An Enhanced Multi-View Deep-Broad Learning Framework for Reliable Retinal Disease Classification

Abstract

Optical coherence tomography (OCT) provides cross-sectional views of retinal structures and is widely used to assess conditions that affect the macula. Automated analysis may help clinicians review large volumes of images, but a model’s performance can be affected by image quality, differences between datasets, class imbalance, and uncertainty in its predictions. This paper proposes a reliability-oriented framework for classifying OCT images into four categories: choroidal neovascularization (CNV), diabetic macular edema (DME), drusen, and normal retina. The proposed pipeline combines basic image-quality screening, three complementary image representations, feature extraction, attention-guided fusion, and a broad learning classification stage. A confidence-based referral step is also included so that unsuitable or uncertain cases can be flagged for expert review rather than automatically assigned a definitive label. Recent research from 2024 to 2026 is reviewed to place the proposal in context. In particular, a 2024 study has already investigated multi-view deep-broad learning for retinal disease assessment, while later OCT studies have examined robustness, hybrid CNN–Transformer architectures, lesion-aware fusion, and calibration. Therefore, the potential contribution of this work is framed around evaluating reliability and referral behaviour within a clearly controlled OCT classification experiment, not around claiming that multi-view deep-broad learning is a new idea. No experimental results are claimed in this manuscript; implementation and validation are required before the framework’s effectiveness can be established.