Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders

Ezgi Kara Timuçin

doi:10.54287/gujsa.1920474

Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders

Abstract

Some of the causes of irreversible visual impairment in the world include retinal diseases like choroidal neovascularization (CNV), diabetic macular edema (DME), and drusen deposits. Even though optical coherence tomography (OCT) is now regarded as the gold-standard of non-invasive imaging modality in the evaluation of the retina, the scarcity of expert annotations of OCT has been a severe bottleneck in supervised deep learning algorithms. In order to overcome this drawback, a few-shot learning system is suggested on the basis of Prototypical Networks with the pretrained ResNet18 encoder trained through episodic meta-learning. Class prototypes are calculated as average embedding of K support samples per class and then classification is performed using the Euclidean distance in a 512 dimensional embedding space. The experimentation is carried out using the publicly available benchmark of Kermany OCT2017 that contains 84495 images and four diagnostic groups CNV, DME, DRUSEN, and NORMAL. The proposed model achieves 99.33 ± 0.15% accuracy at 4-way 5-shot setting evaluated over 200 test episodes. Per-shot accuracies are 97.82 ± 0.36% (1-shot), 99.33 ± 0.15% (5-shot), and 99.18% ± 0.16% (10-shot). Per-class accuracies are 98.9% (CNV), 99.2% (DME), 99.0% (DRUSEN), and 99.3% (NORMAL). t-SNE, prototype space, and Grad-CAM analyses confirm clinically meaningful, well-separated embeddings directed toward pathology-specific retinal structures. These results demonstrate that Prototypical Networks with ImageNet-pretrained encoders surpass all prior few-shot methods on the Kermany benchmark without generative augmentation, achieving near-perfect classification from as few as one labeled example per class and constituting a clinically viable approach for automated retinal diagnosis in annotation-scarce settings.

Keywords

References

Agbareia, R., Omar, M., Zloto, O., Glicksberg, B. S., Nadkarni, G. N., & Klang, E. (2025). Multimodal LLMs for retinal disease diagnosis via OCT: Few-shot versus single-shot learning. Therapeutic Advances in Ophthalmology, 17. https://doi.org/10.1177/25158414251340569
Chen, W.-Y., Liu, Y.-C., Kira, Z., Wang, Y.-C. F., & Huang, J.-B. (2019). A closer look at few-shot classification. In: Proceedings of the Seventh International Conference on Learning Representations (ICLR 2019) (6-9 May 2019), New Orleans.
Finn, C., Abbeel, P., & Levine, S. (2017). Model-agnostic meta-learning for fast adaptation of deep networks. In: D. Precup, & Y. W. Teh (Eds.), Proceedings of the 34th International Conference on Machine Learning (ICML’17) (11-16 August 2017, vol. 70, pp. 1126-1135), Sydney NSW Australia.
Gholami, P., Roy, P., Parthasarathy, M. K., & Garg, V. (2020). OCTID: Optical coherence tomography image database. Computers and Electrical Engineering, 81, 106532. https://doi.org/10.1016/j.compeleceng.2019.106532
He, J., Wang, J., Han, Z., Ma, J., Wang, C., & Qi, M. (2023). An interpretable transformer network for the retinal disease classification using optical coherence tomography. Scientific Reports, 13, 3637. https://doi.org/10.1038/s41598-023-30853-z
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), (27-30 June 2016, 770-778), Las Vegas, NV, USA.
Kermany, D. S., Goldbaum, M., Cai, W., Valentim, C. C. S., Liang, H., Baxter, S. L., McKeown, A., Yang, G., Wu, X., Yan, F., Dong, J., Prasadha, M. K., Pei, J., Ting, M. Y. L., Zhu, J., Li, C., Hewett, S., Dong, J., Ziyar, I., … Zhang, K. (2018). Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell, 172(5), 1122-1131. https://doi.org/10.1016/j.cell.2018.02.010
Kulyabin, M., Zhdanov, A., Nikiforova, A., Stepichev, A., Kuznetsova, A., Ronkin, M., Borisov, V., Bogachev, A., Korotkich, S., Constable, P. A., & Maier, A. (2024). OCTDL: Optical coherence tomography dataset for image-based deep learning methods. Scientific Data, 11, 365. https://doi.org/10.1038/s41597-024-03182-7

Liu, X., Dong, X., Li, T., Zou, X., Cheng, C., Jiang, Z., Gao, Z., Duan, S., Chen, M., Liu, T., Huang, P., Li, D., & Lu, H. (2024). A difficulty-aware and task-augmentation method based on meta-learning for few-shot diabetic retinopathy classification. Quantitative Imaging in Medicine and Surgery, 14(1), 861-876. https://doi.org/10.21037/qims-23-567
Miladinović, A., Biscontin, A., Ajčević, M., Kresevic, S., Accardo, A., Marangoni, D., Tognetto, D., & Inferrera, L. (2024). Evaluating deep learning models for classifying OCT images with limited data and noisy labels. Scientific Reports, 14, 30321. https://doi.org/10.1038/s41598-024-81127-1
Morano, J., Fazekas, B., Sükei, E., Fecso, R., Emre, T., Gumpinger, M., Faustmann, G., Oghbaie, M., Schmidt-Erfurth, U., & Bogunović, H. (2025). Multimodal foundation model and benchmark for comprehensive retinal OCT image analysis. npj Digital Medicine, 8, 576. https://doi.org/10.1038/s41746-025-01852-3
Pan, H., Miao, J., Yu, J., Dong, J., Zhang, M., Wang, X., & Feng, J. (2025). A lightweight model for the retinal disease classification using optical coherence tomography. Biomedical Signal Processing and Control, 101, 107146. https://doi.org/10.1016/j.bspc.2024.107146
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-CAM: Visual explanations from deep networks via gradient-based localization. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV) (22-29 October 2017, pp. 618-626), Venice, Italy. https://doi.org/10.1109/ICCV.2017.74
Snell, J., Swersky, K., & Zemel, R. (2017). Prototypical networks for few-shot learning. In: U. von Luxburg, I. Guyon, S. Bengio, H. Wallach, & R. Fergus (Eds.), Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS'17) (4-9 December 2017, pp. 4080 - 4090), Long Beach California USA.
Sung, F., Yang, Y., Zhang, L., Xiang, T., Torr, P. H. S., & Hospedales, T. M. (2017). Learning to Compare: Relation Network for Few-Shot Learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (18-23 June 2018, pp. 1199-1208), Salt Lake City, UT, USA. https://doi.org/10.1109/CVPR.2018.00131
Tai, C.-Y., Chen, C.-W., Wu, C.-C., Chiu, B.-C., Lin, C.-H. D., Lu, C.-K., Wang, J.-K., & Huang, T.-L. (2025). Improvement Strategies for Few-Shot Learning in OCT Image Classification of Rare Retinal Diseases. In: Proceedings of the IEEE 14th Global Conference on Consumer Electronics (GCCE) (23-26 September 2025, pp. 489-490), Osaka, Japan. https://doi.org/10.1109/GCCE65946.2025.11275047
van der Maaten, L., & Hinton, G. (2008). Visualizing data using t-SNE. Journal of Machine Learning Research, 9(86), 2579-2605.
Vinyals, O., Blundell, C., Lillicrap, T. P., Kavukcuoglu, K., & Wierstra, D. (2016). Matching networks for one shot learning. In: D. D. Lee, U. von Luxburg, R. Garnett, M. Sugiyama, & I. Guyon (Eds.), Proceedings of the 30th International Conference on Neural Information Processing Systems (NIPS'16) (5-10 December 2016, pp. 3637-3645), Barcelona Spain.
Yoo, T. K., Choi, J. Y., & Kim, H. K. (2021). Feasibility study to improve deep learning in OCT diagnosis of rare retinal diseases with few-shot classification. Medical & Biological Engineering & Computing, 59(2), 401-415. https://doi.org/10.1007/s11517-021-02321-1
Zhao, Q., Mai, S. W., Li, Q., Huang, G. C., Gao, M. C., Yang, W. L., Wang, G., Ma, Y., Li, L., & Peng, X. Y. (2023). Automated Classification of Inherited Retinal Diseases in Optical Coherence Tomography Images Using Few-shot Learning. Biomedical and Environmental Sciences, 36(5), 431-440. https://doi.org/10.3967/bes2023.052

Details

Primary Language

English

Subjects

Machine Learning (Other)

Journal Section

Research Article

Authors

Ezgi Kara Timuçin ^*
0000-0003-1596-3427
Türkiye

Publication Date

June 30, 2026

Submission Date

April 1, 2026

Acceptance Date

June 5, 2026

Published in Issue

Year 2026 Volume: 13 Number: 2

DOI

https://doi.org/10.54287/gujsa.1920474

IZ

https://izlik.org/JA34FZ37AC

Cite

RIS / Bibtex

APA

Kara Timuçin, E. (2026). Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders. Gazi University Journal of Science Part A: Engineering and Innovation, 13(2), 784-797. https://doi.org/10.54287/gujsa.1920474

AMA

1.Kara Timuçin E. Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders. GU J Sci, Part A. 2026;13(2):784-797. doi:10.54287/gujsa.1920474

Chicago

Kara Timuçin, Ezgi. 2026. “Few-Shot Classification of Retinal OCT Images Using Prototypical Networks With Pretrained Encoders”. Gazi University Journal of Science Part A: Engineering and Innovation 13 (2): 784-97. https://doi.org/10.54287/gujsa.1920474.

EndNote

Kara Timuçin E (June 1, 2026) Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders. Gazi University Journal of Science Part A: Engineering and Innovation 13 2 784–797.

IEEE

[1]E. Kara Timuçin, “Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders”, GU J Sci, Part A, vol. 13, no. 2, pp. 784–797, June 2026, doi: 10.54287/gujsa.1920474.

ISNAD

Kara Timuçin, Ezgi. “Few-Shot Classification of Retinal OCT Images Using Prototypical Networks With Pretrained Encoders”. Gazi University Journal of Science Part A: Engineering and Innovation 13/2 (June 1, 2026): 784-797. https://doi.org/10.54287/gujsa.1920474.

JAMA

1.Kara Timuçin E. Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders. GU J Sci, Part A. 2026;13:784–797.

MLA

Kara Timuçin, Ezgi. “Few-Shot Classification of Retinal OCT Images Using Prototypical Networks With Pretrained Encoders”. Gazi University Journal of Science Part A: Engineering and Innovation, vol. 13, no. 2, June 2026, pp. 784-97, doi:10.54287/gujsa.1920474.

Vancouver

1.Ezgi Kara Timuçin. Few-Shot Classification of Retinal OCT Images Using Prototypical Networks with Pretrained Encoders. GU J Sci, Part A. 2026 Jun. 1;13(2):784-97. doi:10.54287/gujsa.1920474