PyGlaucoMetrics: An Open-Source Multi-Criteria Glaucoma Defect Evaluation

Abstract

Purpose : The search for the best way to measure glaucomatous vision loss continues. Furthermore, to address the challenge of unreliable ICD codes in clinical center data, we developed and shared an open-source Python repository that standardizes and assesses Visual Field (VF) data. Our repository automatically identifies signs of glaucoma based on predefined criteria.

Methods : Our Python toolbox can categorize glaucomatous VF defects based on the following criteria as outlined in Figure 1a: 1) Hodapp-Anderson-Parrish 2 (HAP2), 2) United Kingdom Glaucoma Treatment Study criteria (UKGTS), and 3) Low-pressure Glaucoma Treatment Study criteria (LoGTS). The analyses were conducted on a publicly available dataset of Humphrey field analyzer (HFA) static perimetry from 7248 unique eyes (3871 patients). Data preprocessing is conducted if necessary to compute mean total deviation (MTD), total deviation probability (TDP), and pattern deviation probability (PDP) using the PyVisualFields library (Figure 1b). Statistical analysis was performed by measuring the variance and reporting the Intraclass Correlation Coefficient (ICC) among three criteria. The toolbox is shared as a GitHub repository (Figure 1) and was added to the PyVisualFields library.

Results : For 28943 VFs, HAP2, UKGTS, and LoGTS identified 17015, 17664, and 17090 VFs as glaucomatous, respectively (Figure 2a). The robustness of these classifications was confirmed by a high ICC of 0.988, indicating strong agreement among the criteria for identifying glaucomatous VFs. Additionally, HAP2 distinct criteria were applied to group VFs into four classes: non-GL, early defect, moderate defect, and severe defect. Multi-class HAP2 identified 41% of total eyes as non-GL while 40%, 6%, and 13% of eyes were classified as early, moderate, and severe defective glaucoma, respectively. Figure 2b showcases the diagnostic outcomes of three employed methods, featuring sensitivity, TD, and PD plots across three distinct scenarios and eyes.

Conclusions : Our open-source Python repository identifies glaucomatous VF defects using widely used criteria, compares glaucoma detection criteria, and demonstrates agreement/disagreement among the methods. It also classifies the stage of VF defects based on HAP2 criteria. Our shared tool provides valuable insights for clinical applications and future research. We will maintain and update this package to ensure its state-of-the-art capabilities.

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