TU-GUA: Architecture with Transformer and Unified Attention for Bone Tumor Segmentation in Radiographs

• Iulle M. G. Neves UFPI
• Mauro A. G. Clark UFPI
• Mateus Assis Veloso UFPI
• João O. B. Diniz IFMA
• José D. L. Araújo UFPI
• Antonio O. de Carvalho Filho UFPI

DOI: https://doi.org/10.5753/eniac.2026.30589

Resumo

O diagnóstico preciso de tumores ósseos em radiografias convencionais é desafiador devido à baixa distinção de contraste e à complexidade das bordas tumorais. Este trabalho propõe a arquitetura TransUNetWithGUA (TU-GUA), que integra um Simple Windowed Transformer no gargalo da rede e o bloco GUA-Fusion para fusão multi-escala com atenção unificada (canal e espacial) e gating adaptativo nas conexões encoder–decoder. O modelo foi avaliado sob protocolo reprodutível e comparado com arquiteturas baseline. Os resultados demonstram que o método proposto alcança o maior recall na detecção de lesões tumorais, característica prioritária para rastreamento oncológico, além de robustez qualitativa em cenários de baixo contraste.

Referências

Bloier, M., Hinterwimmer, F., Breden, S., Consalvo, S., Neumann, J., Wilhelm, N., and Burgkart, R. (2022). Detection and segmentation of heterogeneous bone tumours in limited radiographs. Current Directions in Biomedical Engineering, 8(2):69–72.

Chen, L.-C., Papandreou, G., Schroff, F., and Adam, H. (2017). Rethinking atrous convolution for semantic image segmentation. arXiv preprint arXiv:1706.05587.

Dosovitskiy, A. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929.

He, Y., Pan, I., Bao, B., Halsey, K., Chang, M., Liu, H., and Bai, H. X. (2020). Deep learning-based classification of primary bone tumors on radiographs: a preliminary study. EBioMedicine, 62.

Hu, J., Shen, L., and Sun, G. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 7132–7141.

Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., and Guo, B. (2021). Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 10012–10022.

Oktay, O., Schlemper, J., Folgoc, L. L., Lee, M., Heinrich, M., Misawa, K., Mori, K., McDonagh, S., Hammerla, N. Y., Kainz, B., Glocker, B., and Rueckert, D. (2018). Attention u-net: Learning where to look for the pancreas. arXiv preprint arXiv:1804.03999.

Oliveira, M. V., Matos, C. E. F., Júnior, G. B., and Almeida, J. D. S. (2025). Dual-scale se unet: Uma rede neural convolucional aprimorada com se blocks para segmentação renal em imagens de tc. In Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), pages 760–771. Sociedade Brasileira de Computação (SBC).

Rhyou, S. Y., Bang, C., Cho, Y. J., Bae, H., Ha, Y. J., Baek, S. Y., and Moon, J. E. (2025). Tracking temporal progression of benign bone tumors through x-ray based detection and segmentation. Scientific Reports, 15(1):39491.

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI), pages 234–241. Springer.

Saifullah, S., Dreżewski, R., Yudhana, A., Wielgosz, M., and Caesarendra, W. (2025). Modified u-net with attention gate for enhanced automated brain tumor segmentation. Neural Computing and Applications, 37(7):5521–5558.

Wang, H. et al. (2025). Deep learning models in classifying primary bone tumors and bone infections based on radiographs. NPJ Precision Oncology, 9(1):72.

Woo, S., Park, J., Lee, J., and Kweon, I. S. (2018). Cbam: Convolutional block attention module. In Proceedings of the European Conference on Computer Vision (ECCV), pages 3–19.

Xie, Z., Zhao, K., Yan, X., Wu, S., Mei, J., and Lu, H. (2022). Merged u-net for bone tumors x-ray images segmentation. In 2022 IEEE International Conference on Image Processing (ICIP), pages 1276–1280. IEEE.

Yao, S., Huang, Y., Wang, X., Zhang, Y., Paixao, I. C., Wang, Z., and Song, J. (2025). A radiograph dataset for the classification, localization, and segmentation of primary bone tumors. Scientific Data, 12(1):88.

Zhou, Z., Rahman Siddiquee, M. M., Tajbakhsh, N., and Liang, J. (2018). Unet++: A nested u-net architecture for medical image segmentation. In International Workshop on Deep Learning in Medical Image Analysis, pages 3–11. Springer.

Zuiderveld, K. J. (1994). Contrast limited adaptive histogram equalization. In Heckbert, P. S., editor, Graphics Gems, pages 474–485. Elsevier.