Abstract
Purpose :
Automated multilayer segmentation (MLS) methods determine the desired inner layer boundaries. However, they are prone to layer misidentification errors, especially in eyes with retinal lesions and data with poor quality. The inner layer boundaries for such cases are often not identifiable manually. This study proposes an automated outer boundary of inner plexiform layer (IPL) segmentation approximation method using OCTA volumes.
Methods :
MLS detects IPL segmentation failure using internal limiting membrane (ILM)-IPL and ILM-outer plexiform layer (OPL) angiography slabs generated based on ILM segmentation and the outer boundary of OPL segmentation. This assumes that the ILM and OPL segmentations are correct. It is expected that the local similarity, measured by normalized cross correlation (NCC), between ILM-IPL and ILM-OPL angiography slabs is low if MLS IPL segmentation malfunctioned as these slabs are generated based on maximum projection. If the variance of NCC is smaller than a threshold, then the MLS IPL segmentation is replaced by IPL segmentation approximation as weighted average of the ILM segmentation and OPL segmentation, otherwise MLS IPL segmentation is used.
Performance of the algorithm is evaluated using 161 Angiography volume data over 3x3 mm (76 scans), 6x6 mm (67 scans), 8x8 mm (2 scans), 12x12 mm (7 scans), HD 6x6 mm (6 scans), HD 8x8 mm (3 scans) acquired using CIRRUS™ HD-OCT 6000 with AngioPlex® OCT Angiography (ZEISS, Dublin, CA). Data include a mix of retinal diseases. A clinical grader evaluated each superficial retinal layer (SRL) slab generated with the new algorithm as success or failure.
Results :
Fig 1 shows two examples of SRL using MLS IPL segmentation and IPL approximation where the MLS IPL segmentation is incorrect. The IPL approximation creates a more accurate representation of SRL for these cases. MLS IPL segmentation of 34 (21%) of the scans were replaced by IPL approximation. Success rate of MLS without IPL approximation and with approximation is 79% and 96% respectively.
Conclusions :
We proposed a new IPL segmentation approximation method. Our method creates acceptable SRL slabs when MLS IPL segmentation performance suffers from severe retinal disease or poor image quality. Automated inner retinal segmentation approximation may be a valuable diagnostic tool for retinal diseases.
This is a 2020 Imaging in the Eye Conference abstract.