Extended Deep-Learning Network for Histopathological Image-Based Multiclass Breast Cancer Classification Using Residual Features
Autonomy of breast cancer classification is a challenging problem, and early diagnosis is highly important. Histopathology images provide microscopic-level details of tissue samples and play a crucial role in the accurate diagnosis and classification of breast cancer. Moreover, advancements in deep learning play an essential role in early cancer diagnosis. However, existing techniques involve unique models for each classification based on the magnification factor and require training numerous models or using a hierarchical approach combining multiple models irrespective of the focus of the cell features. This may lead to lower performance for multiclass categorization. This paper adopts the DenseNet161 network by adding a learnable residual layer. The learnable residual layer enhances the features, providing low-level information. In addition, residual features are obtained from the convolution features of the preceding layer, which ensures that the future size is consistent with the number of channels in DenseNet’s layer. The concatenation of spatial features with residual features helps better learn texture classification without the need for an additional texture feature extraction module. The model was validated for both binary and multiclass categorization of malignant images. The proposed model’s classification accuracy ranges from 94.65% to 100% for binary and multiclass classification, and the error rate is 2.78%. Overall, the suggested model has the potential to improve the survival of breast cancer patients by allowing precise diagnosis and therapy.