Purchase this article with an account.
E.S. Barriga, G. Zamora, S. Russell, P. Soliz; Detection of Spectral Variations Using Independent Component Analysis of Hyperspectral Images in Patients With Age–Related Macular Degeneration . Invest. Ophthalmol. Vis. Sci. 2005;46(13):1562.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To demonstrate that changes in retinal spectra can be detected with high resolution spectral (hyperspectral) retinal images utilizing signal processing techniques such as independent component analysis. Because retinal tissue composition in general varies very little from individual to individual, spectral signals that are indicative of a disease or perhaps due to retinal or system toxicity are not early seen. Methods: A total of N=9 subjects (N=3 controls, N=3 with Best’s Disease, and N=3 with Stargardt’s disease) were evaluated. Prior consent from each patient was obtained as stipulated by the University of Iowa Human Subjects Committee. Independent component analysis (ICA) is a mathematical technique for finding underlying components from multidimensional (spectral bands and features) data. ICA has been used in other medical applications to extract signals in the retina due to visual stimulation and to detect brain activation in fMRI and EEG experiments to name a few. ICA is different from other methods in that it finds components that are both statistically independent, and nongaussian. The minimum detectable signal was established with ICA methods and results were compared to more classical techniques, such as multivariate analysis of variance (MANOVA) and principal components analysis (PCA). The ICA was then applied to these 9 cases to classify inter–subject differences, as well as intra–subject variations in tissue types (retinal background, retinal vasculature, etc.). Results: The ICA analysis demonstrated that a variation of less than 5 % can be detected through this robust analytical process. The ICA analysis shows potential for differentiating patients with different types of retinal disease. Using ICA, all three classes of subjects were uniquely classified. Conclusions: The ICA analysis of hyperspectral data shows potential for differentiating patients with different types of retinal disease. It is a powerful tool for extracting a low amplitude signal from complex signals.
This PDF is available to Subscribers Only