Early detection of breast cancer using hybrid of series network and VGG-16

Abstract

Breast cancer is nowadays becoming a serious problem and acts as a main reason for death of women around the world. Hence various devices are being utilized for the detection of breast cancer at an earlier stage and diagnosing it in an earlier stage might even results in complete cure of the disease. Among the wide range of devices available, mammogram is one of the commonly employed and most effective approaches involved in the detection of breast cancer. It records the affected area in the form of mammogram images and these images are processed through image processing techniques for the detection of cancer affected regions. In this paper, novelties have given in all the image processing aspects such as filtering, segmentation, feature extraction and classification. The salt and pepper noises in the mammogram images are eliminated by the usage of novel decision based partial median filter. Then the filtered images are segmented based utilizing a novel technique which is formed on integrating the deep learning techniques of VGG-16 and series network. Features of the segmented images have extracted through BAT-SURF feature extraction, where the orientation of the interest points are extracted using Bat optimization algorithm along with SURF (i.e.) Speeded up Robust Features. It extract most important key points from SURF features and then the extracted image has classified by using the novel Gradient descent decision tree classifier in which a stable learning path provided for easy convergence. Then the performance of the proposed system has analyzed based on the performance metrics like accuracy, specificity, sensitivity, recall, precision, Jaccard coefficient, F score and missed classification. Based on the results obtained, the conclusion of the proposed work has attained enhanced results on comparing with other state of the art approaches. The accuracy value of the proposed hybrid VGG-16 and series network segmentation technique determined as 96.45 and similarly the accuracy value of the proposed Gradient Descent Decision Tree Classification technique has value shows 95.15.