ViFT: Visual field transformer for visual field testing via deep reinforcement learning

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Computer Science

Background:

• Visual field testing (perimetry) is crucial for diagnosing and monitoring visual impairments.
• Prolonged testing times in perimetry can lead to patient fatigue and reduced accuracy.
• There is a need for efficient strategies to shorten visual field testing duration without compromising results.

Purpose of the Study:

• To introduce the visual field transformer (ViFT), an AI-based method for optimizing visual field testing.
• To demonstrate ViFT's ability to control the testing process and learn spatial relationships within the visual field.
• To evaluate ViFT's performance in terms of accuracy and testing time compared to existing methods.

Main Methods:

• Development of a visual field transformer (ViFT) model utilizing deep reinforcement learning.
• Implementation of ViFT to autonomously manage the visual field testing sequence.
• Training ViFT to consider patient perception uncertainty during testing.
• Experimental validation of ViFT on the 24-2 test pattern.

Main Results:

• ViFT achieved comparable or superior accuracy to conventional strategies.
• ViFT reduced visual field testing time by approximately 50%.
• The model learned visual field location relationships without prior information.
• ViFT demonstrated effective control over the testing process, incorporating patient uncertainty.

Conclusions:

• The visual field transformer (ViFT) offers a highly efficient and accurate approach to visual field testing.
• ViFT's deep reinforcement learning framework optimizes testing duration and maintains diagnostic precision.
• This AI-driven method has the potential to improve patient experience and clinical workflow in ophthalmology.