PubMed ID: 42417595
Author(s): Schulz M, Macha N, Zhang S, Pak JW, Storm S, Lopez DA, Blodi BA, Domalpally A. Comparison of Measurement Techniques for Photoreceptor Loss in Geographic Atrophy. Transl Vis Sci Technol. 2026 Jul 1;15(7):8. doi: 10.1167/tvst.15.7.8. PMID 42417595
Journal: Translational Vision Science & Technology, Volume 15, Issue 7, Jul 2026
PURPOSE Ellipsoid zone (EZ) loss is an emerging end point in geographic atrophy (GA) clinical trials. Standardized ground truth is important to train artificial intelligence models in this biomarker. This study compares segmentation and edge detection methods for quantifying EZ and retinal pigment epithelium (RPE) loss on optical coherence tomography (OCT).
METHODS OCT images from 50 eyes with GA were analyzed using segmentation (OCT Explorer) and edge detection (3D Slicer). With segmentation, the EZ and RPE borders were traced to generate thickness maps. For edge detection, graders marked regions of complete EZ or RPE loss on each B-scan. En face maps were generated from both methods and compared. Longitudinal progression was evaluated in 15 eyes with 1 year of follow-up.
RESULTS The mean EZ loss area was 11.09 ± 5.40 mm2 by segmentation and 9.98 ± 5.01 mm2 by edge detection (P < 0.001). The mean RPE loss area was 7.66 ± 4.40 mm2 and 7.48 ± 4.23 mm2, respectively (P = 0.53) The EZ/RPE ratio was larger with segmentation (1.59 vs. 1.42; P < 0.001). Longitudinally, the mean EZ loss change was 1.71 mm²/year vs. 1.45 mm²/year (P = 0.43), and the RPE loss change was 1.74 vs. 1.60 mm²/year (P = 0.23).
CONCLUSIONS Segmentation and edge detection showed high agreement for EZ and RPE loss, but systematic differences were observed in EZ measurement. Segmentation yielded larger EZ areas by including attenuation over drusen, while edge detection provided more conservative boundaries restricted to GA only.
TRANSLATIONAL RELEVANCE Measurement method influences ellipsoid zone quantification, with implications for trial design and artificial intelligence ground truth development.