Abstract
Purpose: :
To determine if a novel automatic segmentation program for spectral domain optical coherence tomography (SDOCT) can be applied to segmentation of the neurosensory retina and the measurement of retinal volume (RV) and retinal thickness (RT) in eyes with diabetic macular edema (DME).
Methods: :
We applied a general segmentation framework based on graph theory and dynamic programming (Chiu et al., IOVS 2012) to segment the boundaries of the internal limiting membrane (ILM) and Bruch’s membrane in bitmap output images of 49 horizontal B-scans per baseline Spectralis SDOCT scan volume for 17 eyes with diabetic macular edema. The automatic segmentation results were compared with RV and CFT results from an intrinsic SDOCT segmentation algorithm.
Results: :
Mean RV in the 1-mm diameter central foveal subfield area was 0.37±0.09 and 0.37±0.07mm³ for intrinsic SDOCT segmentation and our automatic segmentation, respectively. Mean RT of the same area was 476.88±113.68 and 470.80±87.94 µm, respectively. Mean differences in the measured RV and RT were not significantly different between the groups (P=0.9314 and P=0.9040, respectively). Mean RV and RT in surrounding superior, nasal, inferior, and temporal regions (out to 3mm) also did not show significant differences between the groups.
Conclusions: :
This novel automatic segmentation program can be applied to Spectralis output images for automatic segmentation between ILM and Bruch’s membrane and calculation of RV and RT in DME eyes. Further study will be warranted to identify whether automatic segmentation algorithms can be accurately and reproducibly applied for SDOCT segmentation of individual intraretinal layers in eyes with DME-associated pathology, and whether it can be applied to multiple clinical SDOCT platforms.
Keywords: image processing • diabetic retinopathy • edema