Investigating Breast Cancer Detection with Contextual Relationship Embedded CNN in Mammograms

Breast cancer primarily affects women, caused due to the excess growth of malignant breast tissues. The segmentation and early detection process suffered due to the complex and varied nature of breast tissue. To address this challenge, this research proposes a Convolutional Neural Network model with Contextual Relationship Embedding to accurately segment pathological mass regions in mammogram images. In this research work, the mammogram images are collected from datasets and are preprocessed to enhance image quality, noise reduction and contrast enhancement. By using a Deep Convolutional Neural Network, the edges in the highly contrasted regions, complex structure and spatial relationships of the images are gathered by using different operators. The extracted features are concatenated through the Fully Connected-Convolutional Block Attention Module. The contextual relationship embedded features are integrated with the original features, guided by the cross-entropy loss function with contextual relationship constraints. This enables the model to generate more precise decisions for segmentation and boundary identification. The proposed method's efficiency is validated and the proposed model achieves superior performance with an accuracy of 99.59% and an error rate of 0.405%. Overall, this research article concludes that the proposed model is more efficient for breast cancer detection than other existing models.