Bidirectional gated recurrent unit network model can generate future visual field with variable number of input...

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Informatics

Background:

• Visual field tests are crucial for diagnosing and monitoring glaucoma.
• Predicting future visual field test results can improve patient management and treatment strategies.
• Deep learning models offer potential for analyzing complex visual field data.

Purpose of the Study:

• To predict future visual field tests using a bidirectional gated recurrent unit (Bi-GRU) model.
• To evaluate the Bi-GRU model's performance based on the number of input visual field tests and prediction time interval.
• To assess the model's accuracy across different glaucoma severities.

Main Methods:

• Utilized a dataset of 185,858 visual field tests from 23,517 eyes for training and 9,459 tests from 1,053 eyes for testing.
• Developed a Bi-GRU architecture capable of processing 3 to 80 past visual field tests.
• Predicted key metrics: Mean Deviation (MD), Pattern Standard Deviation (PSD), Visual Field Index (VFI), and Total Deviation Value (TDV).

Main Results:

• Prediction errors for MD, PSD, VFI, and TDV were within acceptable ranges (e.g., MD: 1.20-1.68 dB, VFI: 3.64-4.51%).
• Prediction error increased with longer prediction time intervals, though not significantly.
• Prediction errors significantly increased with worsening glaucoma severity.

Conclusions:

• The Bi-GRU model can reliably predict future visual field tests using as few as three previous tests.
• This AI-driven approach shows promise for clinical application in glaucoma management.
• The model's accuracy is influenced by disease severity and prediction horizon.