Statistical Methods to Determine the Reliability of in Vivo Single Voxel MRS Data

Johannes Slotboom¹, Dirk van Ormondt²

Spectral quality and spectral reliability are closely related, but not identical. Although MRS-signal reliability is an issue of highest importance, its notion received, to the best of the authors knowledge, no systematic treatment in the in vivo MRS-literature in the past. Frequently it remains unclear whether clinical in vivo single voxel data MRS data can be trusted, i.e., can be used for diagnostic purposes. In order to make SV-MRS clinically more viable, assessment of spectral reliability of the data should preferably be handled by the MR-scanner system rather than by medical staff. In this contribution, we devise, apply, and test statistical methods for automated data reliability testing. Once the data pass the reliability test, the MRS signal is considered reliable and the variances of the maximum likelihood (ML) estimated parameters approach the Cramr-Rao minimum variance bound (CR-MVB).