Micro Raman imaging in two and three dimensions.

Abstract

Microscopy using Raman scatter as a contrast principle provides chemical-state selective imaging. Because Raman scattering is an intrinsically weak effect, special techniques are required to provide high contrast, artifact-free images. We have developed artifact-removal computer programs for use with Hadamard transform Raman imaging, a multiplexing technique for spectroscopically resolved imaging using a conventional Raman spectrometer. We have also developed image reconstruction programs that provide sharp, confocal-like deblurred images. These programs were tested on images obtained both by the Hadamard technique and by direct imaging using narrow pass band tunable filters. A first-order correction was developed to remove echoes (ghost images) from Raman images obtained by the Hadamard multiplexing technique. The correction was designed to correct for errors in the fabrication of encoding masks. The correction algorithm assumed identical errors at each open/closed (1/0) boundary in the mask. It was demonstrated that the averaging properties of the Hadamard transformation allowed use of this approximation. The nearest-neighbor deblurring technique was implemented for restoration of the central member of stacks of three images obtained by serial sectioning. A fixed (0.4-0.45) fraction of the intensity of the first and third images was subtracted from the central image. Because the technique produces an image with lower signal/noise ratio than the originals, it could only be used on the strongest images available. In such cases, the processed images were sharper than the starting images. Constrained iterative deconvolution was used to process noisier images. Idealized (sinc$\sp2)$ point spread functions were employed. Prior to deconvolution stacks of images were smoothed using the LOWESS algorithm. Unconstrained 2-dimensional (lateral) Fourier deconvolution was applied. Next, axial Fourier deconvolution was applied using a non-negativity constraint. Two or three iterations were necessary. The resulting images contained much less out-of-focus intensity than the starting images. They were of sufficiently high quality that they could be processed through conventional 3-dimensional rendering programs.