Deep-Learning-Based Analysis of Disease-Specific Structural Biomarkers on Retinal Sensitivity in Neovascular Age-Related Macular Degeneration.

PURPOSE: To evaluate the impact of disease-specific biomarkers on pointwise sensitivity (PWS) in patients with active neovascular age-related macular degeneration (nAMD) using two microperimetry (MP) instruments.

METHODS

Participants in this prospective cross-sectional study underwent imaging with a SPECTRALIS HRA+OCT and MP with the photopic MP-3 (NIDEK) and mesopic MAIA (CenterVue). Pigment epithelium detachment (PED), intraretinal and subretinal fluid (IRF, SRF), ellipsoid zone loss (EZL), and subretinal hyperreflective material (SHRM) were quantified using deep learning (DL). DL enabled co-registration between the MP and optical coherence tomography (OCT). Univariate and multivariable mixed-effects models using variable selection including interaction terms were used to assess the effects of the biomarkers on PWS.

RESULTS

Twenty eyes of 20 subjects (mean age, 76.0 years) were included. Sensitivity in the MAIA was lower (-2.87 dB; 95% confidence interval [CI], -3.15 to -2.59) than in the MP-3. Significant interactions between EZL and SRF, EZL and IRF, and between EZL and eccentricity were observed. EZL significantly reduced PWS (-1.30 dB; 95% CI, -2.08 to -0.51) at 0° eccentricity with SRF of 0.15 nL and absent IRF. IRF presence further decreased PWS by -2.57 dB (95% CI, -3.66 to -1.47). SRF negatively impacted PWS (-2.08 dB/nL; 95% CI, -2.53 to -1.64), but showed no association when EZL was present (+0.05 dB/nL; 95% CI, -0.81 to 0.92). SHRM presence and PED volume reduced PWS by -5.13 dB (95% CI, -5.83 to -4.43) and -0.71 dB/nL (95% CI, -0.92 to -0.50), respectively. Eccentricity only impaired function in the presence of EZL (-0.49 dB/°; 95% CI, -0.61 to -0.37).

CONCLUSIONS

DL-quantified biomarkers mostly had a negative impact on PWS in nAMD, particularly IRF and SHRM presence. The above-mentioned biomarkers should be considered when managing patients with nAMD and using MP in trials and clinical settings.