Classification of Breast Histology into Benign and Malignant using ResNet-Variants: A Study with BreakHis Database

Cancer is a harsh disease, and early diagnosis and management are essential to reduce its severity. The happening rate of Breast-Cancer (BC) in women’s community is gradually rising globally due to various causes, and premature detection is essential. Identification of the harshness (benign/malignant) of the cancer is necessary to implement the required treatment. This research aims to develop a pretrained deep-learning (PDL) based technique to detect the benign/malignant class BC from the histology slides. Various stages found in this research include image collection and resizing, feature extraction and feature reduction with $50 \%$ dropout, serial features concatenation and fused features vector generation, and 3 -fold cross-validation and performance confirmation. This work implements the experiments using ResNet-variants using the SoftMax classifier, and this study’s result substantiate that this scheme helps get > 88% accuracy using individual features and $100 \%$ accuracy with fused features.