<![CDATA[Breast cancer is one of the leading causes of death in women, so accurate early detection is key to improving patient survival rates. Although mammography is the standard method for breast cancer screening, manual interpretation of mammogram images still depends on the expertise of radiologists and has the potential to lead to misdiagnosis. This study proposes a concatenate transfer learning-based deep learning approach by combining two Residual Network architectures, namely ResNet50 and ResNet152, to improve feature representation capabilities in mammogram image classification. The dataset used is a combination of MIAS and CBIS-DDSM with two classes, namely benign and malignant. The evaluation was carried out using original test data without augmentation to ensure the objectivity of the results. The experimental results show that the proposed model achieves an average accuracy of 97.09% and outperforms several individual transfer learning models. The main contribution of this study lies in demonstrating that combining deep features from similar but different depth CNN architectures can improve classification stability and accuracy. These findings provide a conceptual basis for the development of more reliable deep learning-based medical decision support systems for early breast cancer detection.]]>
Copyrights © 2026
