# Visual Field Estimation in X-Linked Retinitis Pigmentosa Associated with Retinitis Pigmentosa GTPase Regulator (RPGR) from Image Analysis Using Artificial Intelligence

## TL;DR

This study uses artificial intelligence to estimate visual field data in patients with a specific form of retinitis pigmentosa using OCT scans.

## Contribution

The novel use of AI to predict visual field parameters from OCT scans in RPGR-associated retinitis pigmentosa patients is introduced.

## Key findings

- Ellipsoid zone area (EZA) showed the strongest associations with visual field metrics like mean sensitivity.
- AI segmentation of OCT scans enabled efficient and statistically significant structure-function predictions.
- The method proved effective across a wide range of disease severity and age.

## Abstract

To develop an efficient approach to estimating visual field (VF) in patients with X-linked retinitis pigmentosa (RP) based on macular OCT scans.

Retrospective analysis of patients who were enrolled in a natural history study at Moorfields Eye Hospital (London, United Kingdom).

Male patients with genetically confirmed retinitis pigmentosa GTPase regulator (RPGR)-associated RP.

Visual field raw data were exported and analyzed including Visual Field Modeling and Analysis software. Retinal imaging consisted of OCT macular scans. Paired imaging and VF data acquired within a 1-month range were jointly analyzed. Artificial intelligence (AI) was used to automatically segment and quantify macular ellipsoid zone width (EZW), and ellipsoid zone area (EZA).

Functional parameters from static VF testing such as mean sensitivity (MS) and Hill of Vision analysis that included total volume (VTOT), volume of central 20° (V20), and volume of central 30° (V30) were predicted from EZW and EZA.

Patient age ranged from 5 to 55 years old at baseline. A total of 332 OCT-VF pairs were analyzed. Ellipsoid zone area had the highest conditional R2 (R2c) and most significant associations with MS and V20. There were significant associations between MS and EZW (P = 0.00176), and MS with EZA (P = 0.0009).

This study showed that AI enables efficient acquiring of large amounts of structural OCT parameters, facilitating research and structure-function predictions. The cohort included patients with a wide range of disease severity and statistical significance was achieved with parameters representing a wide range of VF, proving that this method can be applied for patients with milder disease.

Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.