Fast retinal layer identification for optical coherence tomography images

Fast method for identifying the internal limiting membrane (ILM) and retinal pigment epithelium (RPE) from optical coherence tomography images is demonstrated. To avoid unnecessary increment of calculation time, a strong downsampling of the original data set is performed to reduce a number of processed pixels. In ILM segmentation, the obtained data cube is filtered with two different kinds of parameters and two estimates for the position of ILM is determined. A simple smoothness value is determined for both estimates and better estimate is used for future processing. A smaller portion of pixels around estimated ILM are extracted from the down sampled data and filtered again and new estimation for ILM position is determined. That procedure is repeated with smaller portion of pixels around ILM and with different filtering parameters. The principle of RPE segmentation is very much similar with ILM identification. Only the used filtering and processing parameters are changed. Algorithm was tested with eight data sets with good reliability. Over 97% of each scans had smaller segmentation error than 5 pixels. Total required data processing time (ILM and RPE segmentation) for data volume with (600x1500x128) pixels was less than 9 seconds.