[Deep learning to support therapy decisions for intravitreal injections]

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Recent advancements in artificial intelligence (AI) and computer vision have enabled machine learning models to excel in data-rich tasks.
• Deep learning techniques, particularly convolutional neural networks (CNNs), show promise in medical image analysis.

Purpose of the Study:

• To apply a deep convolutional neural network (CNN) for predicting treatment indication for anti-vascular endothelial growth factor (VEGF) medications.
• To evaluate the accuracy of CNN-based prediction using central retinal optical coherence tomography (OCT) scans.

Main Methods:

• A CNN classifier was trained using OCT B-scan images from routine treatments between 2008 and 2016.
• The model was designed to predict treatment indication without human intervention from singular OCT B-scans.

Main Results:

• The CNN achieved over 95% accuracy in predicting the treatment indication for anti-VEGF therapy.
• The model demonstrated successful application of deep learning in analyzing OCT images for clinical decision support.

Conclusions:

• Deep learning, specifically CNNs, can accurately predict anti-VEGF treatment indications from OCT scans.
• While promising, AI-driven classifications require validation by treating physicians due to inherent probabilistic limitations.