Deep Learning Model for Automated Classification of Macular Neovascularization Subtypes in AMD

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Neovascular age-related macular degeneration (AMD) involves macular neovascularization (MNV).
• Accurate classification of MNV subtypes (type 1, 2, 3) is crucial for effective treatment.
• Current classification relies on expert interpretation of structural optical coherence tomography (OCT) images.

Purpose of the Study:

• To develop a deep learning algorithm for classifying MNV subtypes using structural OCT images.
• To evaluate the performance of the deep learning model in treatment-naïve neovascular AMD patients.
• To assess the impact of image data homogenization on classification accuracy.

Main Methods:

• A retrospective cohort of 193 eyes with treatment-naïve neovascular AMD was analyzed.
• Convolutional neural network (CNN)-based deep learning models were trained using cross-validation.
• Structural OCT images were preprocessed, including data homogenization, for classification.

Main Results:

• Homogenized OCT data significantly improved classification performance across all models.
• High sensitivity and specificity were achieved for all MNV subtypes (e.g., Type 1: 96.7% sensitivity, 84.9% specificity).
• Area under the ROC curve (AUC) values ranged from 0.91 to 0.97, indicating strong diagnostic capability.

Conclusions:

• The deep learning model accurately classifies MNV subtypes on structural OCT.
• Data homogenization enhances the model's diagnostic performance.
• This AI tool can assist clinicians in diagnosing MNV subtypes, potentially improving patient outcomes in neovascular AMD.