Balwinder Singh, Dr. Navneet Kaur, Dr. Sikander Singh Cheema

DOI: https://doi.org/10.5281/zenodo.21982052

Retinal diseases, such as Diabetic Macular Edema (DME), Choroidal Neovascularization (CNV), and drusen, are the leading causes of irreversible vision loss worldwide. Early and accurate diagnosis is critical for effective treatment and preservation of vision. Optical Coherence Tomography (OCT) has emerged as a standard, noninvasive imaging modality that provides high-resolution, cross-sectional images of the retina, en-abling detailed morphological assessment. However, the manual interpretation of a large volume of OCT scans is time-consuming, subjective, and requires significant expertise, leading to poten-tial diagnostic delays and inter-observer variability. This study presents an automated diagnostic system using deep learning to classify retinal OCT images. We propose a Convolutional Neural Network (CNN)-based model designed to accurately and efficiently detect the presence of common retinal pathologies. The model was trained and validated on a publicly available dataset of retinal OCT images categorized into four classes: Normal, CNV, DME, and Drusen. This automated system has the potential to serve as a powerful decision support tool, streamline the diag-nostic workflow, facilitate large-scale screening programs, and ultimately improve patient outcomes through timely intervention.

Retinal Disease Classification, Optical Coher-ence Tomography (OCT), Deep Learning, Convolutional Neural Networks (CNN), Diabetic Macular Edema (DME), Choroidal Neovascularization (CNV), Drusen, Automated Diagnosis

This study presented the implementation and evaluation of a MobileNetV3-based model for multi-class classification. The experimental results demonstrated that the model achieved very high accuracy and comparatively low loss during training and validation phases, indicating effective learning of feature representations from the dataset. However, the performance on the test data was lower, with an overall accuracy of 73% and consistently low precision, recall, and F1-scores across all classes. This substantial performance gap highlights a some sort of overfitting. The findings suggest that, although MobileNetV3 is a powerful and efficient architecture, its performance is highly dependent on data quality, distribution, and training strategy. The observed overfitting indicates that the model tends to memorize training data rather than learning robust, generaliz-able patterns. Therefore, further improvements are necessary to enhance real-world applicability. Future work will focus on addressing these limitations through better data preprocessing, balanced dataset construction, application of regularization techniques, and improved training strategies. By incorporating these enhancements, the model’s generalization performance can be significantly improved, making it more reliable for practical deployment.

[1]M. E. Sahin, “A deep learning-based technique for the diagnosis of retinal diseases using OCT images,” Turkish Journal of Science and Technology, 2022. doi: 10.55525/tjst.1128395.
[2]M. Talebzadeh, A. Sodagartojgi, Z. Moslemi, S. Sedighi, B. Kazemi, and F. Akbari, “Deep learning-based retinal abnormality detection from OCT images with limited data,” World Journal of Advanced Research and Reviews, vol. 21, no. 3, 2024. doi: 10.30574/wjarr.2024.21.3.0716.
[3]M. Pekala, N. Joshi, D. E. Freund, N. M. Bressler, D. Cabrera DeBuc, and P. Burlina, “Deep learning based retinal OCT segmentation,” arXiv preprint arXiv:1801.09749, 2018.
[4]J. Kim and L. Tran, “Retinal disease classification from OCT images using deep learning,” in IEEE CIBCB, 2021.
[5]M. Miranda and F. J. Romero, “Antioxidants and retinal diseases,” Antioxidants, vol. 8, no. 12, p. 604, 2019.
[6]M. Berrimi and A. Moussaoui, “Deep learning for identifying and classifying retinal diseases,” in Proc. ICCIS, IEEE, 2020, pp. 1–6.
[7]A. P. Sunija, S. Kar, S. Gayathri, V. P. Gopi, and P. Palanisamy, “OCTNet: A lightweight CNN for retinal disease classification from OCT images,” Computer Methods and Programs in Biomedicine, vol. 200, p. 105877, 2021.
[8]M. Elkholy and M. A. Marzouk, “Deep learning-based classi-fication of eye diseases using convolutional neural network for OCT images,” Frontiers in Computer Science, vol. 5, 2023. doi: 10.3389/fcomp.2023.1252295.
[9]M. Wang, “Research on retinal OCT image classification based on deep learning,” Journal of Computing and Electronic Information Manage-ment, 2023. doi: 10.54097/wnr87znj.
[10]M. Lin, G. Bao, X. Sang, and Y. Wu, “Recent advanced deep learning architectures for retinal fluid segmentation on optical coherence tomog-raphy images,” Sensors, vol. 22, no. 8, 2022. doi: 10.3390/s22083055.
[11]M. Rahimzadeh and M. R. Mohammadi, “ROCT-Net: A new ensemble deep convolutional model for detecting retinal diseases from OCT images,” arXiv:2203.01883, 2022.
[12]S. Sotoudeh-Paima et al., “Multi-scale convolutional neural net-work for automated AMD classification using retinal OCT images,” arXiv:2110.03002, 2021.
[13] Y. Guo et al., “Automated segmentation of retinal fluid volumes from OCT using deep learning,” arXiv:2006.02569, 2020.
[14]M. Noor et al., “RetinaVision: XAI-driven deep learning framework for retinal disease classification,” arXiv:2602.19324, 2026.
[15]K. He et al., “Deep residual learning for image recognition,” in Proc. CVPR, 2016. (Used widely as backbone in OCT classification models)


[16]M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convo-lutional neural networks,” in Proc. ICML, 2019.
[17]A. Howard et al., “Searching for MobileNetV3,” in Proc. ICCV, 2019.
[18]J. De Fauw et al., “Clinically applicable deep learning for diagnosis and referral in retinal disease,” Nature Medicine, vol. 24, no. 9, 2018.
[19]D. S. Kermany et al., “Identifying medical diagnoses and treatable diseases by image-based deep learning,” Cell, vol. 172, no. 5, 2018.
[20]H. Fu et al., “Deep learning-based automated detection of retinal diseases from OCT images,” IEEE Transactions on Medical Imaging, 2019.
[21]Z. Li et al., “Deep learning for detecting retinal diseases from OCT images,” IEEE Access, 2020.
A. Rasti et al., “Deep learning-based classification of retinal OCT images: A survey,” Neural Computing and Applications, 2022.

Balwinder Singh, Dr. Navneet Kaur, Dr. Sikander Singh Cheema (2026). Lightweight MobileNet-Based Retinal Disease Classification Using OCT Images. International Journal of Computer Science Engineering Techniques, 10(4), 57–62. ISSN: 2455-135X. DOI: https://doi.org/10.5281/zenodo.21982052