Progressive Transfer Learning in Hybrid CNN-Transformer Architectures for Robust Diabetic Retinopathy Grading
DOI:
https://doi.org/10.61467/2007.1558.2027.v18i1.1532Keywords:
Diabetic Retinopathy, Medical Image Classification, Progressive Transfer Learning, Hybrid Neural Networks, Deep Learning, Vision Transformers, Artificial Intelligence in Medicine, Automated Triage, Retinopatía diabética, clasificación de imágenes médicas, redes neuronales híbridas, aprendizaje profundo, inteligencia artificial en medicinaAbstract
Automated diabetic retinopathy (DR) grading requires both local feature extraction and global spatial reasoning. While Convolutional Neural Networks and Vision Transformers (ViTs) struggle individually with spatial context and data limitations respectively, we propose an optimized hybrid framework (ProGrad-DR) combining an InceptionV3 backbone with a ViT encoder. To mitigate ViT overfitting on imbalanced data, we introduce a Progressive Transfer Learning regime, pre-training on a binary task before transferring to 5-class grading, and utilizing algorithmic class-weighting instead of synthetic oversampling. Trained on a composite dataset (APTOS/MESSIDOR) and independently tested on IDRiD, our model achieved a State-of-the-Art Quadratic Weighted Kappa of 0.9048 and a Balanced Accuracy of 0.7534. Notably, Mild DR sensitivity reached 80.0% with 99.4% healthy specificity. This stable, generalizable methodology successfully yields a clinically safe automated triage tool for multi-class DR grading across diverse settings.
Spanish-language metadata / Metadatos en español
Título en español:
Aprendizaje por transferencia progresiva en arquitecturas híbridas CNN-Transformer para una clasificación robusta de la retinopatía diabética
Resumen:
La clasificación automatizada de la retinopatía diabética (RD) requiere tanto la extracción de características locales como el razonamiento espacial global. Mientras que las redes neuronales convolucionales (CNN) y los Transformers de visión (ViT) presentan, de forma individual, limitaciones relacionadas con el contexto espacial y la disponibilidad de datos, respectivamente, proponemos un marco híbrido optimizado (ProGrad-DR) que combina una arquitectura base InceptionV3 con un codificador ViT.
Para mitigar el sobreajuste del ViT en datos desbalanceados, introducimos un esquema de aprendizaje por transferencia progresiva, basado en un preentrenamiento mediante una tarea binaria antes de transferir el modelo a una clasificación de cinco clases, utilizando además una ponderación algorítmica de clases en lugar de sobremuestreo sintético. Entrenado con un conjunto de datos compuesto (APTOS/MESSIDOR) y evaluado de forma independiente con IDRiD, nuestro modelo alcanzó un coeficiente kappa ponderado cuadrático de 0.9048 y una exactitud balanceada de 0.7534. Cabe destacar que la sensibilidad para la RD leve alcanzó el 80.0 %, con una especificidad del 99.4 % para los casos sanos.
Esta metodología estable y generalizable permite obtener una herramienta de triaje automatizado clínicamente segura para la clasificación multiclase de la RD en diversos entornos.
Palabras Claves:
Retinopatía diabética, clasificación de imágenes médicas, aprendizaje por transferencia progresiva, redes neuronales híbridas, aprendizaje profundo, Transformers de visión, inteligencia artificial en medicina, triaje automatizado.
Smart citations:
SciteAI.
Dimensions.
Open Alex.
References
Abràmoff, M. D., Garvin, M. K., & Sonka, M. (2010). Retinal imaging and image analysis. IEEE Reviews in Biomedical Engineering, 3, 169–208. https://doi.org/10.1109/RBME.2010.2084567
Karthik, Maggie, & Dane, S. (2019). APTOS 2019 blindness detection [Data set]. Kaggle. https://www.kaggle.com/competitions/aptos2019-blindness-detection
Ashwini, K., & Dash, R. (2023). Grading diabetic retinopathy using multiresolution based CNN. Biomedical Signal Processing and Control, 86, Article 105210. https://doi.org/10.1016/j.bspc.2023.105210
Azad, R., Kazerouni, A., Heidari, M., Aghdam, E. K., Molaei, A., Jia, Y., Jose, A., Roy, R., & Merhof, D. (2024). Advances in medical image analysis with vision transformers: A comprehensive review. Medical Image Analysis, 91, Article 103000. https://doi.org/10.1016/j.media.2023.103000
Banerjee, T., Singh, D. P., & Kour, P. (2026). Advances in deep neural, transformer learning, and kernel-based methods for diabetic retinopathy detection: A comprehensive review. Archives of Computational Methods in Engineering, 33, 2659–2707. https://doi.org/10.1007/s11831-025-10376-8
Bhardwaj, C., Jain, S., & Sood, M. (2021). Deep learning–based diabetic retinopathy severity grading system employing quadrant ensemble model. Journal of Digital Imaging, 34(2), 440–457. https://doi.org/10.1007/s10278-021-00418-5
Bhoyar, V., & Patel, M. (2026). A comprehensive review of deep learning approaches for automated detection, segmentation, and grading of diabetic retinopathy. Archives of Computational Methods in Engineering, 33, 4359–4380. https://doi.org/10.1007/s11831-025-10460-z
Cho, N. H., Shaw, J. E., Karuranga, S., Huang, Y., da Rocha Fernandes, J. D., Ohlrogge, A. W., & Malanda, B. (2018). IDF Diabetes Atlas: Global estimates of diabetes prevalence for 2017 and projections for 2045. Diabetes Research and Clinical Practice, 138, 271–281. https://doi.org/10.1016/j.diabres.2018.02.023
Chowdhury, A. R., Chatterjee, T., & Banerjee, S. (2019). A random forest classifier-based approach in the detection of abnormalities in the retina. Medical & Biological Engineering & Computing, 57(1), 193–203. https://doi.org/10.1007/s11517-018-1878-0
Decencière, E., Zhang, X., Cazuguel, G., Lay, B., Cochener, B., Trone, C., Gain, P., Ordonez, R., Massin, P., Erginay, A., Charton, B., & Klein, J.-C. (2014). Feedback on a publicly distributed image database: The Messidor database. Image Analysis & Stereology, 33(3), 231–234. https://doi.org/10.5566/ias.1155
Fong, D. S., Aiello, L., Gardner, T. W., King, G. L., Blankenship, G., Cavallerano, J. D., Ferris, F. L., III, Klein, R., & American Diabetes Association. (2004). Retinopathy in diabetes. Diabetes Care, 27(Suppl. 1), S84–S87. https://doi.org/10.2337/diacare.27.2007.s84
Girija, R., Deepa, N., & Chinni, S. K. (2025). CBAM-integrated deep neural networks for diabetic retinopathy detection with quadratic weighted kappa evaluation. In 2025 3rd International Conference on Advances in Computation, Communication and Information Technology (ICAICCIT) (pp. 304–309). IEEE. https://doi.org/10.1109/ICAICCIT68829.2025.11433973
Guariguata, L., Whiting, D., Weil, C., & Unwin, N. (2011). The International Diabetes Federation diabetes atlas methodology for estimating global and national prevalence of diabetes in adults. Diabetes Research and Clinical Practice, 94(3), 322–332. https://doi.org/10.1016/j.diabres.2011.10.040
Hagos, M. T., & Kant, S. (2019). Transfer learning based detection of diabetic retinopathy from small dataset [Preprint]. arXiv. https://doi.org/10.48550/arXiv.1905.07203
He, A., Li, T., Li, N., Wang, K., & Fu, H. (2021). CABNet: Category attention block for imbalanced diabetic retinopathy grading. IEEE Transactions on Medical Imaging, 40(1), 143–153. https://doi.org/10.1109/TMI.2020.3023463
Ikram, A., & Imran, A. (2025). ResViT FusionNet model: An explainable AI-driven approach for automated grading of diabetic retinopathy in retinal images. Computers in Biology and Medicine, 186, Article 109656. https://doi.org/10.1016/j.compbiomed.2025.109656
International Diabetes Federation. (2013). Five questions on the IDF Diabetes Atlas. Diabetes Research and Clinical Practice, 102(2), 147–148. https://doi.org/10.1016/j.diabres.2013.10.013
International Diabetes Federation. (2025). IDF Diabetes Atlas (11th ed.). https://diabetesatlas.org/resources/idf-diabetes-atlas-2025/
Khanna, M., Singh, L. K., Thawkar, S., & Goyal, M. (2023). Deep learning based computer-aided automatic prediction and grading system for diabetic retinopathy. Multimedia Tools and Applications, 82, 39255–39302. https://doi.org/10.1007/s11042-023-14970-5
Kwan, C. C., & Fawzi, A. A. (2019). Imaging and biomarkers in diabetic macular edema and diabetic retinopathy. Current Diabetes Reports, 19(10), Article 95. https://doi.org/10.1007/s11892-019-1226-2
Lechner, J., O’Leary, O. E., & Stitt, A. W. (2017). The pathology associated with diabetic retinopathy. Vision Research, 139, 7–14. https://doi.org/10.1016/j.visres.2017.04.003
Li, Y.-H., Yeh, N.-N., Chen, S.-J., & Chung, Y.-C. (2019). Computer-assisted diagnosis for diabetic retinopathy based on fundus images using deep convolutional neural network. Mobile Information Systems, 2019, Article 6142839. https://doi.org/10.1155/2019/6142839
Liu, T., Chen, Y., Shen, H., Zhou, R., Zhang, M., Liu, T., & Liu, J. (2021). A novel diabetic retinopathy detection approach based on deep symmetric convolutional neural network. IEEE Access, 9, 160552–160558. https://doi.org/10.1109/ACCESS.2021.3131630
Liu, Y., Yao, D., Ma, Y., Wang, H., Wang, J., Bai, X., Zeng, G., & Liu, Y. (2025). STMF-DRNet: A multi-branch fine-grained classification model for diabetic retinopathy using Swin-TransformerV2. Biomedical Signal Processing and Control, 103, Article 107352. https://doi.org/10.1016/j.bspc.2024.107352
Majumder, S., & Kehtarnavaz, N. (2021). Multitasking deep learning model for detection of five stages of diabetic retinopathy. IEEE Access, 9, 123220–123230. https://doi.org/10.1109/ACCESS.2021.3109240
Mansour, R. F. (2018). Deep-learning-based automatic computer-aided diagnosis system for diabetic retinopathy. Biomedical Engineering Letters, 8, 41–57. https://doi.org/10.1007/s13534-017-0047-y
Mazlan, N., Yazid, H., Arof, H., & Isa, H. (2020). Automated microaneurysms detection and classification using multilevel thresholding and multilayer perceptron. Journal of Medical and Biological Engineering, 40(2), 292–306. https://doi.org/10.1007/s40846-020-00509-8
Nazih, W., Aseeri, A. O., Atallah, O. Y., & El-Sappagh, S. (2023). Vision transformer model for predicting the severity of diabetic retinopathy in fundus photography-based retina images. IEEE Access, 11, 117546–117561. https://doi.org/10.1109/ACCESS.2023.3326528
Porwal, P., Pachade, S., Kamble, R., Kokare, M., Deshmukh, G., Sahasrabuddhe, V., & Meriaudeau, F. (2018). Indian Diabetic Retinopathy Image Dataset (IDRiD): A database for diabetic retinopathy screening research. Data, 3(3), Article 25. https://doi.org/10.3390/data3030025
Reguant, R., Brunak, S., & Saha, S. (2021). Understanding inherent image features in CNN-based assessment of diabetic retinopathy. Scientific Reports, 11, Article 9704. https://doi.org/10.1038/s41598-021-89225-0
Sambyal, N., Saini, P., Syal, R., & Gupta, V. (2020). Modified U-Net architecture for semantic segmentation of diabetic retinopathy images. Biocybernetics and Biomedical Engineering, 40(3), 1094–1109. https://doi.org/10.1016/j.bbe.2020.05.006
Seoud, L., Chelbi, J., & Cheriet, F. (2015). Automatic grading of diabetic retinopathy on a public database. In X. Chen, M. K. Garvin, J. J. Liu, E. Trucco, & Y. Xu (Eds.), Proceedings of the Ophthalmic Medical Image Analysis Second International Workshop (OMIA 2015) (pp. 97–104). https://doi.org/10.17077/omia.1032
Sinclair, A., Saeedi, P., Kaundal, A., Karuranga, S., Malanda, B., & Williams, R. (2020). Diabetes and global ageing among 65–99-year-old adults: Findings from the International Diabetes Federation Diabetes Atlas, 9th edition. Diabetes Research and Clinical Practice, 162, Article 108078. https://doi.org/10.1016/j.diabres.2020.108078
Skouta, A., Elmoufidi, A., Jai-Andaloussi, S., & Ouchetto, O. (2022). Hemorrhage semantic segmentation in fundus images for the diagnosis of diabetic retinopathy by using a convolutional neural network. Journal of Big Data, 9, Article 78. https://doi.org/10.1186/s40537-022-00632-0
Usman, T. M., Saheed, Y. K., Ignace, D., & Nsang, A. (2023). Diabetic retinopathy detection using principal component analysis multi-label feature extraction and classification. International Journal of Cognitive Computing in Engineering, 4, 78–88. https://doi.org/10.1016/j.ijcce.2023.02.002
Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30, 5998–6008.
Wang, Y.-B., Zhu, C.-Z., Yan, Q.-F., & Liu, L.-Q. (2016). A novel vessel segmentation in fundus images based on SVM. In 2016 International Conference on Information System and Artificial Intelligence (ISAI) (pp. 390–394). IEEE. https://doi.org/10.1109/ISAI.2016.0089
Watkins, P. J. (2003). Retinopathy. BMJ, 326(7395), 924–926. https://doi.org/10.1136/bmj.326.7395.924
Wilkinson, C. P., Ferris, F. L., III, Klein, R. E., Lee, P. P., Agardh, C. D., Davis, M., Dills, D., Kampik, A., Pararajasegaram, R., Verdaguer, J. T., & Global Diabetic Retinopathy Project Group. (2003). Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology, 110(9), 1677–1682. https://doi.org/10.1016/S0161-6420(03)00475-5
Yi, S., Ren, Y., & Shao, D. (2026). CTANet: A hybrid CNN and ViT network with adaptive focus fusion for diabetic retinopathy grading. Biomedical Signal Processing and Control, 113, Article 109106. https://doi.org/10.1016/j.bspc.2025.109106
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 International Journal of Combinatorial Optimization Problems and Informatics
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
