2.3. Postprocessing

Noise correlation matrixes were calculated from the second dynamic noise scan data, which contained more than 2 million samples, with a bandwidth of 128 kHz. The average and standard deviation of the noise correlation matrix were compared between the two setups, as well as under the tuned and detuned conditions of the transmit coil.

First, SNR maps were obtained by using the SNR‐valued reconstruction method of Kellman and McVeigh. 15 Using a dynamic noise scan for the noise covariance matrix ensures the same scaling for the signal and noise allowing for direct SNR calculation. After calculating the noise covariance matrixes, a noise prewhitening step was applied, which ensures a uniform noise for every channel. Thereafter, sensitivity‐weighted reconstruction using equal noise weighting was performed according to Roemer et al. 16 A data mask that captures the entire brain was applied to calculate the mean SNR within the subject.

Second, 1/g‐factor maps were calculated by omitting lines from the fully sampled k‐space. The 1/g‐factor maps were obtained using the coil sensitivity maps and the noise covariance matrixes according to Pruessmann et al. 17 The minimum and average 1/g‐factors were compared between the setups for 2D and 3D parallel imaging accelerations up to 25‐fold (5 × 5) for anterior posterior (AP)‐left right (LR), AP‐feet head (FH) and LR‐FH directions.