The application of multivariate statistical methods to NMR imaging

Statistical methods are used to integrate the information content in multivariate NMR images which had been obtained using different instrumental settings. Empirical and mechanistic models of multivariate pixel values are compared, making use of a novel reparametrisation of the mechanistic parameters to improve numerical stability. A new approach is presented to integrating the information using colour image displays. Projection pursuit is used to identify optimal instrumental settings in synthetic images. Methods are illustrated using seven cross-sectional NMR microscopy images of a blackberry, which was chosen as typical of fruit specimens with hard tissue distributed in a soft matrix.