Purchase this article with an account.
H. Li, J. Lim, J. Liu, D. Wong, N. Tan, C. Y. Cheung, P. Mitchell, A. G. Tan, J. J. Wang, T. Y. Wong; Assessment of a Computerized Nuclear Cataract Grading System. Invest. Ophthalmol. Vis. Sci. 2010;51(13):5666.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To develop and test an automatic grading system for nuclear cataract using slit-lamp images.
Our system was developed using digital slit-lamp images acquired with a digital Topcon DC-1 slit-lamp camera. The automatic nuclear cataract system consists of a segmentation algorithm that detect the anatomical structure of lens, a feature extraction algorithm, and a grade prediction algorithm. Anatomical structure consisting of the contour of lens and contour of nuclear region is detected using a modified active shape model. The 21-dimensional features to assess nuclear opacity were selected according to clinical grading protocol, which include intensity of sulcus, intensity ratio between nucleus and lens, and strength of nucleus edge. Support vector machine regression is employed for grade prediction, in which one hundred images were used as the training set.
The automatic grading system was tested using 5551 slit-lamp images from a population-based study, the Singapore Malay Eye Study. The lens structure was detected successfully for 95% of the tested images. Compared with the Wisconsin cataract grading system, the average differences of the automatic grading system is 0.36 on a 5.0 scale. The t-value is 2.74 and the p-value is 0.006.
An automatic grading system for nuclear cataract has been developed and tested. The system can achieve good agreement with the Wisconsin cataract grading system. The system will save workload for population study and improve objectivity of clinical grading.
This PDF is available to Subscribers Only