Automated method for RNFL segmentation in spectral domain OCT

Abstract

We introduce a method based on optical reflectivity changes to segment the retinal nerve fiber layer (RNFL) in images recorded using swept source spectral domain optical coherence tomography (OCT). The segmented image is used to determine the RNFL thickness. Simple filtering followed by edge detecting techniques can successfully be applied to segment the RNFL from recorded images and estimate RNFL thickness. The method is computationally more efficient than previously reported approaches. Higher computational efficiency allows faster segmentation and provides the ophthalmologist segmented retinal images that better utilize advantages of spectral domain OCT instrumentation. OCT B-scan and fundus images of the retina are recorded for 5 patients. The segmentation method is applied on B-scan images recorded from all patients. An expert ophthalmologist separately demarcates the RNFL layer in the OCT images from the same patients in each B-scan image. Results from automated image processing software are compared to the boundary demarcated by the expert ophthalmologist. The absolute error between the boundaries demarcated by the expert and the algorithm is expressed in terms of area and is used as an error metric. Ability of the algorithm to accurately segment the RNFL in comparison with an expert ophthalmologist is reported.

Optimized retinal nerve fiber layer segmentation based on optical reflectivity and birefringence for polarization-sensitive optical coherence tomography

“…Although the current approach is computationally efficient, it tends to overestimate RNFL thickness as previously mentioned and sometimes it misdetects the RNFL boundary (especially on low quality images). Multiple sophisticated intensity-based RNFL boundary detection algorithms are available and are expected to provide better estimation of intensity-based RNFL boundaries than our current implementation [14][15][16][17][18][19] .…”

“…PSOCT polarimetric processing algorithms are sensitive to slight misdetection of RNFL boundaries. Various sophisticated RNFL segmentation approaches have been reported [14][15][16][17][18][19] . Mujat et al…”

“…Garvin et al 15 presented a 3-D graph search approach for OCT retinal layer segmentation. Paranjape et al 16 introduced a simple and fast segmentation method based on a Sobel filter for automated RNFL boundary detection. Mishra et al 17 developed a two-step scheme which first identifies the approximate locations of the *rylander@mail.utexas.edu; phone 1-512-471-1995 RNFL and then performs a local kernel-based optimization to refine segmentation results.…”