Abstract
Purpose :
To evaluate the ellipsoid zone (EZ) on spectral domain optical coherence tomography (SD-OCT) and its association with visual acuity letter score (VALS) in participants with macular edema due to central retinal or hemiretinal vein occlusion in SCORE2.
Methods :
SD-OCT volume scans were evaluated and segmented at baseline, month 01 (M01), month 06 (M06), month 12 (M12), and month 24 (M24). Segmented layer coordinates were used to generate en face thickness maps showing EZ defect. The area of EZ defect within the central subfield (CSF) was measured using automated machine-learning. Expert graders independently reviewed all SD-OCT scans and performed qualitative assessment of EZ status as normal vs abnormal.
Results :
EZ assessment was not possible at baseline due to signal blockage in over 90% of eyes. At M01, 48.7% of eyes had an EZ defect. This proportion decreased to 39.0% and 32.8% at M06 and M12, and then increased to 44.4% at M24. Mean area of EZ defect within the CSF was 0.07 mm2 (SD 0.16) at M01, 0.05 mm2 (SD 0.12) at M06, 0.03 mm2 (SD 0.09) at M12, and 0.06 mm2 (SD 0.16) at M24. Mean VALS was better in eyes without an EZ defect compared to eyes with an EZ defect at all time points: M01 (70.2 vs 60.4, p<0.0001), M06 (74.6 vs 65.6, p=0.0001), M12 (77.0 vs 67.8, p<0.0001), and M24 (73.3 vs 57.2, p<0.0001). Area of EZ defect correlated negatively with VALS at M01 (-0.41, p<0.0001), M06 (-0.40, p<0.0001), M12 (-0.33, p<0.0001), and M24 (-0.51, p<0.0001). An EZ defect at M01 was associated with poorer VALS at M06 (-0.36, p<0.0001), M12 (-0.35, p<0.0001), and M24 (-0.40, p<0.0001). The presence of an EZ defect was associated with the grader’s assessment of abnormal EZ at all time points (p<0.0001).
Conclusions :
In SCORE2, we developed a machine-learning workflow to measure the EZ defect area in the CSF on SD-OCT. Improvement in mean VALS at M01, M06, and M12 correlated with a decrease in the EZ defect area; EZ defect and VALS worsened at M24 after no protocol-defined treatment for a year. At all visits, VALS was better in eyes without an EZ defect compared to those with an EZ defect. In addition, an EZ defect at M01 was associated with poorer VALS at M06, M12, and M24. Machine-learning assessment of the EZ defect was strongly associated with the reading center’s assessment of an abnormal EZ.
This is a 2020 ARVO Annual Meeting abstract.