Bidirectional Neural Network Model for Glaucoma Progression Prediction.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma progression prediction is crucial for timely intervention.
• Existing deep learning models primarily focus on spatial data, neglecting temporal progression.
• Few studies utilize deep learning for predicting glaucoma's time-based visual field changes.

Purpose of the Study:

• To develop and evaluate a bidirectional recurrent deep learning model (Bi-RM) for predicting progressive visual field diagnoses in glaucoma patients.
• To assess the performance of Bi-RM against established methods like linear regression and term memory techniques.

Main Methods:

• A dataset comprising 5413 eyes from 3321 patients was used for training.
• Five consecutive visual field diagnoses served as input to predict the sixth diagnosis.
• The Bi-RM's precision was compared with linear regression (LR) and term memory (TM) algorithms.

Main Results:

• The Bi-RM achieved significantly lower total prediction error than LR and TM.
• Bi-RM exhibited the least prediction error across most test cases in class prediction.
• The model's performance was unaffected by reliability keys or glaucoma severity.

Conclusions:

• The bidirectional recurrent deep learning model (Bi-RM) offers a promising approach for accurate glaucoma progression prediction.
• Bi-RM outperforms traditional methods in forecasting visual field changes, aiding clinical decision-making.
• This deep learning model provides a robust tool for managing glaucoma patients, irrespective of data reliability or disease stage.