Validation of a semi-automated computer-aided technique for quantifying corneal vascularisation and scarring

To determine the validity and reproducibility of a semi-automated computer-aided method of quantifying corneal vascularisation and scarring.

Software for quantifying corneal vascularisation and scarring was devised and its validity was tested. Three masked clinicians viewed 20 corneal photographs and estimated the percentages of the cornea that were (a) scarred and (b) vascularised. The images were also analysed using the software. Two months later, the photographs were re-graded and a further computer-derived data set was obtained. VALIDITY was assessed using the Pearson correlation coefficient and reproducibility using the Bland-Altman method.

There was good agreement between the mean clinician and software assessment of corneal scarring, Pearson r=0.91. There was more moderate agreement between the mean clinician and software assessment of corneal vascularisation, Pearson r=0.82. Reproducibility: The software was 95% reproducible within 1.36% (mean difference -0.14%, SD 0.62%) when assessing scarring and within 0.83% (mean difference -0.19%, SD 0.33%) when assessing vascularisation. Ophthalmologists were, at best, only reproducible within 15.53% for quantifying corneal scarring and 10.1% for quantifying corneal vascularisation.

Computer-aided analysis of corneal photographs provides a reproducible method for quantifying corneal vascularisation and scarring and has the potential to be a valuable tool for investigating treatments of corneal disease.