Breast cancer classification from histopathological images using dual deep network architecture

Breast Cancer is one of the fatal diseases and leading cause of mortality in women all over the world; moreover, early detection of breast cancer can minimize the risk of death, however accurate detection and classification of breast cancer is critical task. Histopathology is a technique used for a breast cancer diagnosis; histopathological images comprise rich phenotypic information which is utilized for monitoring underlying techniques contributing to patient survival and disease progression outcomes. In recent years, deep learning has achieved success in the medical domain, and further, it has become a primary methodological choice for interpreting and analyzing histology images. The existing approach of histopathological image classification requires a huge amount of labeled data to achieve satisfactory results which face the challenge due to limited annotated data due to cost constraints. A promising mechanism is required to be designed for binary classification; Thus in this research work, Dual Deep Network architecture (DDNA) is designed for lesion identification and binary classification; Dual Deep Network comprises two novel networks i.e. PSNet1 and PSNet2; PSNet1 is designed to extract the dynamic feature and identify the lesion. PSNet2 is designed for binary classification using the PSNet1 feature; further attention module is used for feature mapping and enhancing the feature extraction and network optimization is carried out to enhance the performance. DDNA is evaluated on the BreakHis dataset on image level and patient-level considering the different metrics; also comparative analysis is carried out with various deep learning techniques and varying magnification factors as 40X, 100X, 200X, and 400X. Moreover, the evaluation shows the model’s efficiency which ranges between 99 and 100% considering image level and patient level. © 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.