Meeting Banner
Abstract #1967

Nonlinear Laplacian Eigenmaps Dimension Reduction of in-vivo Magnetic Resonance Spectroscopic Imaging Analysis

Guang Yang1, Felix Raschke2, Thomas Richard Barrick2, Franklyn A. Howe2

1Division of Clinical Sciences, , St. Georges University of London, London, United Kingdom; 2Division of Clinical Sciences,, St. Georges University of London, London, United Kingdom

MRSI has demonstrated great clinical potential as a supplement to standard imaging for non-invasive diagnosis of brain tumours. Pattern recognition(PR) techniques are used to assist MRSI tumour identification and characterisation, and they can be applied to MRSI data with suspected gliomas with an aim to segment regions relating to tumour core, tumour infiltration and normal brain. Dimensionality reduction(DR) is an important prerequisite in any real case of PR. In this work, we advocate the spectral manifold learning method of Laplacian eigenmaps as a DR technique suitable for MRSI datasets, with correlation to standard MRI to aid confirmation of our results.

Abstract PDF Poster PDF

Keywords

according achieve acquisition adjacency advocate allowing applied assigned assist assumes assumption biopsy blue brain characteristics class classes classify clinical component components compute confirmation connected considered constructed consuming contrast core correlation curves cyan define defined definitive determine developed diagnosis difficult dimension dimensional dimensionality distance edges eigenvalues either embedded equivalent ethics euclidean fail findings furthermore generated geodesics gives grade graph green house identification in vivo indicated individual infiltration infiltrative informative informed invasive kernel kingdom known linear local locally magnet manifold manifolds mapped mapping maps mass materials matrix median metabolite mixture must nearest necrotic networks neural next nodes nonlinear offers often outer overlaid overlays patients plots potential prerequisite previous prior procedures processed promising proposed quartiles reasonable recognition reduced reduction reflects remain represent representation representations resolution respectively risks scatter sciences segmentation segmentations sets source space spatial spectra spectral spectroscopic spectroscopy spectrum step structure suggestive suggests suitable suppression suspected theory thick tissue together transform volume wright yang zero