Optical tomography by time-resolved diffuse reflectance measurements

The understanding of the interaction of light with a scattering/absorbing medium provides a foundation needed for developing many applications in diagnostic medicine and industry. The objective of this research was to obtain quantitative depth-resolved information about absorbing constituents in a scattering medium.Initially, the project focussed on quantification in samples where scattering and absorber concentration were variable. Using time resolved reflectance measurements, a series of statistical descriptors of the photon time distributions were calculated. Stepwise multilinear regression was used to formulate linear models from optimal linear combinations of the descriptors. It was found that the scattering coefficient, absorption coefficient and apparent particle diameter could be estimated to within 9, 10 and 7% of their reference values respectively.An array of radial reflectance measurements on layered scattering/absorbing samples was made to obtain information sensitive to sub-surface changes in absorption. As an initial approach to depth-resolved quantification, classical tomographic reconstruction techniques were used. However, due to the ambiguity of the reconstruction, extremely poor estimates of the sub-surface absorption resulted. Chemometric methods were then employed for enhanced quantification. By using stepwise multilinear regression with time-resolved data, the absorption coefficient in the top region of a sample could be estimated to within 2%. However, errors in the absorption coefficient estimations deep within a sample remained high.Further improvements in sample quantification were made by linearizing the reconstruction problem. By using a priori information about sample composition in upper regions, subsequent calibrations for lower regions were directed. Estimations of the absorption coefficient deep within a sample with hierarchical locally weighted calibration were obtainable at greater than 50% accuracy. This represented a 20% improvement at all sample depths over stepwise multilinear regression.Confocal illumination and detection optics was also used for discriminating highly scattered photons from light, which follows a geometric path through a sample. When confocal optics were used together with information from the rising edge of time distribution, little enhancement in quantification was observed in comparison to an integrated signal. This important finding demonstrates that the confocal optical detection should be considered when imaging in scattering/absorbing media