PREDICTABILITY OF SNR AND READER PREFERENCE IN CLINICAL MR IMAGING

Fifty-four independent scans were performed in two volunteers covering one anatomic region in each (the brain and knee) with the purpose of ascertaining the agreement between predicted and measured signal-to-noise ratios (SNR). Systematically varied parameters were number of excitations (NEX), field of view (FOV), section thickness (dz), and the number of phase-encoding steps (Ny). Correlation coefficients of measured versus predicted SNR were 0.82 and 0.86, respectively, in the anatomies studied. Significantly improved correlations were found for data subpopulations in which NEX was held constant. To assess the criteria guiding reader preference, a blinded study was performed in which radiologists were asked to rate images from least to most desirable. In order to quantitatively determine the criteria for reader preference, plots of mean rating versus SNR, voxel volume, and an image quality index [IQI = SNR/(voxel volume)] were performed. The latter was found to be a better predictor of reader preference than either SNR or spatial resolution alone. The data suggests T1-weighted scan protocols yielding SNR of approximately 20 are preferable with any excess SNR being traded for smaller voxel size or shorter scan times. © 1990.