R. Srivastava, X. Gao, F. Yin, D. Wong, J. Liu, C.Y. Cheung, T.Y. Wong, Exploring image gradients for nuclear cataract grading, in Proc. 15th International Conference on Biomedical Engineering, Singapore, 2013
Nuclear Cataract (NC) is the most common type of cataract and can be automatically diagnosed from slit-lamp images of the eye lens. The diagnosis can be based on two cues, 1. brightness and color of the lens, which has been used by most of the researchers and 2. visibility of parts of the lens, which has not been explored much. The main contribution of this paper is in utilizing gray level intensity gradient based features for computerized grading of NC. The idea behind the proposed system, called ACASIA-NC, is that clear visibility of landmarks in a healthy eye leads to distinct edges in the lens region. While for advanced stages of NC, the edges in this region fade, since the landmarks are vaguely visible. To capture edge information in the lens region, features related to grayscale image gradient have been extracted. Experiments performed on a large dataset of over 5000 slit lamp images show that the proposed features outperforms state-of-the-art in automatic NC grading, both in terms of speed and accuracy.