To investigate the dependence of DTI metrics on signal to noise ratio (SNR) in a longitudinal multicenter clinical trial. In a longitudinal trial, SNR may fluctuate with time and cause variations on DTI measures. Changes can be particularly severe after a hardware or software upgrade. Differences among MR platforms can introduce across-scanner bias. Although many simulations have shown that fractional anisotropy (FA) and diffusivities depend on SNR^1-3, experimental tests of this dependence may reveal important systematic differences among sites.Images are acquired from BIRN phantoms⁴ on a monthly basis from each of 27 sites in the SPRINT-MS trial (11 Siemens TIM Trio, 6 Siemens Skyra, 1 GE Signa EXCITE, 7 GE Signa HDxt, 1 GE DISCOVERY MR750 and 1 GE DISCOVERY MR750w) over 15 months. The trial is performed within the NeuroNext (www.neuronext.org) network. Scans were acquired with 2.5mm isotropic resolution with 64 b=700sec/mm² and 8 b=0 volumes. SNR was determined on a voxel-by-voxel basis by taking the ratio of the mean and standard deviation among b=0 volumes for each scan within a 16mm x 16mm (Siemens) or 40mm x 40mm (GE) square region of interest (ROI) at magnetic isocenter. Diffusivities were calculated using log-linear least-squares⁵. For comparison, we used Monte Carlo simulations assuming axially symmetric tensors with mean diffusivities ranging from 1600 x10^-6mm²/s to 2100x10^-6mm²/s. We assume FA = 0, as expected for an isotropic agar phantom. Noise was added to the signal, with SNR varying from 8-100 and 10000 noise realizations for each value of SNR. MD and FA were calculated to compare with results obtained from phantom images. Data analysis and simulations were performed using Matlab (MATLAB R2012a, The MathWorks Inc., Natick, MA 2012).MD and FA as a function of SNR are plotted in Fig.1 and 2. The experimental data points encompass a broad range of MD and SNR among scanners and time points. If we assume that MD is constant across phantoms and over time, we observe an unexpected linear relationship between SNR and MD. Pearson correlation is significant for each scanner type (R=-0.66, -0.91, -0.77 for Trio, Skyra, GE, p<0.0001). FA and SNR correlate for GE (r=-0.25, p=0.002) but not for Siemens scanners. Although the range of SNR was large among scanners and time points, the MD and FA largely overlapped.Measurements of DTI parameters can be biased by noise^1-3,6. We presented an analysis of the effect of SNR on MD and FA on 27 scanners in a longitudinal multicenter study. The dependence of DTI measures on SNR is different for each type of scanner and deviates from the theoretical relationships. The difference among scanners may be due to a number of features which are currently under investigation. There are systematic differences among scanners in terms of eddy current artifact⁷ and reconstruction algorithm that may be responsible for the observed trends. It is possible that the experimentally observed measurements of SNR can be used to control for systematic differences among scanners, but it is important to note that the relationship between SNR and DTI measures may require inclusion of more factors than is typically assumed.Experimental evaluation and simulation showed that a sufficient SNR is essential for a reliable estimation of DTI parameters generally. The findings of this study may prove important to account for inter-scanner biases in a longitudinal study.