Ensemble Deep Learning for Histopathological Breast Cancer Detection

Breast cancer remains one of the leading causes of mortality among women worldwide, and early and accurate diagnosis is essential for effective treatment. In this study, we propose an ensemble deep learning approach for classifying histopathological images of breast cancer using the BreaKHis dataset. Two state-of-the-art convolutional neural network architectures, ResNet50 and DenseNet121, were fine-tuned and combined through multiple ensemble strategies, including stacking with logistic regression, XGBoost, and hard/soft voting. The models were trained on an 80/20 train-validation split, preserving the distribution of benign and malignant classes across all magnification levels (40X, 100X, 200X, and 400X). Experimental results show that the hard voting ensemble achieved the highest accuracy of 98.61%, closely followed by the XGBoost ensemble with an accuracy of 98.55%, both outperforming individual models. These findings highlight the effectiveness of ensemble deep learning in improving classification performance for breast cancer histopathology and suggest its potential for aiding pathologists in clinical decision-making.