AI-Based Response Classification After Anti-VEGF Loading in Neovascular Age-Related Macular Degeneration

Area of Science:

• Ophthalmology, Artificial Intelligence in Medicine, Medical Imaging Analysis

Background:

• Wet age-related macular degeneration (AMD) involves macular neovascularization (MNV), treated with anti-VEGF injections.
• Current treatment assessment relies on optical coherence tomography (OCT) fluid presence and visual acuity, which can be discrepant.
• Objective assessment of treatment response is crucial for managing wet AMD.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) classification model for assessing anti-VEGF treatment response in wet AMD patients.
• To objectively predict both anatomic disease activity and visual prognosis from OCT images at 3 months post-treatment.
• To overcome limitations of subjective assessments in clinical practice.

Main Methods:

• A retrospective study of 120 patients (144 eyes) treated with intravitreal bevacizumab.
• Development of a Siamese neural network (ResNet-18-based) model trained on pre-treatment and 3-month post-treatment OCT image pairs.
• Patients classified into three groups: active disease, good response, and limited response based on fluid presence and visual acuity changes.

Main Results:

• The AI model achieved 95.4% accuracy in classifying treatment response.
• High performance metrics: macro precision (0.948), macro recall (0.949), and macro F1 score (0.948).
• Explainability methods (LayerCAM, SHAP) confirmed the model's focus on pathology-related regions in OCT images.

Conclusions:

• The AI model effectively classifies post-loading treatment response in wet AMD patients.
• The model integrates anatomic disease status and visual prognosis prediction from OCT scans.
• This AI tool offers an objective method for evaluating treatment efficacy in wet AMD.