Automated Detection of ELM Disruption Regions in 3D OCT Images

Abstract

 

Purpose:

 

Disruption of External Limiting Membrane integrity on SD-OCT is associated with lower visual acuity outcome in patients suffering from Diabetic Macular Edema (DME), probably related to disalignment of the photoreceptor inner segments. However, no automated method to estimate ELM from SD-OCT exists.

 

Methods:

 

5 subjects diagnosed with clinically signficant DME were included and underwent macular-centered SD-OCT (Heidelberg Spectralis, 512×19×496 voxels, 10.98×239.66×3.87µm³). 5 subjects w/o retinal thickening and normal acuity were also scanned (Carl Zeiss, 200x200x1024 voxels, 30×30×2 µm³). Automated estimation of local ELM disruption was achieved as follows. First, eleven surfaces are automatically segmented using our standard 3D graph search approach [1], the sub-volume between surface 7 and 11 containing the ELM region is flattened based on the segmented retinal pigment epithelium (RPE) layer, a second, edge-based graph search surface-detection method segments the ELM region in close proximity "above" RPE, and each ELM A-scan is classified into disrupted or not disrupted based on the ELM voxel texture.

 

Results:

 

In patients with DME, large areas of disrupted ELM are clearly visible (Fig 1). In normals, numerous pinpoint areas of 50-100 µm ELM disruption exist, but no larger areas.

 

Conclusions:

 

Automated estimation of ELM Disruption in patients with DME has been achieved. Though this pilot study concerned Spectralis scans, the algorithm can be applied to SD-OCT images from any manufacturer. In normal subjects the ELM has pinpoint disruptions, which apparently does not affect function, though their full import remains unclear. We have started determining the relationship of quantitative ELM disruption to visual acuity in this type of patient.  

 

Keywords: image processing • imaging/image analysis: clinical • imaging/image analysis: non-clinical