Reproducibility of magnetic resonance spectroscopy in correlation with signal-to-noise ratio

An increased amount of myoinositol (mI) relative to creatine (Cr) by proton MR spectroscopy (1H-MRS) measurement gives a useful aid for the diagnosis of Alzheimer's disease (AD). Previous results of test–retest measurement of mI, however, have shown variability more than twice as large as for other metabolites. The aims of this study were to analyze test–retest variability of 1H-MRS measurements in correlation with signal-to-noise ratio (SNR). Ten subjects clinically suspected of mild AD were examined twice (2–14 days apart) with 1H-MRS measurements of voxels placed at anterior and posterior cingulate cortex. The percent differences between two measurements (%differences) of mI/Cr showed a significant linear trend to decrease as average SNR increased, but %differences of N-acetylaspartate (NAA)/Cr and choline (Cho)/Cr did not. The average of %differences was 10.5, 15.0 and 20.8 for NAA/Cr, Cho/Cr, and mI/Cr, respectively, indicating a prominent deterioration of mI/Cr measurement reproducibility, which decreased to 6.96, 15.4 and 9.87, respectively, when the analysis was limited to measurements with SNR over 25. The results indicate that MRS measurements with high SNR should be used to obtain reliable assessments of mI/Cr as accurate diagnostic indicator of AD in clinical MR examinations.