COMPUTER-AIDED DIAGNOSIS OF RETINAL DISEASES USING OPTICAL COHERENCE TOMOGRAPHY IMAGES AND DEEP LEARNING MODELS
Keywords
DOI:
https://doi.org/10.34238/tnu-jst.15358Abstract
This study evaluates nine deep learning models across three architecture families (standard Convolutional Neural Networks, modern Convolutional Neural Networks, Vision Transformers) for retinal optical coherence tomography image classification on the OCT2017 dataset. We fine-tune and benchmark these models under identical conditions across three dimensions: classification performance, lesion localization capability, and robustness to noise. This approach aims to determine whether the models truly learn pathological features or merely rely on spurious correlations. Experimental results demonstrate that ConvNeXt V2-T achieves the highest accuracy (95.45%), while ResNet50 yields the most precise lesion localization in the DME class (LLA 85.68%), and SwinV2-T demonstrates the greatest robustness against noise (with a macro-F1 drop of only 3.2%). Furthermore, the study reveals a clear trade-off among these architectures regarding classification accuracy, clinical interpretability, and robustness. These findings provide an empirical basis for selecting a suitable model based on specific practical requirements in clinical settings.Downloads
References
D. Huang et al., “Optical coherence tomography,” Science, vol. 254, no. 5035, pp. 1178-1181, 1991.
D. S. Kermany et al., “Identifying medical diagnoses and treatable diseases by image-based deep learn ing,” Cell, vol. 172, no. 5, pp. 1122-1131, 2018.
S. P. K. Karri, D. Chakraborty, and J. Chatterjee, “Transfer learning based classification of optical co herence tomography images with deep convolutional neural networks,” in Proc. IEEE EMBC, 2017, pp. 2033-2036.
R. Rasti, H. Rabbani, A. Mehridehnavi, and F. Hajizadeh, “Macular oct classification using a multi scale convolutional neural network ensemble,” IEEE Transactions on Medical Imaging, vol. 37, no. 4, pp. 1024-1034, 2017.
F. Li et al., “Deep learning-based automated detection of retinal diseases using optical coherence tomography images,” Biomedical Optics Express, vol. 10, no. 12, pp. 6204-6226, 2019.
S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” in Proc. ECCV, 2018, pp. 3-19.
Q. Wang, B. Wu, P. Zhu, P. Li, W. Zuo, and Q. Hu, “Eca-net: Efficient channel attention for deep convolutional neural networks,” in Proc. CVPR, 2020, pp. 11534-11542.
S. Woo et al., “Convnext v2: Co-designing and scaling convnets with masked autoencoders,” in Proc. CVPR, 2023, pp. 16133-16142.
A. Wang et al., “Repvit: Revisiting Mobile CNN from ViT Perspective,” in Proc. CVPR, 2024, pp. 15909-15920, doi: 10.1109/CVPR52733.2024.01506.
A. Dosovitskiy et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” in Proc. ICLR, 2021, doi: 10.48550/arXiv.2010.11929.
Z. Liu et al., “Swin transformer v2: Scaling up capacity and resolution,” in Proc. CVPR, 2022, pp. 12009-12019.
K. He, X. Chen, S. Xie, Y. Li, P. Dollár, and R. Girshick, “Masked autoencoders are scalable vision learners,” in Proc. CVPR, 2022, pp. 16000-16009.
A. Holzinger, G. Langs, H. Denk, K. Zatloukal, and H. Müller, “Causability and explainability of arti ficial intelligence in medicine,” WIREs Data Mining Knowl. Discov., vol. 9, no. 4, 2019, doi: 10.1002/widm.1312.
K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. CVPR, 2016, pp. 770-778.
M. Tan and Q. Le, “Efficientnetv2: Smaller models and faster training,” in Proc. ICML, 2021, pp. 10096-10106.
R. R. Selvaraju et al., “Grad-cam: Visual explanations from deep networks via gradient-based localiza tion,” in Proc. ICCV, 2017, pp. 618-626.
S. Abnar and W. Zuidema, “Quantifying attention flow in transformers,” in Proc. ACL, 2020, pp. 4190-4197.
D. Hendrycks and T. Dietterich, “Benchmarking neural network robustness to common corruptions and perturbations,” in Proc. ICLR, 2019, doi: 10.48550/arXiv.1903.12261.
