Impact of Segmentation Techniques on the Performance of Automatic Classification of Dermatological Lesions

• Lucas Silva de Souza UFERSA
• Mikael Johnatan da Silva Azevedo UFERSA
• Angélica Félix de Castro UFERSA
• Amanda Gondim de Oliveira UFERSA

DOI: https://doi.org/10.5753/eniac.2026.31327

Resumo

Este trabalho avalia o impacto de técnicas de segmentação na classificação multimodal de lesões dermatológicas. Foram testados 14 cenários na base HAM10000, combinando quatro métodos clássicos e duas U-Nets com os classificadores ResNet-50 e EfficientNet-B3. Os testes revelaram que o sucesso da segmentação depende do backbone utilizado. Na ResNet-50, o método de Otsu obteve desempenho equivalente às redes profundas, eliminando custos de treinamento. Na EfficientNet-B3, os métodos clássicos degradaram a rede, justificando o uso da U-Net de 5 Blocos, que alcançou o maior Macro F1-Score (0,62) ao mitigar o desbalanceamento de classes.

Referências

Alenezi, F., Armghan, A., e Polat, K. (2023). A Novel Multi-Task Learning Network Based on Melanoma Segmentation and Classification with Skin Lesion Images. In Diagnostics, 13(2), 262. DOI: 10.3390/diagnostics13020262.

Anand, V., Gupta, S., Koundal, D., & Singh, K. (2023). Fusion of U-Net and CNN model for segmentation and classification of skin lesion from dermoscopy images. In Expert Systems with Applications, 213, 119230. DOI: 10.1016/j.eswa.2022.119230

Biswas, S; Raza, S. M. M. e Nguyen, T. U. (2025) Advancing skin lesion classification: the role of SAM-based segmentation in enhancing convolutional neural network performance. In SPIE Digital Library 13605 - Applications of Digital Image Processing XLVIII. DOI: 10.1117/12.3065735.

Davis, L. E.; Shalin, S. C. e Tackett, A. J. (2019). Current state of melanoma diagnosis and treatment. In Cancer Biology & Therapy, p. 1366-1379. Taylor and Francis. DOI: 10.1080/15384047.2019.1640032.

Hasan, M. K., Elahi, M. T. E., Alam, M. A., Jawad, M. T., & Martí, R. (2022). DermoExpert: Skin lesion classification using a hybrid convolutional neural network through segmentation, transfer learning, and augmentation. In Informatics in Medicine Unlocked, 28. DOI: 10.1016/j.imu.2021.100819

Khan, M. A., Sharif, M., Akram, T., Damaševičius, R., & Maskeliūnas, R. (2021). Skin Lesion Segmentation and Multiclass Classification Using Deep Learning Features and Improved Moth Flame Optimization. In Diagnostics, 11(5), 811. DOI: 10.3390/diagnostics11050811

Mahbod, A.; Tschandl, P.; Langs, G.; Ecker, R. e Ellinger, I. (2020). The effects of skin lesion segmentation on the performance of dermatoscopic image classification. In Computer Methods and Programs in Biomedicine, p. 1-12. Elsevier - Science Direct. DOI: 10.1016/j.cmpb.2020.105725

Naik, P. P. (2021). Cutaneous Malignant Melanoma: A Review of Early Diagnosis and Management. In World Journal of Oncology, p. 7-19. Elmer Press. DOI: 10.14740/wjon1349

Saginala, K.; Barsouk, A.; Aluru, J. S. e Rawla, P. (2021). Epidemiology of Melanoma.

In Medical Sciences. MDPI. DOI: 10.3390/medsci9040063

Singh, L; Janghel, R. R. e Sahu, S. P. (2022). An Empirical Review on Evaluating the Impact of Image Segmentation on the Classification Performance for Skin Lesion Detection. In IETE Technical Review. Taylor e Francis. DOI: 10.1080/02564602.2022.2068681

Xie, Y; Zhang, J., Xia, Y. e Shen, C. (2020). A Mutual Bootstrapping Model for Automated Skin Lesion Segmentation and Classification. In IEEE Transactions on Medical Imaging, vol. 39, no. 7, pp. 2482-2493, July 2020, DOI: 10.1109/TMI.2020.2972964.

Waseh, S. e Lee, J. B. (2023). Advances in melanoma: epidemiology, diagnosis, and prognosis. In Frontiers in Medicine, p. 1-14. DOI: 10.3389/fmed.2023.1268479.

Wong, K. C. L.; Syeda-Mahmood, T. e Moradi, M. (2018). Building medical image classifiers with very limited data using segmentation networks. In Medical Image Analysis. DOI: 10.1016/j.media.2018.07.010,