Standardization and analysis of digitized photographic data in the longitudinal documentation of cataractous growth.

Abstract

Age-related cataract formation in man can be documented with slit and retroillumination photographs. With digitization and image analysis of such photographs a cataract may be characterized by a frequency distribution of picture elements over a 255 step gray scale spectrum. Transition from a clear to a cataractous lens may be manifested as a change from a unimodal, Gaussian to a multimodal, non Gaussian frequency distribution respectively. How should one compare and contrast these two distributions, so to accurately describe the extent and significance of a change in lens opacification? The in vitro system of cold cataract formation in the rabbit lens was used as a model of the much slower process of age-related cataract formation in man. As in the human lens undergoing progressive opacification, the frequency distribution (number of pixels vs. intensity of gray) for a digitized image of a clear lens at 26 degrees C is unimodal and Gaussian; that of a fully developed cold cataract at 10 degrees C is multimodal and non-gaussian. In spite of the increasing multimodality of the frequency distribution as the temperature dropped and the cataract grew in density and size, the mean gray density proved to be a valid and useful measure to characterize the distribution and to compare different unaligned images. The Wilcoxon Rank Sum Test proved to be useless in comparing the frequency distributions from cataract images because it proved to be too sensitive to subtle changes in the degree of opacification. Anomalous behavior of the opacification process--i.e. clarification as well as opacification of the lens during cold cataract formation caused all pairs to appear statistically significantly different when in appearance there was no difference. The mean of the frequency distribution is less sensitive to this anomalous behavior and is useful as a comparative index. The method of calculating the threshold of significant change in the mean density of a cataract image is presented.