A fully accurate web-based application for automatic classification of glaucomatous visual field defects using Hodapp-Parrish-Anderson criteria.

Abstract

Purpose : To evaluate a novel web-based application for grading 24-2 Humphrey visual field test results in accordance with Hodapp-Parrish-Anderson (HPA) criteria.

Methods : The Glaucoma Field Defect Classifier (GDFC) was designed and deployed online for researchers to utilise (https://gfdc.app). 168 consecutively recorded visual fields of 89 glaucoma patients attending clinic at a tertiary referral centre were studied. Each visual field was evaluated by two independent researchers and disagreement resolved by a third researcher acting as arbiter; all using printed HPA criteria to make their appraisals. The same fields were inputted into GFDC and outcomes were compared against gold-standard human assessment to gauge accuracy.

Results : For every single perimetry result, GDFC produced output which matched human grading based on HPA criteria. The accuracy of GFDC was therefore 100% for each level of visual field grading (mild, moderate, severe). Sensitivity and specificity were also 100%. Interpretability analysis confirmed the web-application correctly identified test loci on perimetry result charts and classified defects based on result legends.

Conclusions : GDFC, a web-application, exhibits equivalent performance to human graders classifying glaucomatous visual field defects based on HPA criteria. GFDC can facilitate visual field assessment at scale with potential to augment clinical practice and research—where application of explicit perimetry criteria is often precluded by time constraints. The code used to build GFDC is freely available online for others to adapt to support use with alternative visual field result formats.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.