Magnetic resonance spectroscopy (MRS) allows to determine the concentrations of important neurometabolites of the brain in a non-invasive way. To utilize the full diagnostic potential of this method in vivo, it is imperative to know the measurement uncertainty of the determined neurometabolite concentrations, as otherwise statistical fluctuations of the measurand may be easily confused with real biological differences. In MRS it is common to assume so-called Cramér-Rao Lower Bounds as measurement uncertainty. However, Cramér-Rao Lower Bounds only represent the lower boundary of the measurement uncertainty, assuming that the used quantification model is correct and the parameters are well chosen. A statistical standard deviation can usually not be determined, as limits on time and costs prevent repeated measurements on the same patient.
The publication "Assessment of measurement precision in single-voxel spectroscopy at 7 T: Toward minimal detectable changes of metabolite concentrations in the human brain in vivo", which is being published in the 2022 March issue of the journal Magnetic Resonance in Medicine, therefore presents a study design and statistical framework, which allows to determine real standard deviations of neurometabolite concentrations measured with MRS in vivo. The presented framework was applied to in vivo measurements performed at a 7T MR scanner, in which the concentrations of neurometabolites were obtained using different versions of the SPECIAL (SPin ECho full Intensity Acquired Localization) technique. To this end, nine healthy volunteers were each examined in total four times on two different days using the same protocol. Variation of time between these repeated measurements allowed to investigate repeatability and reproducibility individually. As a result, the real uncertainties for the reproducibility of a concentration determination correlated quite well with the respective Cramér-Rao lower bounds for all metabolites considered, but were consistently above these individual measurement uncertainties by a factor of about two to three.
The described study design can easily be applied to other scanners and methods, and hence, will allow the assessment of statistical variances within clinical studies and the reduction of misinterpretations of statistical fluctuations. Simultaneously with the publication, both the generated data and the algorithm used were published in open source databases.
The article was chosen as "Editor's Pick" of the march issue of Magnetic Resonance in Medicine.
https://doi.org/10.1002/mrm.29034
Contact:
Ariane Fillmer, 8.1, E-Mail: Ariane.Fillmer(at)ptb.de