Abstract
Purpose: :
To evaluate a new image alignment algorithm as a technique for improving the repeatability of stereometric parameters (specifically neuroretinal rim area, RA) obtained from scanning laser tomography [Heidelberg Retina Tomograph (HRT)]
Methods: :
Alignment of series of HRT images is inherently difficult because of magnification, rotation and translation shifts, and warping in the optic nerve head (ONH) position during scanning and data acquisition. Heidelberg Engineering has developed a new alignment algorithm to account for these difficulties using automated mosaicing and super–resolution techniques. This study examines the repeatability of rim area (RA) using the new algorithm as compared to the previous version (HRT software module 1.7.0.0). We applied both algorithms to a test–retest series of 74 patients with ocular hypertension or glaucoma (5 images per series) and a longitudinal series of 21 normal subjects (4 to 18 images per series acquired over 3.0 to 7.4 years). Improvement in repeatability was estimated by a reduction in standard deviation of RA afforded by the new alignment algorithm.
Results: :
There was clear qualitative improvement in image repeatability in HRT series where an obvious misalignment event had occurred. However, the difference in repeatability of the RA measurement was not statistically significant in the test–retest data (P=0.24) or the longitudinal normal data (P=0.89). However, in a small sub–sample, where obvious misalignment events had occurred, the improvement in repeatability was marked.
Conclusions: :
A new image alignment algorithm improves the repeatability of some longitudinal series of HRT images of the ONH but does not seem to have a significant effect on the repeatability of the RA parameter. Improvements may manifest in other stereometric parameters or longitudinal analyses that are conducted at a pixel or super–pixel level.
Keywords: imaging/image analysis: non-clinical • optic disc