Multi-scale convolutional neural network for automated AMD classification using retinal OCT images.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Disease Diagnosis
• Ophthalmology and Vision Science

Background:

• Age-related macular degeneration (AMD) is a leading cause of vision loss in older adults.
• Increasing use of optical coherence tomography (OCT) and an aging population strain healthcare resources.
• Deep learning offers potential for automated AMD diagnosis from OCT scans.

Purpose of the Study:

• To propose a multi-scale convolutional neural network (CNN) for automated AMD diagnosis.
• To effectively capture and fuse features from varying scales within OCT images.
• To improve diagnostic performance by addressing inter-scale variations of AMD pathologies.

Main Methods:

• Developed a multi-scale CNN utilizing a feature pyramid network (FPN) structure.
• Trained and evaluated the model on two large datasets: NEH (12,649 images) and UCSD (108,312 images).
• The model diagnoses normal eyes, dry AMD (drusen), and wet AMD (choroidal neovascularization).

Main Results:

• The proposed multi-scale CNN outperformed existing OCT classification frameworks.
• Feature fusion improved performance by 0.4%–3.3% across tested backbone models.
• Gradual learning enhanced accuracy from 87.2% to 93.4%, with heatmap visualization confirming pathology detection at different sizes.

Conclusions:

• The multi-scale CNN demonstrates superior performance for AMD diagnosis using OCT images.
• The architecture effectively detects pathologies of varying sizes, validated by heatmap analysis.
• This AI tool shows promise as a screening aid in healthcare settings for ophthalmologists.