Evaluation of segmentation procedures using spectral domain optical coherence tomography in exudative age-related...

The purpose of this study was to evaluate standardized automated segmentation procedures of spectral domain optical coherence tomography (SD-OCT) in imaging of age-related macular degeneration.

Twenty-nine eyes of 29 patients with neovascular age-related macular degeneration were included. Three groups were assigned, according to the predominant localization of extravasated fluid in the intra-, subretinal, and subretinal pigment epithelium (RPE) compartment. Automated segmentation procedures were evaluated in B scans of 512 × 128 × 1024 and 200 × 200 × 1024 scan patterns using SD-OCT (Cirrus). Alignment errors at the internal limiting membrane, actual RPE, and extrapolation of the physiologic RPE (RPE fit) were graded using a standardized classification system.

The rate of severe alignment failures was 56% and 41% for the 512 × 128 and the 200 × 200 raster pattern, respectively. Internal limiting membrane and actual RPE boundaries were most correctly delineated in the 200 × 200 raster pattern. Retinal pigment epithelium fit alignment was generally poor in 50% of scans. Retinal thickness values defined by internal limiting membrane and actual RPE segmentation were 90% accurate and not compromised by RPE fit misalignment. Subretinal fluid was demarcated most reliably. Alignment errors may occur together with a large spectrum of morphologic alterations.

Automated algorithms of SD-OCT demonstrate a substantial rate of alignment failures in the assessment of exudative age-related macular degeneration pathologies, which are usually associated with misinterpretation of boundaries at the (sub) RPE level.