The emergence of compact optical spectral imaging technologies has motivated the study of their use in a variety of applications, including medical diagnosis and monitoring. In particular, large format CCD focal planes in conjunction with spectrally tunable devices offer enhanced spatial information together with visible and near infrared (NIR) spectroscopic data for the passive, noninvasive, measurement of human skin and near surface tissue characteristics. One such spectral imaging system was recently developed by mating a Liquid Crystal Tunable Filter (LCTF) together with a 2048x2048 silicon CCD focal plane. This system is capable of collecting more than 30 co-registered spectral images spaced every 10 nanometers and spanning 400 to 720 nanometers. This system combines the potential of near infrared diffuse reflectance spectroscopy with the high spatial resolution of traditional optical imaging techniques. Spectral images were acquired of portions of the hands and arms of several test subjects with a variety of features observable. The observations were collected in a "light box" under controlled illumination conditions. Images of a diffuse reflectance standard and instrument dark frames were collected to allow conversion of the raw images to spectral reflectance images. This paper presents examples of the spectral images collected, instrument characteristics and performance, and results of analysis algorithms applied to the data. Results also are shown for a new algorithm extracting the saturated oxygen hemoglobin fraction from these data.
A number of non invasive methods have been developed to characterize parameters in near-surface skin tissue; however, the work has usually been concerned with using either spectral or spatial information. This motivated our study in which both spatial and spectral data are used to extract features for characterizing the spatial distribution of near-surface oxygen saturation. This paper addresses combined physical and statistical models to retrieve the ratio of oxy- and deoxy-hemoglobin in tissues from data collected by an imaging spectrometer. To retrieve the oxygen saturation fraction from the data, algorithms from the literature using two or three wavelengths were compared to our new algorithm using the many more wavelengths (25 to 60) available in imaging spectrometer data, and noise reduction achieved through principal component transformations.
In addition to the analysis of experimental spectral imagery, an oxygen saturation phantom of size 128x128 pixels was simulated. In the forward process, a reflectance image was constructed from an assumed oxygen saturation map and the absorption coefficients of oxy- hemoglobin, deoxy-hemoglobin, melanin and other chromophores. The reflectance data have 60 bands spanning 400 nm to 990 nm with 10 nm intervals in the spectral dimension. Varying amounts of white Gaussian noise was added to the reflectance data to simulate measurement errors in an actual experiment. In the backward process, an oxygen saturation image was reconstructed by applying the algorithm to study the effect of measurement error on the retrieved saturation fraction. The resultant images were evaluated by their mean squared error.