| Computational optical sectioning can reveal the three-dimensional microstructure of biomedical specimens without damaging the object. Conventional microscope images contain information at the focal plane, obscured by out-of-focus information from adjacent planes. Several computational optical sectioning methods have been developed to remove out-of-focus information, and have been applied in a number of published biomedical studies. However, because these mathematical models only approximate the imaging system, computational methods must have inherent limitations. As a consequence, such methods might provide false information about the object. Such limitations have long been ignored or underestimated. In order to understand better the limitations of computational methods, and to avoid misinterpreting their results, a quantitative evaluation of the strengths and weaknesses of such methods must be obtained. The present thesis is the first evaluation of computational methods which employs reliable physical models. An agar-yeast model of one-dimensional diffusion was employed to provide a predictable intensity gradient which was used to evaluate the restoration ability of computational sectioning methods in the low frequency range. For high frequency signals, fluorescent beads were used as the test objects to evaluate the faithfulness of the sectioning results. A specially designed orthogonal illumination system allowed the bead optical section to be directly observed using the same optical imaging system employed in testing the computational sectioning methods. Therefore, any discrepancies between the orthogonal plane image and the results of computational methods could be attributed solely to deficiencies in the computational methods. The experimental results demonstrate that none of the tested computational methods is capable of restoring low frequency 3-D information. Therefore, when computational methods are used to perform quantitative measurement of low frequency signals caution should be employed in interpreting the sectioning results. For high frequency signals, it was found that the computational sectioning results were statistically acceptable with regard to absolute object size; however, the artifacts and noise in such images were exaggerated by all tested methods. Moreover, the resultant images exhibited shape distortion. The effects of imperfections in the experimental point spread function on the resultant image were also evaluated. |
