Classifying Breast Cancer Using a Hybrid Approach

Abstract

Breast cancer remains one of the leading causes of cancer-related mortality among women worldwide, and early diagnosis significantly improves survival rates. Histopathological image analysis plays a crucial role in breast cancer diagnosis; however, manual interpretation is time-consuming and susceptible to inter-observer variability. This study proposes a hybrid magnification-invariant deep learning framework named the Magnification Invariant Classifier (MIC) based on an enhanced UNET encoder-decoder architecture for automated breast cancer histopathology image classification. The proposed framework was evaluated using the publicly available BreaKHis dataset containing 9,109 histopathological images acquired at 40×, 100×, 200×, and 400× magnification levels. The dataset was divided into 70% training, 15% validation, and 15% testing subsets. Data augmentation techniques including rotation, flipping, cropping, and shearing were applied to improve model generalization and reduce overfitting. Experimental results demonstrate that the proposed model achieved an accuracy of 91.23%, precision of 94.51%, recall of 90.99%, F1-score of 92.71%, and an AUC score of 0.94, outperforming UNET, UNET++, VNET, and RESNET models by approximately 5–9% in classification accuracy. The proposed architecture effectively extracts both local and global tissue features while while remaining robust across varying magnification levels. The reduced loss value of 0.0876 further confirms the efficiency and stability of the proposed framework. The findings indicate that the proposed model can serve as an effective computer-aided diagnostic tool for assisting pathologists in accurate and early breast cancer detection.