Structure-Function Correlation of Deep-Learning Quantified Ellipsoid Zone and Retinal Pigment Epithelium Loss and Microperimetry in Geographic Atrophy.

PURPOSE

The purpose of this study was to define structure-function correlation of geographic atrophy (GA) on optical coherence tomography (OCT) and functional testing on microperimetry (MP) based on deep-learning (DL)-quantified spectral-domain OCT (SD-OCT) biomarkers.

METHODS

Patients with GA were prospectively examined by SD-OCT (Spectralis, 97 B-scans) and two microperimetry devices (MP3 and MAIA) in two combined test runs each. DL-algorithms measured the ellipsoid-zone thickness (EZT), ellipsoid-zone loss (EZL), hyper-reflective-foci (HRF) volume, drusen-volume (DV), and retinal-pigment-epithelium loss (RPEL) area. Pointwise co-registration was established between all stimuli and the location on OCT. A multivariable mixed-effect model with variable selection was used to identify pointwise retinal sensitivity (PWS) changes for each biomarker, accounting for age and eccentricity.

RESULTS

Three thousand six hundred stimuli points were collected and correlated with 1940 OCT B-scans in 20 eyes of 20 patients. PWS was significantly lower in stimuli with EZL without RPEL (-2.81, 95% confidence interval [CI] = -3.72 to -1.91 decibel [dB], 0 degrees, P < 0.0001) and in areas with both EZL and RPEL (-10.03, 95% CI = -10.96 to -9.11 dB, 0 degrees, P < 0.0001) compared to areas without any atrophy. Increased EZT had a significant positive effect on PWS (0.34, 95% CI = 0.32 to 0.36 dB/µm, P < 0.0001). Structure-function correlations were consistent throughout all levels of eccentricity with P < 0.001. Drusen and HRF volume, but not age, were associated with reduced PWS.

CONCLUSIONS

Functional impairment by MP was associated with defined morphological changes as quantified by DL on OCT. PR degeneration seen as EZL alone impairs the function on MP examinations. The combination of DL-based SD-OCT biomarker assessment and MP appear suited for evaluation of retinal function beyond visual acuity for disease monitoring.