An Improved Convolutional Network for Breast Cancer Detection

Breast cancer is the most common among all types of cancers impacting women worldwide; early identification is vital in decreasing mortality rates. This study presents a deep learning framework utilizing ResNet architecture, processed with wavelet features, to improve the robustness of breast cancer detection. The framework employs ResNet50 as the backbone for feature extraction and incorporates wavelet-based features to offer complementary information. The model attains enhanced performance through meticulous preprocessing, wavelet transformation, and fine-tuning strategies. A Wavelet-Residual Network (WResNet) is introduced, integrating both convolutional filters and wavelet filters in parallel to effectively capture features in both the spatial as well as frequency domains. The filtered images were integrated and subjected to a MaxPooling layer to extract significant features. INBreast dataset of breast mammography images is used to train the proposed model. The model achieved a high test accuracy of 98.1% with an AUC of 0.99. The results indicate that the proposed WResNet framework offers a promising solution for reliable breast cancer detection in clinical settings.