Identification of retinal cells in in-vivo high resolution images

Abstract

Recent advances in the acquisition of in-vivo high resolution retinal images through the use of Adaptive Optics (AO) have allowed the identification of cellular structures such as cones and rods, in and out of the fovea, in such a way that their histological characteristics can be studied in-vivo and later compared to data obtained post-mortem. In this work, an algorithm is proposed for the detection of photoreceptors; it consists of two stages: Early Cell Detection (ECD), to detect all candidate cells, and Refinement of Cell Detection (RCD), to reduce over-detection of photoreceptors. The algorithm has been tested using synthetic and real images, the latter acquired with an Adaptive Optics Scanning Light Ophthalmoscope (AOSLO). The proposed algorithm was compared against the one developed by Li and Roorda, and both algorithms were tested on synthetic and real images, yielding similar algorithm performance on both kinds of images when they had only cones; however, the algorithm developed by Li and Roorda, when applied to real images having cones and rods, identifies photoreceptors in vascular tissue, in addition to showing low rod detection.