Destilação de Conhecimento Evidencial Adaptativa Empregada na Classificação de Imagens da Cavidade Oral
Resumo
O câncer oral apresenta taxas de sobrevivência superiores a 80% quando diagnosticado precocemente; contudo, esse índice cai para menos de 20% em estágios avançados. Apesar dos avanços recentes indicarem elevado desempenho na classificação de lâminas digitais, a adoção clínica dessas abordagens ainda é limitada por desafios metodológicos relevantes. Este trabalho propõe uma metodologia A-EKD para o diagnóstico confiável de OSCC via destilação de conhecimento evidencial. Introduz-se o Gate (G) para filtrar incertezas do Professor, protegendo o aprendizado do Aluno MobileNetV3 em amostras ruidosas. Sob rigor experimental para evitar vazamento de dados, o modelo atingiu 90,98% de acurácia, superando o estado da arte de 87,1%. A análise estatística mostra um potencial da classificação embasado por um Odds Ratio de 24,5 e uma Err-AUC de 0,8490, validando a premissa de filtrar incertezas. O método mitiga a superconfiança e otimiza a calibração, viabilizando diagnósticos confiáveis através de redes leves, que possuem um menor número de parâmetros.Referências
Agresti, A. (2012). Categorical Data Analysis, volume 792. John Wiley & Sons, 3rd edition.
Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1):1–3.
Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. (2017). On calibration of modern neural networks. In International Conference on Machine Learning (ICML), pages 1321–1330.
Hendrycks, D. and Gimpel, K. (2017). A baseline for detecting misclassified and out-of-distribution examples in neural networks. International Conference on Learning Representations (ICLR).
Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.
Howard, A. et al. (2019). Searching for mobilenetv3. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 1314–1324.
Lambert, B., Forbes, F., Doyle, S., H., D., and Dojat, M. (2024). Trustworthy clinical ai solutions: A unified review of uncertainty quantification in deep learning models for medical image analysis. Artificial Intelligence in Medicine, 150:102830.
McNemar, Q. (1947). Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2):153–157.
Nogueira, M. and Gomes, E. F. (2025). Histopathological imaging dataset for oral cancer analysis: A study with a data leakage warning. In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies, pages 811–818.
Paraíso, E. and Machado, A. (2025). Impacto do balanceamento e regularização na segmentação semântica de imagens histopatológicas. In Anais Estendidos do XXV Simpósio Brasileiro de Computação Aplicada à Saúde, pages 13–18, Porto Alegre, RS, Brasil. SBC.
Prajwal, R., Pawan, S. J., Nazarian, S., Heller, N., Weight, C. J., Duddalwar, V., and Kuo, C.-C. J. (2025). A study on energy consumption in ai-driven medical image segmentation. Journal of Imaging, 11(6).
Rahman, T. Y., Mahanta, L. B., Das, A. K., and Sarma, J. D. (2020). Histopathological imaging database for oral cancer analysis. Data in Brief, 29:105114.
Sensoy, M., Kaplan, L., and Kandemir, M. (2018). Evidential deep learning to quantify classification uncertainty. In Advances in Neural Information Processing Systems (NeurIPS), volume 31.
Tan, M. and Le, Q. (2019). Efficientnet: Rethinking model scaling for convolutional neural networks. arXiv preprint arXiv:1905.11946.
Van Rijsbergen, C. (1975). Information Retrieval. Butterworths.
WHO Classification of Tumours Editorial Board, editor (2024). Head and neck tumours, volume 9 of World Health Organization classification of tumours. IARC, Lyon, 5 edition.
Xiang, L., Gao, J., and Xu, C. (2025). Evidential knowledge distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV).
Zhang, J., Gao, Y., Liu, R., Cheng, X., Zhang, H., and Chen, S. (2025). Can students beyond the teacher? distilling knowledge from teacher’s bias. In Proceedings of the AAAI Conference on Artificial Intelligence, AAAI’25/IAAI’25/EAAI’25. AAAI Press.
Zhao, B. et al. (2022). Decoupled knowledge distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11953–11962.
Brier, G. W. (1950). Verification of forecasts expressed in terms of probability. Monthly Weather Review, 78(1):1–3.
Guo, C., Pleiss, G., Sun, Y., and Weinberger, K. Q. (2017). On calibration of modern neural networks. In International Conference on Machine Learning (ICML), pages 1321–1330.
Hendrycks, D. and Gimpel, K. (2017). A baseline for detecting misclassified and out-of-distribution examples in neural networks. International Conference on Learning Representations (ICLR).
Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531.
Howard, A. et al. (2019). Searching for mobilenetv3. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 1314–1324.
Lambert, B., Forbes, F., Doyle, S., H., D., and Dojat, M. (2024). Trustworthy clinical ai solutions: A unified review of uncertainty quantification in deep learning models for medical image analysis. Artificial Intelligence in Medicine, 150:102830.
McNemar, Q. (1947). Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2):153–157.
Nogueira, M. and Gomes, E. F. (2025). Histopathological imaging dataset for oral cancer analysis: A study with a data leakage warning. In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies, pages 811–818.
Paraíso, E. and Machado, A. (2025). Impacto do balanceamento e regularização na segmentação semântica de imagens histopatológicas. In Anais Estendidos do XXV Simpósio Brasileiro de Computação Aplicada à Saúde, pages 13–18, Porto Alegre, RS, Brasil. SBC.
Prajwal, R., Pawan, S. J., Nazarian, S., Heller, N., Weight, C. J., Duddalwar, V., and Kuo, C.-C. J. (2025). A study on energy consumption in ai-driven medical image segmentation. Journal of Imaging, 11(6).
Rahman, T. Y., Mahanta, L. B., Das, A. K., and Sarma, J. D. (2020). Histopathological imaging database for oral cancer analysis. Data in Brief, 29:105114.
Sensoy, M., Kaplan, L., and Kandemir, M. (2018). Evidential deep learning to quantify classification uncertainty. In Advances in Neural Information Processing Systems (NeurIPS), volume 31.
Tan, M. and Le, Q. (2019). Efficientnet: Rethinking model scaling for convolutional neural networks. arXiv preprint arXiv:1905.11946.
Van Rijsbergen, C. (1975). Information Retrieval. Butterworths.
WHO Classification of Tumours Editorial Board, editor (2024). Head and neck tumours, volume 9 of World Health Organization classification of tumours. IARC, Lyon, 5 edition.
Xiang, L., Gao, J., and Xu, C. (2025). Evidential knowledge distillation. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV).
Zhang, J., Gao, Y., Liu, R., Cheng, X., Zhang, H., and Chen, S. (2025). Can students beyond the teacher? distilling knowledge from teacher’s bias. In Proceedings of the AAAI Conference on Artificial Intelligence, AAAI’25/IAAI’25/EAAI’25. AAAI Press.
Zhao, B. et al. (2022). Decoupled knowledge distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11953–11962.
Publicado
01/06/2026
Como Citar
ABREU JÚNIOR, Carlos Alberto Matias de; RIBEIRO, Thiago Pirola; NASCIMENTO, Marcelo Zanchetta do.
Destilação de Conhecimento Evidencial Adaptativa Empregada na Classificação de Imagens da Cavidade Oral. In: SIMPÓSIO BRASILEIRO DE COMPUTAÇÃO APLICADA À SAÚDE (SBCAS), 26. , 2026, Ouro Preto/MG.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 383-394.
ISSN 2763-8952.
DOI: https://doi.org/10.5753/sbcas.2026.21038.
