Deep Learning Estimation of 24-2 Visual Field Map From Optic Nerve Head Optical Coherence Tomography Angiography

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Visual field (VF) testing is crucial for diagnosing and monitoring glaucoma and other optic nerve head (ONH) conditions.
• Optical coherence tomography angiography (OCTA) provides detailed microvascular information of the ONH.
• Current VF testing can be time-consuming and may not capture subtle changes effectively.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for estimating 24-2 visual field (VF) maps from OCTA optic nerve head (ONH) en face images.
• To compare the performance of DL models against traditional linear regression (LR) methods.

Main Methods:

• Trained DL models on 3148 VF OCTA image pairs from 994 participants, utilizing radial peripapillary capillary (RPC), superficial, and choroidal ONH vascular density (VD) layers.
• Estimated 24-2 mean deviation (MD), pattern standard deviation (PSD), total deviation (TD), and pattern deviation (PD) values.
• Assessed model accuracy using mean absolute error (MAE) and Pearson correlation coefficient (R).

Main Results:

• DL models significantly outperformed LR models in estimating VF values across all tested ONH layers (P <0.001).
• Using RPC, DL achieved an R of 0.79 and MAE of 1.77 dB for MD estimation.
• DL models using combined ONH layers showed slight improvements for MD and TD estimations compared to individual layers.

Conclusions:

• Deep learning models applied to OCTA images demonstrate high accuracy in estimating 24-2 visual field maps.
• Leveraging ONH microvascular information from OCTA via DL holds potential for reducing the frequency of VF testing.
• This AI-driven approach may offer a more efficient method for monitoring optic nerve health.