Evaluation of illumination wavelengths for fluorescence detection of atherosclerosis.

Abstract

The fluorescence-emission spectra of normal arterial tissues and atherosclerotic plaques are distinct. The fluorescence spectra of arterial tissues are also dependent on the illumination wavelength. Which illumination wavelengths are best for identifying arterial tissues? To answer this question, a system for measuring fluorescence spectra at different illumination wavelengths was constructed and the fluorescence spectra of in-vitro aorta specimens were acquired with several illumination wavelengths between 270 nm and 500 nm. The Mahalanobis distance, a statistical measure of separability between two class density-distributions, was used to compare spectral features of normal vessel and atherosclerotic plaques at each of the illumination wavelengths. The Mahalanobis distance results indicated that the separability was largest for illumination wavelengths in the 314 nm to 334 nm range. In addition, the atherosclerotic plaque class was further subdivided into three diseased subclasses--fibrous plaques, complicated plaques and hard calcified plaques. The largest Mahalanobis distance between normal vessel and soft plaque (fibrous plaque and complicated plaque) again occurred with illumination in the wavelength range 314 nm to 334 nm. Conversely, the hard-calcified plaque class was most separable from the normal vessel class with illumination wavelengths longer than 380 nm. Since the best illumination wavelength ranges for soft plaques and hard calcified plaques were different, tissue identification was evaluated for the three-class case--normal artery versus soft plaques versus calcified plaques. Analysis with the generalized multiple-discriminant measure, a multiple-class extension of the Mahalanobis distance, indicated that illumination in the 300 nm to 314 nm range and the 400 nm to 458 nm range had the greatest three-class separability. Several multiple-class classifiers were also evaluated at each illumination wavelength. These classifiers did not reveal any wavelengths that were clearly best for accurate tissue-type identification. Finally, class separability and classification performance of combined features from multiple illumination wavelengths were also examined. The combination of features from multiple illumination wavelengths was found to increase separability and to improve the overall classification capability.