Premium
Segmentation of MRS signals using ASPECT (Analysis of SPectra using Eigenvector deComposition of Targets)
Author(s) -
Roebuck Joseph R.,
Windham Joe P.,
Hearshen David O.
Publication year - 1994
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.597289
Subject(s) - eigenvalues and eigenvectors , segmentation , decomposition , artificial intelligence , pattern recognition (psychology) , spectral line , image segmentation , computer science , mathematics , computer vision , physics , chemistry , quantum mechanics , astronomy , organic chemistry
Efforts to minimize the effects of partial volume contamination (PVC) in in vivo magnetic resonance spectroscopy (MRS) have focused upon improving the sensitivity and efficiency of spatially localized MRS measurements. Such improvements may improve spatial resolution and reduce the time required to acquire multiple spectra, however, PVC can affect in vivo spectra at any resolution. In this paper, a model for segmenting in vivo MRS signals compromised by PVC in selected applications is introduced. The segmentation algorithm used is linear and is based on filters originally developed for image processing applications. The model is developed from first principles and evaluated using computer simulations. It is suited for segmenting multivoxel or chemical shift imaging data, and can be used with spectra acquired at any spatial resolution. It is used to estimate the size of the partial volumes contributing to a voxel compromised by PVC and the spatially selective signal components that would be expected to arise from these partial volumes if they could be measured directly. Several spectral perturbants present in in vivo MRS measurements violate the linearity assumptions underlying the model and produce systematic errors that must be accounted for. A number of perturbants are discussed, and the potential in vivo applications of the model are illustrated using solvent‐suppressed 1 H–CSI spectra from the normal human brain.