Abstract:
Breast cancer remains one of the most prevalent and life-threatening diseases, with early detection playing a crucial role in improving patient outcomes. Deep learning techniques have emerged as powerful tools for automating cancer diagnosis, yet challenges persist in optimizing accuracy and interpretability. This study employs a hybrid deep learning model integrating Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks for classifying breast cancer histopathological images from the Kaggle dataset. The dataset comprises 277,524 high-resolution images labelled as benign and malignant, requiring extensive preprocessing, including image augmentation and noise reduction, to enhance generalizability. The CNN component extracts spatial features, while BiLSTM captures temporal dependencies across multi-view pathology images. The model is trained using Adam optimizer with a learning rate of 0.0001 and a batch size of 32 for 50 epochs, employing a 5-fold cross-validation strategy. The proposed model achieves an accuracy of 92.7%, surpassing traditional CNNs (86.8%) and ResNet-50 (90.2%), with an AUC-ROC score of 0.91, demonstrating superior feature extraction and classification capabilities. The findings highlight the effectiveness of hybrid architectures in improving breast cancer detection, suggesting the potential for further refinement through Grad-CAM visualizations and expansion to multi-class cancer datasets.
