Three-dimensional Reconstruction using a Single Two-dimensional Retinal Image

We have been developing a computer-aided diagnosis(CAD)system to detect abnormalities in color retinal images. Techniques have been proposed for automatic extraction of blood vessels(BVs), automatic measurement of diameter of BVs, erasing BVs, and detection of optic disk region in the works. In this study, these techniques were applied to automatic 3D reconstruction using a single 2D fundus image. The BVs were extracted by using mathematical morphology operator from the green component of the RGB value of color retinal images. Next, the BVs were erased from the image and then erased pixels were interpolated by using the RGB values of pixels in surrounding region. The diameter of BV was defined as minimum length of line segment between the BV walls. 3D configuration of optic disk region was reconstructed using intensity of pixel values in BV-erased image as information of the depth. The 3D fundus image was able to express that blood vessels passed along the vitreous side of the retina and the optic disk’s concave shape. It can be used as an assisting tool for test subjects to understand the anatomy of the fundus image.