Deep learning based eye disease classification using Optical Coherence Tomography (OCT) images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical Coherence Tomography (OCT) is vital for early ocular disease detection.
• Automated classification of retinal diseases from OCT images is an active research area.
• Deep learning offers potential for improving diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate deep learning models for multi-class classification of retinal diseases using OCT images.
• To compare the performance of different Convolutional Neural Network (CNN) architectures.
• To assess the impact of preprocessing, data augmentation, and class imbalance handling.

Main Methods:

• Four CNNs (VGG16, VGG19, ResNet50, InceptionV3) were evaluated on 7314 OCT images.
• Images were classified into Choroidal Neovascularization (CNV), Diabetic Macular Edema (DME), Drusen, and Normal.
• Models were tested with and without preprocessing, data augmentation, and latent feature space SMOTE for class imbalance.

Main Results:

• VGG-based models, particularly VGG19, showed the most consistent performance.
• VGG19 achieved high internal accuracy (up to 97.54%) with preprocessing and SMOTE.
• External validation revealed reduced generalizability, with VGG19 accuracy at 96.07%.

Conclusions:

• Deep learning models show potential for automated retinal disease classification from OCT images.
• Preprocessing and class imbalance handling are crucial for optimal performance.
• Single-center datasets limit generalizability; multi-center evaluation is needed for clinical deployment.