Comprehensive Performance Evaluation of Deep Learning Models for Optimizing Breast Cancer Classification on Mammogram Datasets
Sharmin Akhter, Jiangjiang Liu · 2025
Breast cancer is one of the most common and life-threatening diseases for women around the world, making early detection crucial for better patient outcomes. In recent years, deep learning models have demonstrated considerable promise in the classification and diagnosis of breast cancer using medical imaging. This research focuses on the development of advanced deep learning models for classifying breast cancer images utilizing a mammogram dataset. A recently published dataset of 745 original images was compiled for model training. Multiple deep learning architectures were examined to determine the best-performing model for accurate classification. Among them, DenseNet201 outperformed other models with the highest accuracy of 95.58%, followed by Xception and MobileNet V2, which achieved accuracy rates of 95.30% and 93.98%, respectively. However, VGG19 and ResNet50 yielded comparatively lower accuracy rates of 73.75% and 66.42%. Furthermore, we evaluated our model using the INbreast dataset, a widely recognized benchmark for breast cancer classification, to assess its generalization capability. The model demonstrated remarkable performance, achieving an accuracy of 98.85%, along with superior precision, recall, and F1-score, showcasing its robustness in real-world clinical applications. These results highlight the potential of our proposed deep learning model in assisting radiologists with early breast cancer detection and improving diagnostic accuracy.