Clinical Co-occurrence Priors for Multilabel Classification in Brazilian Gastroscopy Images

• Willgnner Ferreira Santos UFG
• Paulo Victor dos Santos UFG
• Daniela Medeiros Milhomem Cardoso HGG
• Marcella Scoczynski Ribeiro Martins UTFPR
• Ronaldo Martins da Costa UFG

DOI: https://doi.org/10.5753/eniac.2026.30033

Resumo

This work evaluates relational formulations for multilabel gastroscopy classification, motivated by finding co-occurrences in upper gastrointestinal endoscopy. On a Brazilian dataset of 1,990 images evaluated across five core clinical labels under stratified five-fold cross-validation with a ResNet50 backbone, we compare independent binary cross-entropy (M0 BCE), Asymmetric Loss (M0 ASL), Classifier Chains (M1), explicit co-occurrence loss regularization (M2), and ML-GCN (M3). Co-occurrence regularization (M2) obtains the highest mean macro-F1 (0.537 ± 0.041), whereas ML-GCN underperforms (0.323± 0.023). However, paired bootstrap resampling shows M2’s confidence interval overlaps with plain BCE (0.517±0.045), indicating no formulation conclusively outperforms ResNet50+BCE. The primary contribution is a rigorous benchmark and informative null result on relational modeling in small clinical label regimes.

Referências

Bravo, J. I., Gomez, A. M., and Rodriguez, C. E. (2023). Spatio-temporal deep learning models for anatomical quality assessment in upper gastrointestinal video endoscopy. Computers in Biology and Medicine, 161:107005.

Cao, M., Lin, K., Wang, S., and Zhao, P. (2025). Multitask learning framework for autoimmune gastritis diagnosis and anatomical localization in gastroscopy. IEEE Transactions on Biomedical Engineering, 72(3):954–963.

Chen, Z.-M., Wei, X.-S., Wang, P., and Guo, Y. (2019). Multi-label image recognition with graph convolutional networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5177–5186.

Ferreira, R. A., Souza, F. M., and Martins, M. S. R. (2024). Deep learning frameworks for endoscopic image quality and pathology classification in public healthcare systems. IEEE Transactions on Medical Imaging, 43(1):412–422.

He, K., Zhang, X., Ren, S., and Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778.

Hossain, M. A., Rahman, M. M., and Islam, M. S. (2023). Comparative evaluation of convolutional and vision transformer architectures for endoscopic polyp classification and segmentation. IEEE Access, 11:94512–94526.

Kipf, T. N. and Welling, M. (2017). Semi-supervised classification with graph convolutional networks. In International Conference on Learning Representations (ICLR).

Klang, E., Barash, Y., Soffer, S., Zimlichman, E., Eliakim, R., and Kopylov, U. (2023). Deep learning applications in upper gastrointestinal endoscopy: A systematic review and meta-analysis. Digestive and Liver Disease, 55(4):445–453.

Krenzer, A., Shah, Z., Troya, J., and Puppe, F. (2026). Evaluation of loss functions and class weighting schemes for multilabel medical image classification. In Lecture Notes in Computer Science (LNCS), volume 14526, pages 276–288. Springer Nature.

Li, C., Wang, R., Xu, J., and Zhang, Y. (2022). Anatomical landmark and region recognition in upper gastrointestinal endoscopy: A benchmark study. Medical Image Analysis, 78:102401.

Liu, X., Chen, H., Zhou, W., and Wang, Y. (2024). Kyoto gastritis score recognition via multi-output efficientnet models in upper endoscopy. Computer Methods and Programs in Biomedicine, 245:108021.

Lopes, R. V., Almeida, T. P., and Santos, P. V. (2024). Self-supervised visual representation learning for gastric atrophy and intestinal metaplasia screening. IEEE Signal Processing Letters, 31:1405–1409.

Read, J., Pfahringer, B., Holmes, G., and Frank, E. (2011). Classifier chains for multi-label classification. Machine Learning, 85(3):333–359.

Ridnik, E., Ben-Baruch, E., Zamir, N., Noy, A., and Friedman, I. (2021). Asymmetric loss for multi-label classification. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 82–91.

Sechidis, K., Tsoumakas, G., and Vlahavas, I. (2011). On the stratification of multi-label data. In Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases (ECML PKDD), pages 145–158. Springer.

Silva, L. M., Santos, P. H., Oliveira, G. R., and Costa, R. M. (2023). Artificial intelligence in gastrointestinal endoscopy: A brazilian multi-center diagnostic benchmark. IEEE Journal of Biomedical and Health Informatics, 27(2):890–899.

Wei, J., Zhang, H., Li, R., and Sun, Y. (2025). Multi-granularity collaborative network for hierarchical pathology classification in upper gastrointestinal endoscopy. Pattern Recognition, 158:111045.

Yang, Y., Zhao, Q., Lin, X., and Wang, H. (2023). Automated identification of gastritis and intestinal metaplasia via deep learning in esophagogastroduodenoscopy. Frontiers in Oncology, 13:1007237.

Zhang, M., Wang, J., Liu, Y., Chen, X., and Zhao, L. (2023). Gastric lesion classification and early cancer detection in endoscopic images using multi-scale convolutional networks. IEEE Access, 11:28450–28461.