Purchase this article with an account.
M. M. Schulze, T. L. Simpson, N. Hutchings; Cross-Correlating Between Grading Scales of Redness - Is It Possible?. Invest. Ophthalmol. Vis. Sci. 2007;48(13):5369.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Objective image processing metrics were derived to allow comparison between 4 grading scales of bulbar hyperemia.
Image metrics of fractal dimension (FD, describing vessel branching) and % pixel coverage (PC, describing vessel coverage) were selected to describe the severity of bulbar hyperemia for each grading scale (McMonnies/Chapman-Davies, CCLRU, Efron & CCLR). Due to size and resolution differences between scales, two sets of image metrics were obtained. The first was determined for the original image size and the largest region of interest (ROI) within the bulbar conjunctiva. The second was determined after downscaling the images (to match for size & resolution across scales) and for a fixed ROI size. For all images, thresholding, filtering and background subtraction were applied to account for differences between perspective of the scales and image noise. The association between each set of image metrics and the reference image grades was determined with Pearson correlation coefficients.
Correlations were high between reference image grades and both sets of objective metrics (all Pearson’s r’s ≥0.85, p<0.05). The level of correlation between the image metrics and the reference grades was dependent upon the ROI considered. The pixel coverage for the fixed ROI had a lower correlation for the McMonnies/Chapman-Davies scale and a higher correlation for the CCLR scale (p≤0.05). Correlations between FD and reference images increased for the CCLR & CCLRU scales with the fixed ROI and reduced for all others (all scales, p≤0.05). Correlations for the CCLR scale were least affected by the pre-processing procedures (CCLR: range 0.97 - 0.99, CCLRU: 0.93 - 0.99, Efron: 0.95 - 0.99, McMonnies/Chapman-Davies: 0.85 - 0.90).
Fractal metrics and pixel coverage enable comparison between various grading scale references. However, the complexity of pre-processing required, because of non-standardized image acquisition in particular, makes complete calibration (over the full range) of all scales problematic.
This PDF is available to Subscribers Only