PCA-DRIVEN MULTISTAGE-MULTISCALE FUSION FRAMEWORK FOR BREAST HISTOPATHOLOGY IMAGE CLASSIFICATION

Abstract

Breast cancer is worldwide existing one of the critical diseases among the women. This is spreading quite drastically and many people facing with this disease. Early diagnosis, prognosis and treatment are only the way with which we can fight and overcome this and also save the life of someone. There are different diagnosis systems already developed and also being developed to diagnose breast cancer at early stage, but the performance of these systems is a major concern. Developing a lightweight and more efficient diagnosis system with is quite difficult. To improve the performance, we have proposed a principal component analysis (PCA) with attention based Deep Convolutional Neural Network (DCNN). There is PCA driven feature extraction approach is adopted. The input image is directly feeded to the base Deep network model after performing augmentation and extracted features at every stage are concatenated then PCA reduce the features maps with more significant feature adaptation, finally these features are enhanced to put more discrimination through the attention network. The proposed model is reporting a superior accuracy of 97.68% as that of existing state of the methods.