Wavelet Skip U-Net: High-Frequency Enhancement for Diabetic Retinopathy Lesion Segmentation

• Lucas Araújo Gonçalves UFMA
• Geraldo Braz Junior UFMA
• João Dallyson Sousa de Almeida UFMA
• Marcos Melo Ferreira UFMA

DOI: https://doi.org/10.5753/sbcas.2026.21515

Abstract

Diabetic Retinopathy (DR) is one of the leading causes of vision loss in diabetic patients, making the early detection of lesions such as exudates and hemorrhages essential. This work proposes a deep learning architecture based on U-Net, incorporating a pre-trained EfficientNet-B4 encoder and a Wavelet Enhancement Module integrated into the first skip connection. The proposed approach uses two-dimensional discrete wavelet decomposition (2D DWT—Haar) to capture edge and texture details, thereby facilitating the segmentation of small lesions. Experiments conducted on the IDRiD dataset demonstrated that including of the wavelet module improved the Dice coefficient for the hemorrhage class from 62,3% to 65,1% on the test set, outperforming classical architectures such as the standard U-Net and DeepLabV3+.

References

Almattar, W., Luqman, H., and Khan, F. A. (2023). Diabetic retinopathy grading review: Current techniques and future directions. Image and Vision Computing, 139:104821.

Chagas, T. A., Dos Reis, M. A., Leivas, G., Santos, L. P., Gossenheimer, A. N., Melo, G. B., Malerbi, F. K., and Schaan, B. D. (2023). Prevalence of diabetic retinopathy in brazil: a systematic review with meta-analysis. Diabetology & Metabolic Syndrome, 15(1):34.

Dias, M., Santos, C., Aguiar, M., Welfer, D., Pereira, A., and Ribeiro, M. (2023). Um novo método baseado em detector de dois estágios para segmentação de instância de lesões retinianas usando o modelo mask r-cnn e a biblioteca detectron2. In Seminário Integrado de Software e Hardware (SEMISH), pages 1–12. SBC.

Fleiss, J. L., Levin, B., Paik, M. C., et al. (1981). The measurement of interrater agreement. Statistical methods for rates and proportions, 2(212-236):22–23.

Gaikwad, A. and Gautam, A. (2024). Segmentation of hard exudates and hemorrhages from diabetic retinopathy images using residual u-net with squeeze and excite blocks. In 2024 IEEE International Conference on Image Processing (ICIP), pages 3064–3069. IEEE.

Kaizer, A. C. and Machado, A. M. C. (2025). Aplicação de técnicas de aprendizado profundo na classificação da retinopatia diabética. In Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), pages 49–54. SBC.

Lin, T.-Y., Goyal, P., Girshick, R., He, K., and Dollár, P. (2017). Focal loss for dense object detection. In Proceedings of the IEEE international conference on computer vision, pages 2980–2988.

Maji, D., Sigedar, P., and Singh, M. (2022). Attention res-unet with guided decoder for semantic segmentation of brain tumors. Biomedical Signal Processing and Control, 71:103077.

Ministério da Saúde (2025). Vigitel Brasil 2006-2024: vigilância de fatores de risco e proteção para doenças crônicas por inquérito telefônico: estimativas sobre frequência e distribuição sociodemográfica de fatores de risco e proteção para doenças crônicas nas capitais dos 26 estados brasileiros e no Distrito Federal entre 2006 e 2024. Technical report, Brasil. Ministério da Saúde. Secretaria de Vigilância em Saúde e Ambiente. Departamento de Análise Epidemiológica e Vigilância de Doenças Não Transmissíveis, Brasília. Acesso em: 5 mar. 2026.

Moreira, F., Schaan, B., Schneiders, J., Reis, M., Serpa, M., and Navaux, P. (2020). Impacto da resolução na detecção de retinopatia diabética com uso de deep learning. 494–499.

Pereira, A., Santos, C., Aguiar, M., Welfer, D., Dias, M., de Menezes, R., and Santana, D. (2024). Enhanced semantic segmentation of retinal microlesions through r2u-net architecture. In Simpósio Brasileiro de Computação Aplicada à Saúde (SBCAS), pages 13–24. SBC.

Porwal, P., Pachade, S., Kamble, R., Kokare, M., Deshmukh, G., Sahasrabuddhe, V., and Meriaudeau, F. (2018). Indian diabetic retinopathy image dataset (idrid): a database for diabetic retinopathy screening research. Data, 3(3):25.

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, pages 234–241. Springer.

Sambyal, N., Saini, P., Syal, R., and Gupta, V. (2020). Modified u-net architecture for semantic segmentation of diabetic retinopathy images. Biocybernetics and Biomedical Engineering, 40(3):1094–1109.

Tan, M. and Le, Q. (2019). Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, pages 6105–6114. PMLR.

Tan, T.-E. and Wong, T. Y. (2023). Diabetic retinopathy: Looking forward to 2030. Frontiers in endocrinology, 13:1077669.

Ullah, Z., Usman, M., Latif, S., Khan, A., and Gwak, J. (2023). Ssmd-unet: Semi-supervised multi-task decoders network for diabetic retinopathy segmentation. Scientific Reports, 13(1):9087.

Wong, D. C. S., Kiew, G., Jeon, S., and Ting, D. (2021). Singapore eye lesions analyzer (selena): the deep learning system for retinal diseases. In Artificial intelligence in ophthalmology, pages 177–185. Springer.

Wu, J., Li, J., Yang, J., and Mei, S. (2025). Wavelet-integrated deep neural networks: A systematic review of applications and synergistic architectures. Neurocomputing, page 131648.

Xue, J., Yan, S., Qu, J., Qi, F., Qiu, C., Zhang, H., Chen, M., Liu, T., Li, D., and Liu, X. (2019). Deep membrane systems for multitask segmentation in diabetic retinopathy. Knowledge-Based Systems, 183:104887.