Macular OCT Classification Using a Multi-Scale Convolutional Neural Network Ensemble.

Area of Science:

• Medical Image Analysis
• Ophthalmology
• Artificial Intelligence

Background:

• Retinal optical coherence tomography (OCT) is crucial for diagnosing eye conditions.
• Computer-aided diagnosis (CAD) systems are essential for early detection and monitoring of retinal pathologies.
• Existing CAD systems require improvement for complex macular diseases.

Purpose of the Study:

• To develop and evaluate a novel CAD system for identifying normal retinas and common macular pathologies (dry age-related macular degeneration, diabetic macular edema) using OCT images.
• To implement a multi-scale convolutional mixture of experts (MCME) ensemble model for enhanced feature learning and classification.
• To assess the system's performance on local and public OCT datasets.

Main Methods:

• A novel multi-scale convolutional mixture of experts (MCME) ensemble model was developed.
• The MCME model utilizes a new cost function for discriminative and fast learning of image features from multiple-scale sub-images.
• Convolutional neural networks were applied within a mixture model incorporating correlated multivariate components for expert interactions.

Main Results:

• The MCME model achieved high classification performance on two distinct macular OCT datasets.
• A precision rate of 98.86% and an area under the receiver operating characteristic curve (AUC) of 0.9985 were obtained on average with four scale-dependent experts.
• The model demonstrated effective identification of normal retinas and the targeted macular pathologies.

Conclusions:

• The proposed MCME-based CAD system shows significant promise for accurate and efficient diagnosis of retinal pathologies from OCT images.
• This approach can assist ophthalmologists in the early detection and management of ocular diseases.
• The MCME model offers a robust and data-driven solution for medical image analysis in ophthalmology.