Purchase this article with an account.
N. O'Leary, D. P. Crabb, P. G. Schlottmann, D. F. Garway-Heath; A Statistical Method for Detecting Change in Series of Stratus OCTTM Measurements. Invest. Ophthalmol. Vis. Sci. 2007;48(13):3335.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To illustrate a quantitative method for detecting change in series of retinal nerve fibre layer thickness (RNFLT) measurements from the StartusOCTTM. To compare the false positive rate of the new method with analysis of average and sectoral RNFLT changes.
Statistic Image Mapping (SIM) is a proven method used in neuroimaging to quantify areas of change in images of the brain and has recently been applied to retina tomography images1. We adapt these techniques to process longitudinal series of 256 RNFLT measurements on fixed diameter scanning circles about the ONH from the StratusOCTTM. The technique produces maps of rates of RNFLT thinning using a permutation technique based solely on the patient’s own data. Clusters of significant change are assessed by using a method that accommodates multiple statistical testing at several points at once: this allows for the detection of small (local) areas of activity whilst controlling for overall false positive change. Results from this method are compared with ordinary linear regression of mean RNFLT (and mean sectoral RNFLT) over time in series of images from 20 subjects; these StratusOCTTM series were acquired in a short period of time as part of test-retest study and thus provide a test-bed for false positive detection.
2 out of 20 subjects were falsely determined as having significant thinning in RNFLT over time using ordinary linear regression of average and sectoral thickness values whilst only 1 false positive result was recorded for SIM at the same level of statistical significance.
Statistic image mapping can be used to quantify change in series of RNFLT measures from longitudinal StratusOCTTM images. Results from this sample suggest that the new technique, designed for detecting relatively small localised thinning, has equivalent specificity to methods that by definition will only detect global or large changes. Sensitivity and utility of the method will be established in other longitudinal data sets. Any statistical method for detecting change in StratusOCTTM image series is dependent on good quality imaging, robust RNFLT segmentation algorithms and successful ‘alignment’ of the scanning circles in time.1. Patterson AJ, Garway-Heath DF, Strouthidis NG, Crabb DP (2005). Invest Ophthalmol Vis Sci 46: 1659-1667.
This PDF is available to Subscribers Only