Deep learning applications in breast cancer diagnosis: Enhancing mammography evaluation through convolutional neural networks
Saruchi Saruchi, Anupam Mittal, Geetika Sharma, Sushil Kamboj, Tarak Kurana · 2025
Breast cancer is still one of the biggest women&s;s health challenges in the world, where mortality rates are high only when detected early, thus calling for innovative ways for early detection. Although mammography is the gold standard for breast cancer screening, its interpretation is complicated and susceptible to interobserver variability, and it is also prone to false positives and negatives, leading to diagnostic delays. Automating feature extraction and classification with artificial intelligence (AI) and deep learning has emerged as a promising solution to improving the analysis of mammography. The performance of convolutional neural networks (CNNs) is extraordinarily good for image recognition tasks, especially when dealing with mammographic images. Furthermore, deep learning models trained using DDSM or CBIS-DDSM datasets are shown to achieve high accuracy distinguishing normal, benign, and malignant tissues. By combining classifiers, including SVM and ELM, with the advanced models like AlexNet, we show that the accuracy can reach up to 100%, sensitivity can reach up to 99.58%, and specificity can reach up to 95.61%, outperforming the state-of-the-art work and existing methodologies for breast mass classification on benchmark datasets (MIAS and INbreast). These results suggest that accuracy can improve by 5–10% and specificity by over 7% compared to previous approaches. Transfer learning and preprocessing are used to improve model performance. However, challenges still include the lack of abundant annotated datasets, the need for the model to be robust across many populations, and the interpretation of deep learning systems. To overcome limitations, continued research and innovation in AI-driven mammography will need to continue to accomplish the goal of revolutionizing breast cancer diagnostics and therefore improve patient outcomes.