|
A new technique for making a three dimensional map of the optic nerve head is expected to have a major impact on the way in which glaucoma is diagnosed and treated. The new technique, contour photography, allows the health of the optic nerve head to be objectively evaluated every six months during the patient's routine office visit. In contour photography, a set of parallel lines of light are projected into the patient's eye and the back of the eye is photographed using a standard camera that is available in almost all ophthalmologist's offices. The three dimensional information is encoded in the positions of the photographed lines, and is decoded by treating each stripe as the intersection of a plane of light with the fundus. At present, a trained human observer identifies the edges of the stripes. In order to decrease the data extraction time, several automated edge detection algorithms were examined for their suitability in the analysis of contour photographs, and the best was extensively evaluated using Monte Carlo simulation. The accuracy and reproducibility of the edge position estimate in images with various amounts of film grain noise were measured for many values of the edge detector parameter and of the signal parameter, its modulation, m. When normalized by the amount of film grain noise, the relationship between reproducibility and l/m was found to be linear over the range of parameters likely to be encountered in contour photography. The accuracy was found to be independent of the amount of film grain noise, and linearly related to 1/m. By estimating m for each edge, the accuracy could be treated as a correctable systematic error of the edge detection process. A sample calculation which used parameter values that are likely to be found in contour photography showed that the automated edge detection process would be expected to produce a random variation in the measurement of the depth of the optic nerve head surface whose standard deviation is 0.5 micrometers, or 0.05 percent of the deepest part of the surface.
|
