Deep learning models to predict primary open-angle glaucoma.

Area of Science:

• Ophthalmology
• Computer Science
• Medical Data Analysis

Background:

• Glaucoma is a leading cause of irreversible blindness globally.
• Visual field (VF) testing is crucial for monitoring glaucoma progression.
• Existing prediction methods often use single time points and binary classification, limiting accuracy.

Purpose of the Study:

• To develop and evaluate deep learning models for predicting the time-to-glaucoma conversion.
• To leverage longitudinal visual field data, incorporating temporal and spatial information.
• To address limitations of binary classification in time-to-event prediction for glaucoma.

Main Methods:

• Implementation of several deep learning approaches, including Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks.
• Utilizing longitudinal visual field data from the Ocular Hypertension Treatment Study (OHTS) dataset.
• Developing models that naturally handle temporal dependencies and spatial patterns in VF data.

Main Results:

• The proposed CNN-LSTM model demonstrated superior performance compared to other examined models.
• The models effectively incorporated temporal and spatial information from longitudinal VF data.
• Accurate prediction of time-to-glaucoma conversion was achieved.

Conclusions:

• Deep learning models, particularly CNN-LSTM, are effective for predicting glaucoma conversion using longitudinal visual field data.
• Incorporating temporal and spatial dynamics improves the accuracy of glaucoma progression prediction.
• This approach offers a more robust method for identifying individuals at risk of glaucoma.