Abstract
Purpose: :
To examine if the image quality metric as currently determined by Mean Pixel Height Standard Deviation (MPHSD) in scanning laser topography [Heidelberg Retinal Tomography (HRT)] is an accurate estimate of repeatability and true image quality.
Methods: :
MPHSD is an image quality metric used on the HRT software to quantify the repeatability of mean topography images. Previously we have developed a computer based simulation to replicate noise inherent in topography images of the ONH (Patterson et al 2005 IOVS 46: 1659–67.). The simulation applies translational, rotational and Gaussian noise components. This noise was applied to 74 single topography images each from a distinct ocular hypertensive or glaucomatous patient. A simulated set of 3 single topographies was generated for each ONH by the application of an identical set of 3 noise components to each of the single topographies. It was from these 74 sets of 3 single topographies that 74 mean topography images were generated. The associated individual pixel height standard deviation values were calculated at each pixel in all mean topographies. From these a MPHSD value was calculated for the corresponding mean topography. Since the noise was identical we would expect the MPHSD values to be identical if it were an accurate estimate of true image quality.
Results: :
The range of values for MPHSD for the 74 patients was very large (8.7 to 106.4 µm). The distribution of pixel height standard deviations values for each mean topography was found to be positively skewed. Qualitatively there seems to be a clear association between the number of 'features' and 'edges' in an image and the size of the MPHSD.
Conclusions: :
The range of MPHSD values varied considerably despite identical noise being applied to all single topographies. As the distribution of pixel height standard deviation values is skewed the mean may not be an appropriate summary measure to characterise the data. We plan to evaluate other metrics to summarise the repeatability of topography images.
Keywords: imaging/image analysis: non-clinical • optic disc • image processing