Classification of breast cancer using MLO and CC images with deep learning method

This study was conducted to compare ResNet50 and InceptionV3 architectures, which are popular deep learning models for breast cancer detection. Breast cancer, one of the most common malignant neoplasms among women, can cause significant increases in mortality rates if not diagnosed in time. Therefore, accurate evaluation of mammographic images used for early detection is of great importance. However, manual interpretation of mammography images carries various error risks and can affect the accuracy of the results. In this context, artificial intelligence and deep learning-based automated diagnosis systems are emerging as an important alternative for breast cancer detection in medical imaging. In this study, using the Digital Database for Screening Mammography (DDSM) dataset, the breast cancer classification performances of both models on mammographic images were examined. In this process, mammographic images from both Cranio-Caudal (CC) and Medio-Lateral Oblique (MLO) angles were used to compare the models on categorical accuracy metrics. The results show that ResNet50 and InceptionV3 models perform differently in breast cancer classification. While the ResNet50 model provides higher accuracy rates, especially on CC images, the InceptionV3 model performs consistently and competitively on both CC and MLO images. While the results show that both models can be effectively used in automatic breast cancer diagnosis, there are performance differences in different data types and image angles. This study provides important findings on the usability of AI-based automated diagnostic systems in breast cancer diagnosis and provides potential contributions to the dissemination of these systems in clinical applications.