Automatic Detection and Classification of Diabetic Retinopathy from Optical Coherence Tomography Angiography Images using Deep Learning-A Review

Abstract

Diabetic retinopathy (DR), a microvascular
complication of diabetes, has become a major global health
problem, affecting vision and potentially leading to blindness if
left untreated. Optical Coherence Tomography Angiography
(OCTA) has become a transformative imaging technique for the
detection and analysis of the choriocapillaris and retinal
microvasculature, enabling the identification of preclinical
microvascular abnormalities that precede visible DR symptoms.
This review examines the role of machine learning (ML) and
deep learning (DL) learning methods in OCTA-based DR
classification. We summarize recent advances in convolutional
neural networks (CNNs) for automated feature extraction and
accurate diagnosis, as well as the various OCTA datasets used in
these studies. The advantages of OCTA imaging over fundus
photography, particularly for early-stage DR detection, are
highlighted. Furthermore, we propose a novel DL-based system
for DR classification that compares its performance with
traditional ML methods based on manual feature extraction.
Challenges related to clinical delivery, such as data variability,
model interpretability, and integration into clinical workflows,
are also discussed. Finally, we highlight future research
directions to address these challenges and improve the adoption
of Deep Learning models for OCTA-based DR diagnosis