Automated segmentation of the optic disc from stereo color photographs using physiologically plausible features

Abstract

Purpose: To evaluate a novel automated segmentation algorithm for cup-to-disc segmentation from stereo color photographs of patients with glaucoma for the measurement of glaucoma progression.

Methods: Stereo color photographs of the optic disc were obtained by using a fixed stereo-base fundus camera in 58 eyes of 58 patients with suspected or open-angle glaucoma. Manual planimetry was performed by three glaucoma faculty members to delineate a reference standard rim and cup segmentation of all stereo pairs and by three glaucoma fellows as well. Pixel feature classification was evaluated on the stereo pairs and corresponding reference standard, by using feature computation based on simulation of photoreceptor color opponency and visual cortex simple and complex cells. An optimal subset of 12 features was used to segment all pixels in all stereo pairs, and the percentage of pixels assigned the correct class and linear cup-to-disc ratio (LCDR) estimates of the glaucoma fellows and the algorithm were compared to the reference standard.

Results: The algorithm was able to assign cup, rim, and background correctly to 88% of all pixels. Correlations of the LCDR estimates of glaucoma fellows with those of the reference standard were 0.73 (95% CI, 0.58-0.83), 0.81 (95% CI, 0.70-0.89), and 0.86 (95% CI, 0.78-0.91), respectively, whereas the correlation of the algorithm with the reference standard was 0.93 (95% CI, 0.89-0.96; n = 58).

Conclusions: The pixel feature classification algorithm allows objective segmentation of the optic disc from conventional color stereo photographs automatically without human input. The performance of the disc segmentation and LCDR calculation of the algorithm was comparable to that of glaucoma fellows in training and is promising for objective evaluation of optic disc cupping.

Fig. 1

Stereo color pair of the optic nerve head of the left eye after alignment and cropping to 512×512pixels. The left and right stereo images can be fused by holding the page at about 30cm (12″) distance from the eye and gazing into infinity.

Fig. 2

A, B, C: Gradings of an optic disc stereo pair by three faculty glaucoma specialists. Rim in grayish, cup in whitish, shown on the left image of the stereo pair from Figure 1. D: reference standard created from A, B, C with white, cup, gray, rim, and black, background

Figure 3

Example of the variety of color opponency steerable Gaussian filterbank kernels. Kernels oriented as if the local gradient was horizontal. First row, from left to right dark-bright opponency kernels for zeroth order, first order 0° to the local gradient, first order 90° to the local gradient, second order 0° to the local gradient, second order 60° to the local gradient and second order 150° to the local gradient, at scale σ 32 pixels. Second row, same for scale σ 64 pixels, and third row, 128 pixels. Next three rows same for blue-yellow opponency kernels, and last three rows, same for red-green kernels. Smaller scales not shown because they are difficult to see. These kernel images also represent the response of each of the features to an impulse function.

Fig. 4

Stereo pair and its most discriminant pixel features. A,B: the left and right stereo image of the stereo pair. Rows below that show images of the most dominant twelve features in rank order (see Table 1 for a description of these features).

Fig. 5

Classification of stereo pairs by glaucoma faculty, three glaucoma fellows, and pixel feature classification. Rows 1,2,3,4: small, medium and large disc excavation and excavation with inferior notching. From left to right: left and right images of the stereo pair; reference standard by three glaucoma faculty; grading by fellow A, by fellow B and by fellow C; and grading by pixel feature classification.

Fig. 6

Scatterplots of linear cup-to-disc ratio estimates from all 58 stereo pairs by three glaucoma fellows and the pixel feature classification algorithm (vertical axes), against the reference standard grading by three glaucoma faculty (horizontal axes). Clockwise, fellow A, B, the algorithm and fellow C respectively. ‘r’ is the correlation coefficient for each of these estimates with the reference standard.