A Hybrid Vgg19–Xgboost Adaptive Transfer Learning Framework for High-Accuracy Multiclass Retinal Disease Classification Using Optical Coherence Tomography
International Journal of Medical Toxicology and Forensic Medicine,
Vol. 16 (2026),
1 January 2026
,
Page 1-9
https://doi.org/10.22037/ijmtfm.v16.53260
Abstract
Background: Early detection of retinal abnormalities is crucial to avoid irreversible visual loss and enable timely clinical interventions using Optical Coherence Tomography (OCT). Multiclass OCT classification remains difficult to automate because of inter-class similarity, limited samples for rare retinal diseases, and significant class imbalance.
Materials and Methods: This paper introduces a new adaptive transfer-learning method, HybVGG-XGB, which combines the fine-tuned VGG19 deep feature extraction with Extreme Gradient Boosting (XGBoost) to classify retinal images into multiclass categories. The proposed framework uses VGG19 to learn discriminative retinal representations and then transfers these features to XGBoost for nonlinear multiclass decision-making, unlike conventional transfer-learning methods that work directly with a fully connected or Softmax classification head.
Results: An experimental subset consisting of 1,618 OCT images from seven categories was used in the experiments: age-related macular degeneration (AMD), diabetic macular edema (DME), epiretinal membrane (ERM), normal (NO), retinal artery occlusion (RAO), retinal vein occlusion (RVO), and vitreomacular interface disease (VID). The proposed HybVGG-XGB framework achieved an accuracy of 88.48%, a macro-F1 score of 0.7890, a sensitivity of 0.7540, a specificity of 0.9782, an ROC-AUC of 0.9903, and a PR-AUC of 0.9342. The findings suggest that task-adaptable convolutional representations, combined with gradient-boosted classification, offer an innovative approach to retinal OCT classification under class imbalance and for computer-aided retinal screening.
Conclusion: The results provide preliminary evidence that combining convolutional feature extraction with gradient-boosted classification is a suitable approach for retinal OCT analysis. However, the degree of minority-class representation and issues of patient-level independence and statistical uncertainty make it difficult to draw conclusions about clinical screening utility.
- Optical Coherence Tomography (OCT), retinal disease classification, hybrid deep learning, transfer learning, VGG-19, XGBoost, gradient boosting, medical image analysis, Computer-Aided Diagnosis (CAD), multi-class classification.
How to Cite
References
[1] Agra CLM, Ventura CV, Da Fonte LD, Oliveira MRV, Leal M, Assunção MELSM, et al. Structural and vascular assessment of the optic nerve head and macula in chronic Chagas disease. Sci Rep. 2025;15(1). [DOI: 10.1038/s41598-025-86582-y]
[2] Bruno P, Quarta A, Calimeri F. Continual learning in medicine: a systematic literature review. Neural Process Lett. 2025;57(1). [DOI: 10.1007/s11063-024-11709-7]
[3] Camacho P, Ribeiro E, Pereira B, Nascimento J, Rosa PC, Henriques J, et al. DNA methyltransferase expression (DNMT1, DNMT3A, and DNMT3B) as a potential biomarker in age-related macular degeneration. J Clin Med. 2025;14(2):559. [DOI: 10.3390/jcm14020559]
[4] Cen L, Ji J, Lin J, Ju S, Lin H, Li T, et al. Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks. Nat Commun. 2021;12(1). [DOI: 10.1038/s41467-021-25138-w]
[5] Pachade S, Porwal P, Kokare M, Deshmukh G, Sahasrabuddhe V, Luo Z, et al. RFMiD: retinal image analysis for multi-disease detection challenge. Med Image Anal. 2025;99:103365. [DOI: 10.1016/j.media.2024.103365]
[6] Domalpally A, Fickweiler W, Levine SR, Goetz KE, VanderBeek BL, Lee A, et al. Data harmonization, standardization, and collaboration for diabetic retinal disease (DRD) research: report from the 2024 Mary Tyler Moore Vision Initiative Workshop on Data. Transl Vis Sci Technol. 2024;13(10):4. [DOI: 10.1167/tvst.13.10.4]
[7] Gharehchopogh FS, Ucan A, Ibrikci T, Arasteh B, Isik G. Slime Mould Algorithm: a comprehensive survey of its variants and applications. Arch Comput Methods Eng. 2023;30(4):2683-723. [DOI: 10.1007/s11831-023-09883-3]
[8] Gholami S, Jannat F, Thompson AC, Ong SSY, Lim JI, Leng T, et al. Distributed training of foundation models for ophthalmic diagnosis. Commun Eng. 2025;4(1). [DOI: 10.1038/s44172-025-00341-5]
[9] Hu K, Zhang Z, Niu X, Zhang Y, Cao C, Xiao F, et al. Retinal vessel segmentation of color fundus images using multiscale convolutional neural network with an improved cross-entropy loss function. Neurocomputing. 2018;309:179-91. [DOI: 10.1016/j.neucom.2018.05.011]
[10] Kaushik K, Bhardwaj A, Cheng X, Dahiya S, Shankar A, Kumar M, et al. Residual network-based deep learning framework for diabetic retinopathy detection. J Database Manag. 2025;36(1):1-21. [DOI: 10.4018/JDM.368006]
[11] He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2016. p. 770-8. [DOI: 10.1109/CVPR.2016.90]
[12] Kermany DS, Goldbaum M, Cai W, Valentim CCS, Liang H, Baxter SL, et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell. 2018;172(5):1122-31. [DOI: 10.1016/j.cell.2018.02.010]
[13] Khan R, Chen C, Zaman A, Wu J, Mai H, Su L, et al. RenalSegNet: automated segmentation of renal tumor, veins, and arteries in contrast-enhanced CT scans. Complex Intell Syst. 2025;11(2). [DOI: 10.1007/s40747-024-01751-2]
[14] Nagel ID, Heinke A, Agnihotri AP, Yassin S, Cheng L, Camp AS, et al. Comparison of a novel ultra-widefield three-color scanning laser ophthalmoscope to other retinal imaging modalities in chorioretinal lesion imaging. Transl Vis Sci Technol. 2025;14(1):11. [DOI: 10.1167/tvst.14.1.11]
[15] Vidivelli S, Padmakumari P, Parthiban C, DharunBalaji A, Manikandan R, Gandomi AH. Optimizing deep learning models for ophthalmological disorder classification. Sci Rep. 2025;15(1). [DOI: 10.1038/s41598-024-75867-3]
[16] El Harti M, Zaamoun S, Jai-Andaloussi S, Ouchetto O. Classification of eye disorders using deep learning and machine learning models. In: International Conference on Advances in Computing Research; 2024. p. 184-194. Cham: Springer Nature Switzerland. [DOI: 10.1007/978-3-031-56950-0_16]
[17] Kulyabin M, Zhdanov A, Nikiforova A, Stepichev A, Kuznetsova A, Borisov V, et al. OCTDL: Optical Coherence Tomography Dataset for Image-Based Deep Learning Methods [dataset]. Mendeley Data. 2024;Version 4. [DOI: 10.17632/sncdhf53xc.4]
- Abstract Viewed: 19 times
- 53260 Downloaded: 15 times