybrid genetic algorithm and particle swarm optimization (HGAPSO). This technique helped solve the scalability issue in GA and the local optima in PSO, with the hybrid algorithm converging faster with better results. The resultant optimized network gives an accuracy of 99.72% in classifying the images as normal, benign or malignant.