BCD-TransNet: Automatic breast cancer detection and classification using transfer learning approach

Abstract

Breast Cancer (BC) is a predominant form of cancer diagnosed in women and one of the deadliest diseases. The important cause of death owing to the cancer amongst women is BC. However, the existing ML techniques are very challenge evaluate the performance of the classification of BC and difficult task for early diagnosis. To overcome this challenge, transfer learning framework have been broadly applied to histopathological images for classifying tumour. So, in this research a novel BC Detection using Transfer learning network (BCD-TransNet) is introduced to identify and classify BC stages. Initially, the histopathological images from BreakHis dataset are pre-processed using stationary wavelet based Retinex (SWR) for eliminating the noise and progress the image quality. The noise-free images are segmented using the Hybrid Greedy Snake-Krill Herd Optimization (HGS-KHO) algorithm. The BCD-TransNet model that incorporates with five different pre-trained networks in which the knowledge attained by each model is transfer to next network for extracting the most relevant features. This detection model has two different phases namely first level classification for identifying benign and malignant cells and the second level classification for identifying the different types in benign and malignant. Finally, the ML-based Decision tree is used to detect the stages of breast tumour. From the simulation analysis, the BCD-TransNet present well accuracy of 99.31% for the classification of breast tumour. The proposed Transfer learning-based BCD-TransNet model improves the overall accuracy 2.11%, 13.31%, 1.82% better than DLA-EABA, Pa-DBN-BC, TTCNN respectively.

Keywords: Greedy snake optimization; Krill Herd optimization algorithm; breast cancer; decision tree; stationary wavelet based Retinex.