MIT-Net: Arquitetura Híbrida Transformer–U-Net para Segmentação de Lesões Mamárias em Ultrassom

• Leandro Iglesias Moura de Freitas UFMA
• Aristófanes Correa Silva UFMA
• Anderson Silva Lopes UFMA

DOI: https://doi.org/10.5753/sbcas.2026.21557

Resumo

O câncer de mama é a neoplasia mais incidente entre mulheres em 157 países, e o diagnóstico precoce é o principal determinante do prognóstico. A ultrassonografia ocupa papel central no rastreamento, mas a delimitação manual de lesões é lenta, subjetiva e sujeita a variabilidade inter-observador significativa, devido ao ruído speckle e a artefatos acústicos. Este trabalho propõe a MIT-Net, uma arquitetura híbrida encoder-decoder que combina o encoder MiT-B5, pré-treinado no ImageNet-1k [Deng et al. 2009], com um decoder customizado inspirado na U-Net, com upsampling progressivo e conexões residuais. Avaliada no conjunto de dados BUSI com 647 imagens anotadas, a MIT-Net alcançou coeficiente Dice de 90,03 % (± 1,22) e IoU de 84,18 % (± 1,57).

Referências

Al-Dhabyani, W., Gomaa, M., Khaled, H., and Fahmy, A. (2020). Dataset of breast ultrasound images. Data in Brief, 28:104863.

Aslam, M., Ahmad, L., Rahman, N. U., and Zaki, A. S. (2025). HA-Net: Hierarchical attention network for breast tumor segmentation in ultrasound images. Scientific Reports, 15(1):39633.

Berg, W. A., Gutierrez, L., NessAiver, M. S., et al. (2004). Diagnostic accuracy of mammography, clinical examination, US, and MR imaging in preoperative assessment of breast cancer. Radiology, 233(3):830–849.

Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. (2009). Imagenet: A large-scale hierarchical image database. In 2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255. IEEE.

Loshchilov, I. and Hutter, F. (2017). Sgdr: Stochastic gradient descent with warm restarts. In International Conference on Learning Representations (ICLR).

Loshchilov, I. and Hutter, F. (2019). Decoupled weight decay regularization. In International Conference on Learning Representations (ICLR).

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. In MICCAI, pages 234–241.

Sung, H., Ferlay, J., Siegel, R. L., et al. (2021). Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: A Cancer Journal for Clinicians, 71(3):209–249.

Wang, T., Liu, J., and Tang, J. (2025). A cross-scale attention-based U-Net for breast ultrasound image segmentation. Journal of Imaging Informatics in Medicine, 38(5):2851–2864.

Xie, E., Wang, W., Yu, Z., et al. (2021). SegFormer: Simple and efficient design for semantic segmentation with transformers. In NeurIPS, volume 34, pages 12077–12090.

Zhang, H., Lian, J., and Ma, Y. (2025a). FET-UNet: A fine-grained edge-aware transformer UNet for breast ultrasound lesion segmentation. Physica Medica, 133:104969.

Zhang, Y., Li, Y., Zheng, J., et al. (2025b). Multi-level feature attention network for medical image segmentation. Expert Systems with Applications, 263:125785.