An automatic method of delineating the position and thickness of the four outer retinal hyperreflective bands in optical coherence tomography images was developed using ImageJ. A manual method was developed in order to validate the automatic method.
Optical Coherence Tomography (OCT) is a non-invasive method of measuring the light reflective properties of material. OCT is particularly effective for revealing the retinal structure of the eye because the tissue is mostly transparent to light allowing a full depth of imaging. OCT makes reflectance measurements along a single depth path called an A-scan. Multiple A-scans are combined to create a 2D image of a slice of the retina called a B-scan. Multiple B-scans are combined to create a 3D volume of data.
The condition of the four outer hyperreflective bands in the retina have been shown to correlate with various disease states. The recovery from the disease states after treatment is also apparent in the condition of these bands.
An automatic method for the determination of the position and thickness of the bands was developed using ImageJ. The method is focused on determining information about the bands rather than a delineation of regions containing the bands. The method utilizes multiple steps that determine information about the bands and analyze the information for quality prior to passing it on to the next step. The first step is to bound the retina using a stack of Gaussian filtered images; the second step is to find potential locations of the bands; the third step is to refine the band positions and to get an initial thickness using region and edge detection image processing techniques; the fourth step is to refine the band position and thickness using Gaussian function fitting; the fifth and last step is to filter the function fit values. The Gaussian function fitting is particularly important as it accounts for the interaction of reflective response of the underlying tissue.
The image stack structure of ImageJ is a natural methodology for a presentation of the B-scan images of the OCT. The stack structure was again used to graph the A-scan data along with the functions fit to that data to determine band position and thickness. Not having to go to an external program for the graphing of the plots and having the data displayed as a sequence of images that can be rapidly scrolled through allows a visual evaluation of the effectiveness of the fit for each A-scan.
The ImageJ sample code of the Bezier Curve and a GUI example extending the StackWindow were the starting points for creating a manual band specification interface to generate data for comparison to the automatic method of band extraction. The Bezier curve example was extended to a non-uniform spline fit with double buffered window. The GUI StackWindow example was extended to create an interface with multiple radio button, checkbox and keyboard interface options. Many of the ImageJ interface capabilities are still passed through to the standard interface such as zoom and pan.
Automatic, Segmentation, Ophthamology, Retina, OCT
Presenting author: Douglas H. Ross, Kenneth R. Sloan
Organisation: University of Alabama at Birmingham (UAB)
co-authors: Douglas H. Ross
Kenneth R. Sloan