# A Structure–Function Dataset With Known Rates of Glaucomatous Progression

## TL;DR

This paper introduces an open dataset of glaucoma progression with known rates, combining structural and functional eye measurements for benchmarking diagnostic methods.

## Contribution

The paper presents a synthetic, open-access dataset with real-world progression rates for glaucoma, enabling benchmarking of diagnostic methods.

## Key findings

- The dataset includes 162 eyes with longitudinal SAP and RNFL data over 5 years.
- Progression rates in the dataset align with clinical data when assessed using PoPLR.
- The synthetic data preserves structure-function relationships from real clinical data.

## Abstract

This article describes the Lions Eye Institute Structure–Function Dataset (LEI-SFD), an open, synthetic dataset of both structural and functional measurements from eyes with glaucoma that are progressing at known rates.

The LEI-SFD contains static automated perimetry (SAP) and retinal nerve fiber layer (RNFL) thickness data (circumpapillary RNFL [cpRNFL]) that is progressing at known rates for 10 visits over 5 years. Measurements at the 10th visit and progression rates are taken from curated clinical data collected at Lions Eye Institute, Perth, Australia. Measurement noise is added based on existing literature.

The dataset contains 162 eyes with a mean (SD) baseline mean deviation of −2.6 (4.0) dB (minimum, −17.6; maximum, 1.6) and a mean (SD) cpRNFL thickness of 77.7 (13.5) microns (minimum, 43; maximum, 109). The average number of progressing SAP locations per eye is 20.8, with a mean pointwise rate of −0.6 dB/y. Using Permutation of Pointwise Linear Regression (PoPLR) to assess progression on the resulting datasets gives similar classification results to those published on clinical data.

This open dataset contains longitudinal, linked structural and functional data with known progression rates. By using visit data and progression rates from real eyes to seed its synthetic generation, relationships between structure and function in current clinical data should be preserved, but with ground-truth progression rates being known.

This open dataset will allow the assessment of the performance of methods for determining glaucomatous progression that use both structure and function on a common benchmark.