Abstract
Purpose :
To investigate the benefits of spatial filters in the detection of visual field (VF) progression.
Methods :
VF sequences with controlled progression rates were simulated based on the longitudinal VF data of 500 glaucoma patients. One VF per patient was selected as the simulation baseline for stable and progressing sequences (15 fields/sequence, 6 months apart). The number of simulated progressing clusters was determined by the probability distribution of progressing points in the patients’ data. Spatial correlations filters based on anatomical and spatial distances of VF test points were applied to the simulated data (Figure 1). Mean deviations (MD), means of total deviation (TD) clusters, and pointwise TD values were analyzed with linear regression (LR) models to monitor global, regional, and local trends, respectively. VF progression was defined as the detection of one negative LR slope (three for local trends) with probability<T. The values of T were adjusted to achieve 95% specificity when testing in stable fields. The time to detect 80% progressing fields and the detection agreements (Fleiss' kappa) between global, regional, and local trends using filtered and unfiltered data were compared.
Results :
The mean MD progression rates (±SD) of the simulated stable and progressing VF sequences were -0.03 (±0.05) and -0.21 (±0.10) dB/year. The average time (±SD) to detect 80% progression with filtered data were 7.0 (±0.4), 5.0 (±0.3) and 4.8 (±0.2) years for global, regional, and local trends methods, respectively. The time required for regional and local trends methods to detect progression with unfiltered data (5.5±0.3 and 6.2±0.3 years) was significantly longer (p<0.001). With spatial filters, the mean detection agreements (±SD) between the three methods after 2, 3, and 5 years (5, 7, and 11 fields) were 0.47 (±0.03), 0.64 (±0.01) and 0.73 (±0.01), respectively. These were significantly better (p<0.0001) than using unfiltered data, which had lower agreements of 0.31 (±0.03), 0.46 (±0.04), 0.59 (±0.03) after 2, 3, and 5 years.
Conclusions :
Spatial correlation filters based on anatomical and spatial distances of VF test points can improve the time to detect VF progression with regional and local trends methods, but not with the global trend method. The detection agreements between global, regional, and local trends improve when using such filters.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.