Purpose: Compare estimates of the prevalence of nuclear cataract made by subjective means using the Wilmer system and our new objective classification scheme. Methods: The classifier was developed using digitized 35mm slides acquired with a Topcon SL5-E slitlamp camera. It consists of a neural network that was trained using the subjective scores rendered by the Wilmer system. Our procedure consisted of: (1) specification of the region of interest, (2) image preprocessing, (3) extraction of classification features, and (4) use of these features to yield opacification grades. For the comparative prevalence study, a series of photographs of the eyes of 200 patients acquired at our Salisbury Eye Evaluation (SEE) clinic were digitized and evaluated using the previously trained neural network. To assess reproducibility, 48 pairs of photos of the same eyes taken on the same day were evaluated. Results; Prevalence of nuclear cataract estimated by the Wilmer system for at least grade 2 was 44.3%. The objective classifier agreed closely. In the reproducibility evaluation, the objective classifier displayed confidence intervals that were approximately one half those of the Wilmer system. Conclusions: We have developed and tested a sensitive, reproducible classification system for nuclear cataract. Estimates of the prevalence of nuclear cataract made using this system compare well with those made using the Wilmer classification system. Better reproducibility of the objective classifier suggests a higher effective sensitivity which would be invaluable in assessment of progression. Supported by N1H Grant R01EY10857-02. None.
|Original language||English (US)|
|Journal||Investigative Ophthalmology and Visual Science|
|State||Published - Dec 1 1997|
ASJC Scopus subject areas
- Sensory Systems
- Cellular and Molecular Neuroscience