Automated Diabetic Retinopathy Lesion Classification using ONH-Based Multi-Scale Feature Extraction and SVM

This work set out to automate diabetic retinopathy lesion classification while addressing a specific, recurring weakness in existing systems: the misclassification of the Optical Nerve Head as an exudate. By excluding the ONH region before feature extraction, and combining this with multi-scale Gabor and Local Binary Pattern features, the proposed framework gave its SVM classifiers meaningfully better information to work with, which translated into a marked accuracy improvement over the existing method, from 93% to 100% on the evaluated test set, alongside complete sensitivity and specificity. The multi-label classification stage further allows the system to report every lesion type present in an image rather than a single generic abnormality flag, which matters for genuine clinical usefulness. Testing on both public datasets and real-time clinical images adds confidence that the approach generalizes beyond curated benchmark data. Future work could extend this framework with deep learning-based feature extraction, severity grading across DR stages, larger multi-source clinical datasets, and eventual deployment as a clinical screening tool integrated with electronic health records.

[1] Abd El-Fattah EE, Saber S, Youssef ME, Eissa H, El-Ahwany E, Amin NA, et al. AKT-AMPKα-mTOR-dependent HIF-1α activation is a new therapeutic target for cancer treatment: A novel approach to repositioning the antidiabetic drug sitagliptin for the management of hepatocellular carcinoma. Front Pharmacol. 2022;12:720173. https://doi.org/10.3389/fphar.2021.720173. [2]  Abdelhady R, Mohammed OA, Doghish AS, Hamad RS, Abdel-Reheim MA, Alamri MMS, et al. Linagliptin, a DPP-4 inhibitor, activates AMPK/FOXO3a and suppresses NF-κB to mitigate the debilitating effects of diethylnitrosamine exposure in rat liver: Novel mechanistic insights. FASEB J. 2024;38:e23480. https://doi.org/10.1096/fj.202302461RR. [3]   Abdelhamid AM, Saber S, Youssef ME, Gaafar AGA, Eissa H, Abd-Eldayem MA, et al. Empagliflozin adjunct with metformin for the inhibition of hepatocellular carcinoma progression: Emerging approach for a new application. Biomed Pharmacother. 2022;145:112455. https://doi.org/10.1016/j.biopha.2021.112455. [4]  Adak S, Magdalene D, Deshmukh S, Das D, Jaganathan BG. A review on mesenchymal stem cells for treatment of retinal diseases. Stem Cell Rev Rep. 2021;17:1154–1173. https://doi.org/10.1007/s12015-020-10090-x. [5]   Akhter MH, Ahmad I, Alshahrani MY, Al-Harbi AI, Khalilullah H, Afzal O, et al. Drug delivery challenges and current progress in nanocarrier-based ocular therapeutic systems. Gels. 2022;8:82. https://doi.org/10.3390/gels8020082. [6]  Ambekar A, Sahoo J, Singh K. Breaking barriers in ocular drug delivery for uveitis: Advanced drug delivery systems, challenges and future prospects. Exp Eye Res. 2025;257:110456. https://doi.org/10.1016/j.exer.2025.110456. [7]   Antonetti DA, Silva PS, Stitt AW. Current understanding of the molecular and cellular pathology of diabetic retinopathy. Nat Rev Endocrinol. 2021;17:195–206. https://doi.org/10.1038/s41574-024-01053-0 [8]  Appell MB, Pejavar J, Pasupathy A, Rompicharla SVK, Abbasi S, Malmberg K, et al. Next-generation therapeutics for retinal neurodegenerative diseases. J Control Release. 2024;367:708–736. https://doi.org/10.1016/j.jconrel.2024.01.063. [9]   Asadi S, Rahimi Z, Kohsari M, Babajani F, Amiri M, Jalilian N, et al. Assessing the possible association between MTHFR (rs1801133) and GPx-1 (rs1050450) polymorphisms with the risk of type 2 diabetes, diabetic neuropathy, and diabetic retinopathy. Mol Biol Rep. 2024;51:583. https://doi.org/10.1007/s11033-024-09519-0. [10]   Berrocal MH, Acaba-Berrocal L. Early pars plana vitrectomy for proliferative diabetic retinopathy: Update and review of current literature. Curr Opin Ophthalmol. 2021;32:203–208. https://doi.org/10.1097/ICU.0000000000000760.