Reproducibility of 3D proton spectroscopy in the human brain.

The inter- and intrasubject reproducibility of the metabolite levels of N-acetylaspartate (NAA), creatine (Cr), and choline (Cho), obtained with three-dimensional (3D) multivoxel proton spectroscopy (1H-MRS), was analyzed in eight healthy volunteers. Serial, back-to-back measurements on a phantom showed the methodology and instrumentation to be highly reproducible, with a median coefficient of variation (CV) of 3.8%. In the human brain, the metabolite levels' variability was larger, with intrasubject median CVs for a total of 1876 signal voxels of 13.8%, 18.5%, and 20.1% for NAA, Cr, and Cho, respectively. These variations possibly arise from small, unavoidable, +/-1-2 mm volume-of-interest (VOI) repositioning uncertainties, which vary each 0.75-cm(3) voxel's partial fluid/gray/white-matter fractions. Comparing the CVs between eight subjects in a total of 324 selected voxels gave total interindividual CVs of 15.6%, 23.3%, and 24.4%, compared with intraindividual CVs in the same voxels of 14.4%, 14.8%, and 15.3%, for NAA, Cr, and Cho, respectively. Replacing the signal(s) from each voxel by the average of itself with its six canonical neighbors reduces the intrasubject median CVs to 8.3%, 9.5%, and 9.7%. The measurement uncertainties can be reduced at a cost of either spatial resolution (by using larger voxels) or time (by performing serial follow-ups).