Automatic Detection of Choroidal Thickness Using Graph Theory and Directional Curvelet Transform From Normal and Diabetic Retinopathy OCT Images.

Abstract

The segmentation of choroidal layer could guide ophthalmologists in order to diagnose most of the pathologies of the eye like diabetic retinopathy (DR). Manually segmentation of this layer in optical coherence tomography (OCT) images is time-consuming, tiresome and dependent on human errors. To overcome these difficulties in this paper we have introduced new computer aided diagnosis based method for automatic segmentation of this layer. Methods: In this study we used curvelet transform, KSVD dictionary learning and Lucy-Richardson algorithm in order to speckle noise removal and enhancement of the OCT images. Then, the graph theory is used to determine the location of the inner choroidal boundary (ICB). In order to find the outer choroidal boundary (OCB), we defined the image histogram in a specific range depend on the average brightness of the image. The area between ICB and OCB considered as choroidal layer. Results: Our proposed method was evaluated on 60 EDI-OCT (Enhanced Depth Imaging Optical Coherence Tomography) images and by comparing the automatic segmentations with manual segmentations of ophthalmologists the average Dice’s Coefficient was 92.14% ± 3.30%. Also, by applying the latest presented open source algorithm by Mazzaferri et al on our dataset the mean Dice’s Coefficient calculated as 55.75 ± 14.54. Conclusions: In both normal eyes and eyes with diabetic retinopathy, we observed a great agreement between the manual segmentations and our automatic segmentations. Also the choroid was thinner in eyes with diabetic retinopathy. The experimental results show significant superiority of our proposed method over the proposed algorithm by Mazzaferri et al. Automatic segmentation of choroidal layer could also be useful for large-scale quantitative studies of the choroid.