Deep Learning-Based DR Screening from Oct Using Layer-Aware Thickness and Texture Maps

Document Type

Conference Proceeding

Source of Publication

Proceedings International Symposium on Biomedical Imaging

Publication Date

4-8-2026

Abstract

Early detection of retinal disease is vital because structural damage can progress without symptoms and lead to irreversible vision loss. Optical coherence tomography (OCT) is well suited to this need since it resolves the retina into discrete layers with micrometer precision. Focusing on diabetic retinopathy (DR), we propose a pipeline that converts OCT layer segmentation into a pixel-aligned thickness image and fuses it with normalized grayscale intensity and local binary pattern (LBP) microtexture as a three-channel input to a compact transformer optimized end to end. The classifier learns complementary evidence from all three channels and, without pretraining, surpasses competitive convolutional neural network (CNN) and transformer baselines with accuracy 99.12% ± 1.99%, specificity 99.33% ± 2.11%, sensitivity 98.95% ± 2.22%, F1 score 99.20% ± 1.78%, and AUC 99.96% ± 0.11%. Ablations confirm that intensity, texture, and morphology contribute in a balanced manner, leading to the observed gains. These results highlight the potential of multimodal, layer-aware OCT modeling to support earlier and more reliable DR screening in routine workflows.

ISBN

[9798331577636]

ISSN

1945-7928

Publisher

IEEE

Volume

2026-April

Disciplines

Computer Sciences | Medicine and Health Sciences

Keywords

diabetic retinopathy (DR), local binary patterns (LBP), Optical coherence tomography (OCT), retinal layer thickness mapping, transformer-based classification

Scopus ID

105041684710

Indexed in Scopus

yes

Open Access

no

Share

COinS