Automatic Segmentation of Corneal Microlayers on Optical Coherence Tomography Images

Abstract

Purpose To propose automatic segmentation algorithm (AUS) for corneal microlayers on optical coherence tomography (OCT) images. Methods Eighty-two corneal OCT scans were obtained from 45 patients with normal and abnormal corneas. Three testing data sets totaling 75 OCT images were randomly selected. Initially, corneal epithelium and endothelium microlayers are estimated using a corneal mask and locally refined to obtain final segmentation. Flat-epithelium and flat-endothelium images are obtained and vertically projected to locate inner corneal microlayers. Inner microlayers are estimated by translating epithelium and endothelium microlayers to detected locations then refined to obtain final segmentation. Images were segmented by trained manual operators (TMOs) and by the algorithm to assess repeatability (i.e., intraoperator error), reproducibility (i.e., interoperator and segmentation errors), and running time. A random masked subjective test was conducted by corneal specialists to subjectively grade the segmentation algorithm. Results Compared with the TMOs, the AUS had significantly less mean intraoperator error (0.53 ± 1.80 vs. 2.32 ± 2.39 pixels; P < 0.0001), it had significantly different mean segmentation error (3.44 ± 3.46 vs. 2.93 ± 3.02 pixels; P < 0.0001), and it had significantly less running time per image (0.19 ± 0.07 vs. 193.95 ± 194.53 seconds; P < 0.0001). The AUS had insignificant subjective grading for microlayer-segmentation grading (4.94 ± 0.32 vs. 4.96 ± 0.24; P = 0.5081), but it had significant subjective grading for regional-segmentation grading (4.96 ± 0.26 vs. 4.79 ± 0.60; P = 0.025). Conclusions The AUS can reproduce the manual segmentation of corneal microlayers with comparable accuracy in almost real-time and with significantly better repeatability. Translational Relevance The AUS can be useful in clinical settings and can aid the diagnosis of corneal diseases by measuring thickness of segmented corneal microlayers.

Pathological-Corneas Layer Segmentation and Thickness Measurement in OCT Images

“…Manual measurement of the thickness is challenging for three reasons. First, manual segmentation has low repeatability and reproducibility, 11 and it is time-consuming. 11 – 13 Second, points of each boundary need to be corrected for both distance and direction using multiple refractive indices for the cornea, but mistakenly points are corrected for only distance using one refractive index for the cornea.…”

“…First, manual segmentation has low repeatability and reproducibility, 11 and it is time-consuming. 11 – 13 Second, points of each boundary need to be corrected for both distance and direction using multiple refractive indices for the cornea, but mistakenly points are corrected for only distance using one refractive index for the cornea. 14 Third, the thickness measurement is measured as the difference in the axial coordinates, which is not accurate.…”

“…In addition, they focus on the regular-shaped cornea whose layer boundaries can be modeled using prior known models. 11 – 13 , 15 – 26 In these methods, the corneal layer boundaries are modeled using circles, 22 parabolas, 11 – 13 , 16 , 17 , 19 fourth-order polynomials, 15 , 18 ellipses, 24 , 25 or Zernike polynomials. 23 Some of these methods first extract the highest intensity points of each boundary from the OCT image or its gradient and fit them to a model to estimate the boundary.…”

“… 23 Some of these methods first extract the highest intensity points of each boundary from the OCT image or its gradient and fit them to a model to estimate the boundary. 11 , 13 , 15 , 17 , 22 The accuracy of these methods depends on the quality of the OCT images and they are most likely to fail in images with low signal to noise ratio (SNR). In other methods, instead of extracting the highest intensity points, graph theory and dynamic programming or level sets are used to search for the minimum-cost paths in the image, according to a designed cost function, which is more likely to pass through the boundary.…”

“…We compare our method to three counterpart methods. 11 – 13 We trained our method using 60 images, and evaluate it using 130 OCT images from multiple datasets. We use a 2D ray-tracing algorithm to correct the locations of the layer boundary due to the refraction of the imaging light, and then we measure the thickness profiles for different corneal layers using the shortest distance.…”