Abstract
Purpose:
To develop a computer program that determines whether there is a retinal nerve fiber layer (RNFL) defect in a given fundus image and, if there is, where it presents.
Methods:
The given red-free fundus image was converted to non-extorted image (the optic disc center was at the level of the macula). The intensity profile was normalized to enhance the contrast. The region of interest (ROI) was set as the circumferential area surrounding the optic disc (internal diameter: 2 disc diameters [DD], external diameter: 3 DD). The temporal half of the ROI was converted to a polarimetric image. After removing the blood vessels with a Frangi filter, the differential gradients were calculated along the angular axis. The average curvature was calculated sector-by-sector based on the gradient data, and the local maximum value was obtained. If the maximum value was greater than the cut-off value, the sector was considered to be an RNFL defect. The images of 100 normal healthy controls and 100 open-angle glaucoma patients were enrolled as age- and sex-matched samples. When both of a subject’s eyes were eligible, the image of one eye was randomly selected. Maximum curvatures of two groups were compared and an ROC analysis was performed to determine the reliability of this system and to set the optimum cut-off value, as the minimum acceptable sensitivity was set to 70.0%.
Results:
The mean ages of the control group and glaucoma group subjects were 57.5 ± 12.8 and 58.7 ± 11.2 years, respectively (p=0.456). In the control group, fifty-two of the 100 subjects were male, while in the glaucoma group, 45 of 100 were male (p=0.322). In the glaucoma group, the mean deviation was -4.90 ± 5.40 dB. There was a significant difference of maximum curvature between the two groups (14.37 ± 5.13 in control, 20.67 ± 10.56 in glaucoma group, p<0.001). In the ROC analysis, the area under curve (AUC) was 0.711. The probability of asymptomatic significance (less than 0.001) and the asymptomatic confidence interval (95% CI; 0.639 - 0.782) suggested that the maximum curvature is highly associated with RNFL defect. Considering a minimum acceptable sensitivity of 70.0%, the specificity, for a cut-off value of 14.84 was determined to be 62.0%.
Conclusions:
The proposed computer program could be an effective screening and early-detection tool for physicians.
Keywords: 550 imaging/image analysis: clinical •
549 image processing