Datensatz

Zusammenfassung

Introduction Critical usability assessment criteria for Magnetic Resonance Spectroscopic Imaging (MRSI) include spectral resolution and signal-to-noise ratio (SNR). In practice, scan time limitations dictate a spatial resolution which is rather coarse compared to pure imaging. This leads to voxel bleeding artefacts reflected in imperfections of the Spatial Response Function (SRF). To reduce them, target-driven overdiscrete reconstruction has been investigated and found to lead to favourable trade-offs for MRSI [1, 2]. This approach furthermore allows to correct for inhomogeneities of the main magnetic field (B0) in a quasi-continuous manner, yielding SNR increase [3]. In this work, the effect on SNR and spectral resolution is investigated in detail and a technical explanation is given. Methods An intermediate spectroscopic image (iSI) with a higher-than-nominal spatial resolution but low SNR is created after application of the first part of the reconstruction operator [3]. The second part then amounts to application of the SRF target function to the iSI, i.e. a spatial averaging. Prior to that, the spectra in the iSI are frequency-aligned according to a separately acquired B0 map on a subvoxel level. The example data presented are a 1H MRSI acquired at 7T with a nominal 20x16 resolution using the FIDLOVS [4] sequence with and without water suppression. Results Fig. 1B contains a subset of in vivo example spectra after eddy current [5] and phase correction from the indicated voxel positions (Fig. 1A, underlying B0 distribution included). Mean SNR improvement by a factor of 4.2 by B0 correction on the intermediate spectroscopic image is obvious. The effect is due to the destruction of previously existing noise correlation between neighboring subvoxels (Fig. 2A). The distribution of SNR of the NAA peak at 2.0 ppm and the fit results of a Voigt line fit to the water peak in the non-water suppressed scan are shown in Fig. 2B, indicating narrower line shapes and hence improved spectral resolution. Conclusions When using target-driven overdiscrete reconstruction, major SNR and line width improvement may be achieved simultaneously by performing B0 correction at the level of the iSI. The additional effort of acquiring the necessary B0 map is usually small compared to the MRSI measurement.