Abstract
Purpose: :
To evaluate HRT topographic change analysis (TCA) with false discovery rate control (FDR) and compare it to statistic image mapping of the retina (SIM), and super-pixel progression and event analysis based on re-sampling (SPEAR) that control for type-1 error.
Methods: :
HRT TCA does not control for type-1 error while simultaneously testing each super-pixel in HRT topographies for glaucomatous changes. We use FDR to address this limitation of HRT TCA. FDR controls type-1 error by adjusting the p-value. P-value cutoffs for TCA with FDR were estimated at false positive rate=5% and were applied to each super-pixel to identify topographic locations with significant reduction in retinal height (red-pixels). For each follow-up, ratio of total no. of red-pixels within the optic disk to disk size (red-pixels ratio) was calculated. Progression was defined as the presence of one or more follow-ups with red-pixel ratio > false positive rate. TCA with FDR was compared to HRT TCA, SIM, and SPEAR. For HRT TCA, progression was defined as the presence of a red-pixel cluster ≥ 27 super-pixels within the optic disk in one or more follow-ups*. For SIM and SPEAR, progression was defined as the presence of a significant red-pixel cluster within the optic disk. All eligible eyes from the UCSD Diagnostic Innovations in Glaucoma Study (DIGS) with ≥4 HRT-II exams were used for evaluating the techniques. Of the 245 eyes from 167 participants, 36 eyes progressed by stereophotographs or showed likely progression on SAP Guided Progression Analysis (progressors). All other eyes were considered to be non-progressing (stable patients). Specificities were estimated in 20 DIGS longitudinal normal eyes (normals).
Results: :
Conclusions: :
TCA with FDR provides the highest specificity with a moderate sensitivity and also improves the specificity of HRT TCA. Compared to SIM, SPEAR gives higher specificity and moderate sensitivity. While SIM and SPEAR are computationally intensive, FDR requires only simple calculations to control type-1 error. Low specificity among all the methods in stable patients indicates possible early glaucomatous change detection.
Keywords: imaging/image analysis: non-clinical • computational modeling