Diversity-Driven CNN–Transformer Ensemble for Robust Single-View Mammography Screening

• Rodrigo Barbosa de Oliveira UFPE
• Tsang Ing Ren UFPE

DOI: https://doi.org/10.5753/eniac.2026.31356

Resumo

Breast cancer screening in resource-constrained settings may involve digitized screen-film mammography and incomplete imaging protocols. This work investigates ROI-level mass classification in single-view digitized mammography using a hybrid CNN–Transformer ensemble. The proposed framework combines ConvNeXt, DenseNet, and Swin-Transformer models under a patient-aware split on the CBIS-DDSM mass subset, reducing the risk of patientlevel data leakage. The ensemble achieved an AUC of 0.8529, Recall of 0.8250, Precision of 0.8049, and F1-score of 0.8148, improving the sensitivity–precision balance over individual backbones. An error-correlation analysis suggests complementary failure patterns between convolutional and attention-based architectures. Stratification by BI-RADS breast density further indicates that Transformer-based models may provide greater robustness in dense tissue categories, although validation on additional cohorts remains necessary. The proposed method should therefore be interpreted as a reproducible ROI-level benchmark for mass classification rather than a complete clinical screening system.

Referências

Al-Antari, M. A., Al-Masni, M. A., and Kim, T.-S. (2020). Deep learning computer-aided diagnosis for breast lesion classification in digital mammograms. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, pages 31–40. Springer.

Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., and Houlsby, N. (2021). An image is worth 16x16 words: Transformers for image recognition at scale. International Conference on Learning Representations (ICLR).

Kebede, S. R. et al. (2024). Dual view deep learning for enhanced breast cancer screening using mammography. Scientific Reports, 14(1):3839.

Lee, J. G., Jun, S., Cho, Y.-W., Lee, H., Kim, G. B., Seo, J. B., and Kim, N. (2017). Deep learning in medical imaging: General overview. Korean Journal of Radiology, 18(4):570–584.

Liao, L. and Aagaard, E. M. (2024). An open codebase for enhancing transparency in deep learning-based breast cancer diagnosis utilizing CBIS-DDSM data. Scientific Reports, 14(1):27318.

Manzari, O., Ahmad, H., et al. (2024). Hybrid CNN and transformer framework for breast cancer classification using mammography images. Scientific Reports, 14(1):1208.

Petrini, D. G. P. et al. (2022). Breast cancer diagnosis in two-view mammography using end-to-end trained efficientnet-based convolutional network. Nuclear Engineering and Technology, 54(1):325–336.

Petrini, D. G. P. and Kim, H. Y. (2025). Optimizing breast cancer detection in mammograms: A systematic evaluation of deep learning architectures. arXiv preprint arXiv:2503.19945.

Sawyer-Lee, R., Gimenez, F., Hoogi, A., and Rubin, D. . T. C. I. A. (2016). Curated breast imaging subset of digital database for screening mammography (cbis-ddsm) [data set].