Data analysis of multivariate magnetic resonance images I. A principal component analysis approach

Abstract Grahn, H., Szeverenyi, N.M., Roggenbuck, M.W., Delaglio, F. and Geladi, P., 1989. Data analysis of multivariate magnetic resonance images. I. A principal component analysis approach. Chemometrics and Intelligent Laboratory Systems , 5: 311–322. Principal components were used as a novel way to condense information in magnetic resonance images. The procedure is capable of extracting the most significant information from stacks of congruent images and of condensing them into a few principal component images. The results are highly visual in nature, and demand little mathematical expertise from the user. The related loading plots can be used as a criterion for the experimental parameter settings. The score plots provide better differentiation of materials than any scatter plot of images obtained directly from pulse sequence parameters with a given pulse repetition time (TR) and echo time (TE) combination. Two examples are shown to illustrate the clinical interpretation. They are a phantom containing solutions of water and glycerin with widely varying relaxation properties, and also of a developing chicken embryo.