Glaucoma Rose Plot Analysis: Detecting Early Structural Progression Using Angular Histograms

Purpose: To evaluate the novel Rose Plot Analysis (RPA) in the analysis and presentation of glaucoma structural progression data.

Design: Case-control image analysis study using retrospective retinal imaging series.

Subjects: Subjects with open-angle glaucoma with at least 5 registered spectral-domain OCT scans.

Methods: Glaucoma RPA was developed, combining a novel application of angular histograms and dynamic cluster analysis of circumpapillary retinal nerve fiber layer (cRNFL) OCT data. Rose Plot Analysis plots were created for each eye and each visit. Significant clusters of progression were indicated in red. Three masked clinicians categorized all RPA plots (progressing, not progressing), in addition to measuring the significant RPA area. A masked OCT series assessment with linear regression of averaged global and sectoral cRNFL thicknesses was conducted as the clinical imaging standard.

Main outcome measures: Interobserver agreement was compared between RPA and the clinical imaging standard. Discriminative ability was assessed using receiver-operating characteristic curves. The time to detection of progression was compared using a Kaplan-Meier survival analysis, and the agreement of RPA with the clinical imaging standard was calculated.

Results: Seven hundred fourty-three scans from 98 eyes were included. Interobserver agreement was significantly greater when categorizing RPA (κ, 0.86; 95% confidence interval [CI], 0.81-0.91) compared with OCT image series (κ, 0.66; 95% CI, 0.54-0.77). The discriminative power of RPA to differentiate between eyes that were progressing and not progressing (area under the curve [AUC], 0.97; 95% CI, 0.92-1.00) was greater than that of global cRNFL thickness (AUC, 0.71; 95% CI, 0.59-0.82; P < 0.0001) and equivalent to that of sectoral cRNFL regression (AUC, 0.97; 95% CI, 0.92-1.00). A Kaplan-Meier survival analysis showed that progression was detected 8.7 months sooner by RPA than by global cRNFL linear regression (P < 0.0001) in progressing eyes but was not sooner than with sectoral cRNFL (P = 0.06). Rose Plot Analysis showed substantial agreement with the presence of significant thinning on sectoral cRNFL linear regression (κ, 0.715; 95% CI, 0.578-0.853).

Conclusions: Rose Plot Analysis has been shown to provide accurate and intuitive, at-a-glance data analysis and presentation that improve interobserver agreement and may aid early diagnosis of glaucomatous disease progression.