Deep Learning Approach Predicts Longitudinal Retinal Nerve Fiber Layer Thickness Changes.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis and monitoring rely on detecting structural changes like retinal nerve fiber layer (RNFL) thinning.
• Longitudinal data analysis is crucial for understanding disease progression patterns.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for predicting RNFL thickness changes in glaucoma patients.
• To compare the performance of DL models against traditional regression methods for RNFL thickness prediction.

Main Methods:

• Utilized longitudinal optical coherence tomography (OCT) data from 251 glaucoma patients across two studies (DIGS and ADAGES).
• Trained and evaluated four models: linear regression (LR), support vector regression (SVR), gradient boosting regression (GBR), and a custom 1D convolutional neural network (CNN).
• Employed patient-level data splitting for robust evaluation and assessed prediction accuracy using Mean Absolute Error (MAE) and R-squared (R²).

Main Results:

• The gradient boosting regression (GBR) model demonstrated strong performance in predicting pointwise RNFL thickness changes (MAE = 5.2 μm, R² = 0.91).
• The custom 1D CNN achieved superior results in predicting average global and sectoral RNFL thickness changes (MAEs from 2.0-4.2 μm, R² from 0.94-0.98).
• The custom DL models showed consistent performance across diverse demographics and disease severities.

Conclusions:

• Deep learning models, especially the custom 1D CNN, offer a promising approach for accurate and high-resolution prediction of RNFL thickness changes in glaucoma.
• These models have the potential to serve as clinical decision support tools for earlier glaucoma diagnosis and improved disease management.
• The incorporation of longitudinal OCT imaging in DL models enhances their reliability for tracking glaucoma progression over time.