Magnetic resonance spectroscopy and pattern recognition in central nervous system disorders

Abstract

713

Objectives: The study combines 1H MRS and the partial least squares discriminant analysis (PLS-DA). Its aim is to explore the usefulness of multivariate methods of the data analysis as the diagnostic support tool in child neurology.

Methods: 77 spectra from the patients with inflammatory/demyelinating diseases (ID group; 17 spectra), neurometabolic disease (NM; 50 spectra) and cerebral palsy (CP; 10 spectra), were done using the 2T MRI/MRS system operating at a proton resonance frequency of 81.3 MHz (PRESS sequence). The multivariate analyses were performed in SIMCA-P 11 Demo (Umetrics, Sweden).

Results: The X data matrix was subjected to PLS-DA with the OSC preprocessing. The analysis resulted in a model with two significant components (R2X=0.229, R2Y=0.593 and Q2=0.489). In Figure 1a three distinct clusters are observed. As seen in the X-weights plot (Figure 1b) the first component is dominated by the spectral range 0.82-1.62 ppm, which corresponds to the increased mobile lipids, lactate and alanine levels in the NM group. The second component is dominated by the range of 1.95-2.03 ppm; it separates the ID group from the NM and CP ones and corresponds mainly to the acetyl moieties from N-acetyl-aspartate; the negative correlation indicates lower NAA levels in NM and CP than in the ID group. The spectral range 3.63-3.9 ppm is positively correlated with CP and corresponds to the myoinositol and glutamine/glutamate signals.

Conclusions: 1H MRS coupled with PR methods is much faster than biochemical screening of body fluids - such a combination may be useful as a supporting diagnostic tool.

Research Support: Supported by grant No. KBN 2P05E06829.