Abstract
Purpose: :
In this study we compare the performance of a standard cortical grading methodology to that of a computer-based measurement methodology.
Methods: :
We used available statistical methods and image analysis techniques to create a grading algorithm for digital retro-illumination images. Severity grades assigned to simulated retro-illumination images by trained graders applying the Wilmer Ophthalmological Institute grading methodology were compared to both grades produced by the grading algorithm and the known severity of each image to assess agreement and bias. A measure of between- and within-grader variability resulted from using five graders to grade each image twice.
Results: :
The results of the simulation study indicate that the grading algorithm yields severity estimates with smaller bias when cataract severity is less than approximately 6 severity units on a 0 to 16 severity scale. However, the graders are less biased on higher severity images. On average, both methodologies have a bias equal to 0.77 severity units. The severity estimate from the grading algorithm has zero variability while the average within-grader variance of estimates from the human graders is 0.80.
Conclusions: :
Trained grader-based cortical cataract severity measurement methods perform well in a simulation study setting and the bias of the estimate is not substantially affected by the subjectivity of the assessment. Computer-based approaches, however, can improve upon bias and variability of grader-based methods in data sets limited to low severity cataracts.
Keywords: clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • cataract