Ocular Imaging Techniques for Diabetic Retinopathy

Diabetic retinopathy is a microvascular complication of diabetes that can lead to vision loss if undetected or untreated. Modern ocular imaging techniques provide non-invasive and high-resolution visualisation of retinal structures and vasculature, enabling early detection, risk stratification and monitoring of disease progression. Traditional methods include colour fundus photography, which captures two-dimensional images of the retinal surface, and dye-based angiography, where intravenous fluorescein or indocyanine green highlights vascular leakage and perfusion deficits. Optical coherence tomography (OCT) revolutionised retinal assessment by generating cross-sectional images of retinal layers, while OCT angiography (OCTA) further expanded capabilities by mapping blood flow in three dimensions without dye injection. Advances in image processing and artificial intelligence have augmented these modalities with automated lesion detection, quantitative vascular metrics and predictive models for progression risk. Collectively, these technologies support personalised screening intervals, guide therapeutic decisions and serve as end points in clinical trials for new interventions. Their integration into clinical practice holds global significance in reducing the burden of vision-threatening diabetic retinopathy, especially in regions with limited access to specialised ophthalmic care.

Research from Nature Portfolio

Recent studies have leveraged large-scale retinal image datasets to develop generalisable deep learning models for multiple ocular and systemic conditions. A self-supervised foundation model trained on over one million unlabelled fundus images demonstrated strong performance when adapted to detect sight-threatening retinopathy and predict systemic outcomes, reducing the need for extensive expert annotation. Another work has introduced a deep learning system that predicts the time to progression of diabetic retinopathy over a five-year horizon directly from fundus photographs. By integrating multiethnic cohorts and validating in real-world clinical settings, this system achieved high concordance with longitudinal outcomes and offers a pathway to extend personalised screening intervals without compromising early detection.

Ocular Imaging Techniques for Diabetic Retinopathy publication trend

The graph below shows the total number of articles in ocular imaging techniques for diabetic retinopathy across all publications each year (not limited to Nature Index journals).

Technical terms

Fundus photography: Two-dimensional colour imaging of the retina used to document surface lesions and haemorrhages.

Fluorescein angiography: Dye-based method that visualises retinal blood flow and leakage by injecting fluorescein into the bloodstream.

Optical coherence tomography angiography (OCTA): Dye-free imaging technique that generates volumetric maps of retinal and choroidal blood flow by detecting motion contrast.

Self-supervised learning: A machine learning approach where a model is pre-trained on unlabelled data using surrogate tasks to learn general representations.

Foundation model: A large pre-trained neural network that can be fine-tuned for multiple downstream imaging tasks with minimal additional annotation.

References

1. A foundation model for generalizable disease detection from retinal images. Nature (2023).
2. A deep learning system for predicting time to progression of diabetic retinopathy. Nature Medicine (2024).
3. Fundus-DeepNet: Multi-label deep learning classification system for enhanced detection of multiple ocular diseases through data fusion of fundus images. Information Fusion (2024).
4. Optical coherence tomography angiography. Progress in Retinal and Eye Research (2017).
5. Split-spectrum amplitude-decorrelation angiography with optical coherence tomography. Optics Express (2012).