Fig 1.
(a) Near infra-red image; (b) Volume or 3D reconstruction of the retina from OCT scans; (c) A B-scan image (cross-section of the retina through the green line in Fig a and b), (d) Retinal Layers are delineated in a B-scan image, (e) and (f) show Optic disc centred enface and SD-OCT B-scan respectively, (g) and (h) show drusen and hyper-reflective intra-retinal spots (HIS) in SD-OCT B-scan respectively.
Fig 2.
The flow diagram of the proposed classification method.
Fig 3.
The basic flow diagram of the segmentation algorithm.
The basic flow diagram of our proposed 3D segmentation.
Fig 4.
MZ-EZ boundary detection steps of SD-OCT B-Scan Image.
(a) SD-OCT B-Scan image; (b) Image after applying Canny edge detection highlighting edge pixels; (c) Image with highlighted edge pixels having positive intensity gradient; (d) Image with the candidate pixels in the ROI which is defined by the upper and lower boundaries depending on the target boundary; (e) The partial graph of the full connected graph representation of the boundary detection problem; (f) A magnified image of the red region of (d), each colour represents a different pixels-group and black circles represent the end pixels which are the nodes of the graph; (g) Image with highlighted pixel-groups lying on the shortest path of the graph; (h) The MZ-EZ Boundary (yellow line) is detected after fitting a curve.
Fig 5.
An example of HIS segmentation.
(a) an SD-OCT B-scan (b) manual ILM-RNFL boundary (red line) and HIS (green colour) (c) automatically detected ILM-RNFL boundary (red line) and HIS (green colour).
Fig 6.
An example of drusen segmentation.
(a) an SD-OCT B-scan with delineation of drusen by the blue color (b) Drusen in 3D view of an SD-OCT volume of an AMD patient.
Fig 7.
The curviness of different MZ-EZ boundaries.
The curviness of different MZ-EZ boundaries (red colour) with a different curve using our proposed method. RBC boundary is in green colour.
Fig 8.
A volumetric image of the retina and the choroid.
(a) A 3D render image of the retina with choroid constructed from an SD-OCT volume. (b) Segmented layers of the retina and choroid (c) The complex of the EZ, IZ, and RPE in a different colour in the gray-scale retinal SD-OCT image.
Table 1.
Constructed datasets from four sources.
Table 2.
Performance of four state-of-the-art and proposed methods on partial Duke dataset (D1) considering only normal and DME patients (because the Venhuizen et al., Lemaitre et al., and Sidibe et al. dataset only used these images).
Table 3.
The metric in mean (standard deviation) of 10 iterations on 15-fold cross-validation test for the proposed classification model using Random Forest.
Table 4.
The accuracy for different machine-learning algorithms for the classification model based on the proposed features on dataset D2.