Habib Rostami

Breast cancer is the second most common cancer in women and is becoming more prevalent worldwide. Regular screenings are crucial for early treatment. Digital Mammography (DM) is the most common method for breast cancer screening, and Contrast- Enhanced Spectral Mammography (CESM or CM) helps detect hidden abnormalities, especially in dense breast tissue. In this work, two methods for the classification and segmentation of breast lesions were presented, including a semi-automatic method and an automatic method. In the semi- automatic method, the doctor delineates the tumor boundaries, and then the type of tumor (benign or malignant) is determined using a classifier model. We also introduced an automatic method, where the segmentation of tumor boundaries is performed automatically using a U-Net neural network, followed by tumor classification. Additionally, a new architecture (JointNet) was proposed for classification, which simultaneously extracts local and global features from the input, resulting in improved classification accuracy. Results: The best result using the semi-automatic segmentation method is 94.74% (on ROI Images) and in the automatic we've reached 80.65% accuracy. Conclusions: Results of our experiments suggest that for breast cancer lesion classification, lesion shape has more effect than their tissue.