FRCS-Net: Superando a Cauda Longa em Radiografias de Tórax Via Aprendizado em Dois Estágios e Ranking Sensível ao Custo
Resumo
A radiografia de tórax é fundamental para triagem, mas o desbalanceamento severo de classes em bases de dados públicas prejudica a detecção de patologias raras. Abordagens baseadas em entropia cruzada ignoram a cauda longa, enquanto técnicas de reamostragem introduzem ruído em cenários multirrótulo. Este trabalho propõe a Focal-Ranking Cost-Sensitive (FRCS) Loss, uma função de perda híbrida desenhada para priorizar a severidade clínica. Utilizou-se uma arquitetura DenseNet-121 em protocolo de treinamento de dois estágios (estabilização via perda assimétrica e refinamento de ordenação), validado em um subconjunto curado da base NIH ChestX-ray14. Os experimentos demonstram que o método supera o modelo base, elevando o F1-Score da classe hérnia (a mais rara) de 0,00 para 0,63, sem degradação significativa na especificidade global. A abordagem confirma que a imposição de restrições de ordenação por pares é superior à classificação tradicional para garantir a segurança clínica em diagnósticos de cauda longa.
Referências
Chen, K., Lei, W., Zhao, S., et al. (2023). PCCT: Progressive Class-Center Triplet Loss for Imbalanced Medical Image Classification. IEEE Journal of Biomedical and Health Informatics, 27(4):2026–2036. DOI: 10.1109/JBHI.2023.3240136.
Ciello, A. D., Franchi, P., Contegiacomo, A., et al. (2017). Missed lung cancer: when, where, and why? Diagnostic and Interventional Radiology, 23(2):118–126. DOI: 10.5152/dir.2016.16187.
Gefter, W. B. and Hatabu, H. (2023). Reducing Errors Resulting From Commonly Missed Chest Radiography Findings. Chest, 163(3):634–649. DOI: 10.1016/j.chest.2022.12.003.
Gefter, W. B., Post, B. A., and Hatabu, H. (2023). Commonly Missed Findings on Chest Radiographs. Chest, 163(3):650–661. DOI: 10.1016/j.chest.2022.10.039.
Gonçalves, J. V. L., de Souza, D. V., dos Santos, C. I. A. T., do Nascimento, C. E. T., da Cruz, L. B., Junior, D. A. D., and Diniz, J. O. B. (2024). D.IAgnóstica: Ferramenta CADx para diagnóstico de doenças pulmonares em imagens radiológicas. In Anais do XXIV Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), Porto Alegre, RS, Brasil. SBC.
Hanif, M. S., Bilal, M., Alsaggaf, A. H., et al. (2025). Enhancing Multi-Label Chest X-Ray Classification Using an Improved Ranking Loss. Bioengineering, 12(6):593. DOI: 10.3390/bioengineering12060593.
Hu, Z., Hongyu, C., Mei, W., and Wang, Y. (2024). Fetransnet: An Enhanced Lung Disease Classification Approach Combining Efficientnet and Transformer with Adaptive Focal Loss. DOI: 10.2139/ssrn.4884895.
Instituto Nacional de Câncer (2022). Estimativa 2023: incidência de câncer no brasil. Disponível em: [link]. Acesso em: 18 fev. 2026.
Kim, D. (2023). CheXFusion: Effective Fusion of Multi-View Features using Transformers for Long-Tailed Chest X-Ray Classification. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2694–2702, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00285.
Lai, H., Yao, Q., He, Z., et al. (2024). Long-Tailed Multi-Label Classification with Noisy Label of Thoracic Diseases from Chest X-Ray. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5, Athens, Greece. IEEE. DOI: 10.1109/ISBI56570.2024.10635361.
Lamoureux, C., Hanna, T. N., Sprecher, D., et al. (2021). Radiologist errors by modality, anatomic region, and pathology for 1.6 million exams: what we have learned. Emergency Radiology, 28(6):1135–1141. DOI: 10.1007/s10140-021-01959-6.
Li, X., Xu, X., Liu, Y., and Zhao, X. (2024). CheX-DS: Improving Chest X-ray Image Classification with Ensemble Learning Based on DenseNet and Swin Transformer. In 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 5295–5301, Lisbon, Portugal. IEEE. DOI: 10.1109/BIBM62325.2024.10822262.
Lin, Y.-C. and Chen, Y.-S. (2026). Weighted Stratification in Multi-label Contrastive Learning for Long-Tailed Medical Image Classification. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, pages 677–687, Cham. Springer Nature Switzerland. DOI: 10.1007/978-3-032-05169-1.
Liu, Z., Cheng, Y., and Tamura, S. (2023). Multi-Label Local to Global Learning: A Novel Learning Paradigm for Chest X-Ray Abnormality Classification. IEEE Journal of Biomedical and Health Informatics, 27(9):4409–4420. DOI: 10.1109/JBHI.2023.3281466.
Marques, A. G. M. and Machado, A. M. C. (2023). Convolutional neural networks with approximation of Shapley values for the classification and interpretation of pneumonia in X-ray images. In Anais do XXIII Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), Porto Alegre, RS, Brasil. SBC.
Nguyen-Mau, T.-H., Huynh, T.-L., Le, T.-D., et al. (2023). Advanced Augmentation and Ensemble Approaches for Classifying Long-Tailed Multi-Label Chest X-Rays. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2721–2730, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00288.
Park, W., Park, I., Kim, S., et al. (2023). Robust Asymmetric Loss for Multi-Label Long-Tailed Learning. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2703–2712, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00286.
Prinster, D., Mahmood, A., Saria, S., et al. (2024). Care to Explain? AI Explanation Types Differentially Impact Chest Radiograph Diagnostic Performance and Physician Trust in AI. Radiology, 313(2):e233261. DOI: 10.1148/radiol.233261.
Ranjan, N., Balkhande, B., Tembhare, L., et al. (2025). Addressing diagnostic resource imbalance in pulmonary tuberculosis detection from chest radiographs through cost-aware learning. Indian Journal of Tuberculosis, 72:S69–S76. DOI: 10.1016/j.ijtb.2025.10.011.
Ridnik, T., Ben-Baruch, E., Zamir, N., et al. (2021). Asymmetric Loss For Multi-Label Classification. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 82–91, Montreal, QC, Canada. IEEE. DOI: 10.1109/ICCV48922.2021.00015.
Sujay, J. K., Surakshith, D. T., Uday, T. Y., et al. (2024). Hybrid Approach for Handling Class Imbalance on Medical Data. In 2024 International Conference on Data Science and Network Security (ICDSNS), pages 1–6, Tiptur, India. IEEE. DOI: 10.1109/ICDSNS62112.2024.10691062.
Tanno, R., Barrett, D. G. T., Sellergren, A., et al. (2025). Collaboration between clinicians and vision–language models in radiology report generation. Nature Medicine, 31(2):599–608. DOI: 10.1038/s41591-024-03302-1.
Tsaniya, H., Fatichah, C., and Suciati, N. (2023). Comparison of sampling methods for handling imbalance data in deep learning-based predictions of chest X-ray abnormality tags. In 2023 the 7th International Conference on Medical and Health Informatics (ICMHI), pages 6–10, Kyoto, Japan. ACM. DOI: 10.1145/3608298.3608300.
Tyagi, M., Roy, S., and Bansal, V. (2022). Custom Weighted Balanced Loss function for Covid 19 Detection from an Imbalanced CXR Dataset. In 2022 26th International Conference on Pattern Recognition (ICPR), pages 2707–2713, Montreal, QC, Canada. IEEE. DOI: 10.1109/ICPR56361.2022.9956580.
Vimala, A. G., Legapriyadharshini, N., Dhanalakshmi, R., et al. (2024). Cost Sensitive Learning using Chest X-Ray with CNN for Covid-19 Detection with Lung Diseases which Lead to Class Imbalance. In 2024 5th International Conference on Image Processing and Capsule Networks (ICIPCN), pages 489–495, Dhulikhel, Nepal. IEEE. DOI: 10.1109/ICIPCN63822.2024.00086.
Zhang, M., Hu, X., Gu, L., et al. (2025). A New Benchmark: Clinical Uncertainty and Severity Aware Labeled Chest X-Ray Images With Multi-Relationship Graph Learning. IEEE Transactions on Medical Imaging, 44(1):338–347. DOI: 10.1109/TMI.2024.3441494.
Ciello, A. D., Franchi, P., Contegiacomo, A., et al. (2017). Missed lung cancer: when, where, and why? Diagnostic and Interventional Radiology, 23(2):118–126. DOI: 10.5152/dir.2016.16187.
Gefter, W. B. and Hatabu, H. (2023). Reducing Errors Resulting From Commonly Missed Chest Radiography Findings. Chest, 163(3):634–649. DOI: 10.1016/j.chest.2022.12.003.
Gefter, W. B., Post, B. A., and Hatabu, H. (2023). Commonly Missed Findings on Chest Radiographs. Chest, 163(3):650–661. DOI: 10.1016/j.chest.2022.10.039.
Gonçalves, J. V. L., de Souza, D. V., dos Santos, C. I. A. T., do Nascimento, C. E. T., da Cruz, L. B., Junior, D. A. D., and Diniz, J. O. B. (2024). D.IAgnóstica: Ferramenta CADx para diagnóstico de doenças pulmonares em imagens radiológicas. In Anais do XXIV Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), Porto Alegre, RS, Brasil. SBC.
Hanif, M. S., Bilal, M., Alsaggaf, A. H., et al. (2025). Enhancing Multi-Label Chest X-Ray Classification Using an Improved Ranking Loss. Bioengineering, 12(6):593. DOI: 10.3390/bioengineering12060593.
Hu, Z., Hongyu, C., Mei, W., and Wang, Y. (2024). Fetransnet: An Enhanced Lung Disease Classification Approach Combining Efficientnet and Transformer with Adaptive Focal Loss. DOI: 10.2139/ssrn.4884895.
Instituto Nacional de Câncer (2022). Estimativa 2023: incidência de câncer no brasil. Disponível em: [link]. Acesso em: 18 fev. 2026.
Kim, D. (2023). CheXFusion: Effective Fusion of Multi-View Features using Transformers for Long-Tailed Chest X-Ray Classification. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2694–2702, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00285.
Lai, H., Yao, Q., He, Z., et al. (2024). Long-Tailed Multi-Label Classification with Noisy Label of Thoracic Diseases from Chest X-Ray. In 2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5, Athens, Greece. IEEE. DOI: 10.1109/ISBI56570.2024.10635361.
Lamoureux, C., Hanna, T. N., Sprecher, D., et al. (2021). Radiologist errors by modality, anatomic region, and pathology for 1.6 million exams: what we have learned. Emergency Radiology, 28(6):1135–1141. DOI: 10.1007/s10140-021-01959-6.
Li, X., Xu, X., Liu, Y., and Zhao, X. (2024). CheX-DS: Improving Chest X-ray Image Classification with Ensemble Learning Based on DenseNet and Swin Transformer. In 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 5295–5301, Lisbon, Portugal. IEEE. DOI: 10.1109/BIBM62325.2024.10822262.
Lin, Y.-C. and Chen, Y.-S. (2026). Weighted Stratification in Multi-label Contrastive Learning for Long-Tailed Medical Image Classification. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, pages 677–687, Cham. Springer Nature Switzerland. DOI: 10.1007/978-3-032-05169-1.
Liu, Z., Cheng, Y., and Tamura, S. (2023). Multi-Label Local to Global Learning: A Novel Learning Paradigm for Chest X-Ray Abnormality Classification. IEEE Journal of Biomedical and Health Informatics, 27(9):4409–4420. DOI: 10.1109/JBHI.2023.3281466.
Marques, A. G. M. and Machado, A. M. C. (2023). Convolutional neural networks with approximation of Shapley values for the classification and interpretation of pneumonia in X-ray images. In Anais do XXIII Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), Porto Alegre, RS, Brasil. SBC.
Nguyen-Mau, T.-H., Huynh, T.-L., Le, T.-D., et al. (2023). Advanced Augmentation and Ensemble Approaches for Classifying Long-Tailed Multi-Label Chest X-Rays. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2721–2730, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00288.
Park, W., Park, I., Kim, S., et al. (2023). Robust Asymmetric Loss for Multi-Label Long-Tailed Learning. In 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), pages 2703–2712, Paris, France. IEEE. DOI: 10.1109/ICCVW60793.2023.00286.
Prinster, D., Mahmood, A., Saria, S., et al. (2024). Care to Explain? AI Explanation Types Differentially Impact Chest Radiograph Diagnostic Performance and Physician Trust in AI. Radiology, 313(2):e233261. DOI: 10.1148/radiol.233261.
Ranjan, N., Balkhande, B., Tembhare, L., et al. (2025). Addressing diagnostic resource imbalance in pulmonary tuberculosis detection from chest radiographs through cost-aware learning. Indian Journal of Tuberculosis, 72:S69–S76. DOI: 10.1016/j.ijtb.2025.10.011.
Ridnik, T., Ben-Baruch, E., Zamir, N., et al. (2021). Asymmetric Loss For Multi-Label Classification. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 82–91, Montreal, QC, Canada. IEEE. DOI: 10.1109/ICCV48922.2021.00015.
Sujay, J. K., Surakshith, D. T., Uday, T. Y., et al. (2024). Hybrid Approach for Handling Class Imbalance on Medical Data. In 2024 International Conference on Data Science and Network Security (ICDSNS), pages 1–6, Tiptur, India. IEEE. DOI: 10.1109/ICDSNS62112.2024.10691062.
Tanno, R., Barrett, D. G. T., Sellergren, A., et al. (2025). Collaboration between clinicians and vision–language models in radiology report generation. Nature Medicine, 31(2):599–608. DOI: 10.1038/s41591-024-03302-1.
Tsaniya, H., Fatichah, C., and Suciati, N. (2023). Comparison of sampling methods for handling imbalance data in deep learning-based predictions of chest X-ray abnormality tags. In 2023 the 7th International Conference on Medical and Health Informatics (ICMHI), pages 6–10, Kyoto, Japan. ACM. DOI: 10.1145/3608298.3608300.
Tyagi, M., Roy, S., and Bansal, V. (2022). Custom Weighted Balanced Loss function for Covid 19 Detection from an Imbalanced CXR Dataset. In 2022 26th International Conference on Pattern Recognition (ICPR), pages 2707–2713, Montreal, QC, Canada. IEEE. DOI: 10.1109/ICPR56361.2022.9956580.
Vimala, A. G., Legapriyadharshini, N., Dhanalakshmi, R., et al. (2024). Cost Sensitive Learning using Chest X-Ray with CNN for Covid-19 Detection with Lung Diseases which Lead to Class Imbalance. In 2024 5th International Conference on Image Processing and Capsule Networks (ICIPCN), pages 489–495, Dhulikhel, Nepal. IEEE. DOI: 10.1109/ICIPCN63822.2024.00086.
Zhang, M., Hu, X., Gu, L., et al. (2025). A New Benchmark: Clinical Uncertainty and Severity Aware Labeled Chest X-Ray Images With Multi-Relationship Graph Learning. IEEE Transactions on Medical Imaging, 44(1):338–347. DOI: 10.1109/TMI.2024.3441494.
Publicado
01/06/2026
Como Citar
PEDREIRA, Gustavo; FERNANDES, Henrique; ZANCHETTA, Marcelo.
FRCS-Net: Superando a Cauda Longa em Radiografias de Tórax Via Aprendizado em Dois Estágios e Ranking Sensível ao Custo. In: SIMPÓSIO BRASILEIRO DE COMPUTAÇÃO APLICADA À SAÚDE (SBCAS), 26. , 2026, Ouro Preto/MG.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 37-48.
ISSN 2763-8952.
DOI: https://doi.org/10.5753/sbcas.2026.20305.
