We have been developing image processing methods to determine the structure and composition of macromolecular assemblies and cellular organelles in the electron microscope. The purpose of this project is to apply multivariate statistics to classify noisy images used both in three-dimensional reconstructions and elemental mapping. The noise is due to low-dose requirements for recording micrographs without significant radiation damage or due to the limited signal that is available in the elemental maps. To determine macromolecular structure, principal component analysis is applied to calculate orthogonal eigenimages that are classified according to their information content. The orientation angles between the images are then calculated, which enables a three-dimensional reconstruction to be made. Elemental distributions are obtained by means of electron energy loss spectroscopic imaging (EELS) where spectra are collected at each pixel in an image to give a three-dimensional data cube in x, y, and E (energy loss). Multivariate statistics are being applied to determine correlations between different features in energy loss spectra and between different features in the resulting elemental maps. - image processing, electron micrographs, multivariate statistics, macromolecular structure, electron energy loss spectroscopy
