Diffusion imaging of post-mortem (PM) human brain has important applications, including validating diffusion signal models and investigating anatomy with very high-resolution data¹. Diffusion-weighted steady state free precession (DW-SSFP) has been demonstrated to be ideally suited for PM imaging at both 3T² and 7T³. However, at 7T challenges arise due to B1 inhomogeneity, which causes the true flip angle to vary across the brain compared to the nominal flip angle. In this work we compare data acquired with 2N diffusion-weighted directions at a single nominal flip angle against N directions repeated at two nominal flip angles⁴. In both strategies, the nominal flip angles (or equivalently, transmit voltages) have been chosen to optimize CNR while achieving maximum whole brain coverage. Data were acquired of PM human brains (n=4) diagnosed with motor neuron disease. PM intervals were under 72 hours. Brains were submerged in Fomblin for susceptibility matching. DW-SSFP data were acquired at 0.85mm isotropic on a human 7T Siemens whole body scanner using a 32-channel receive head coil. DTI data were acquired over 120 diffusion directions with a q-value of 300 mm^-1 (achieving diffusion contrast equivalent to b_eff ~ 6000 s/mm²). Different flip angle strategies can be analyzed with respect to the highly reproducible distribution of B1 across the brain. If we assume homogeneous B1 transmission, one can calculate the optimum flip angle as that which maximizes CNR². At 7T, increasing or decreasing the transmit voltage determines the position of this flip angle within the brain (Figure 1). In the single flip angle strategy, the transmit voltage was set to position the optimal flip angle midway between the center of the brain and the cortex. While the center and periphery are both acquired far from the optimal flip angle, this has the effect of maximizing the average CNR across the range of B1 inhomogeneities. In the dual flip angle approach, the transmit voltage is set so optimal flip angles are at the center of the brain and the periphery for each acquisition, respectively³. Single flip data were acquired twice to match total scan time with the dual flip approach. All data were processed using a modified version of DTIFIT to account for the DW-SSFP signal model⁵. Data were white matter masked before analysis. A measured B1 map was binned to serve as regions of interest for analysis of signal behavior in terms of transmit inhomogeneity (Figure 2).

First we consider the behavior of the three different ranges of voltage settings separately in terms of dispersion and coherence of the estimated orientation. Figure 3 shows orientation dispersion and coherence (mean ± standard error across all brains) vs. B1 for the three anatomically driven transmit voltages (based on a single set of 120 directions for each voltage). Low voltage produces high dispersion and low coherence in the cortex and low dispersion and high coherence in the center; high voltage produces opposite results. The medium voltage data appear to adopt the best of the other two approaches: comparably high coherence and low dispersion in the center and the cortex compared with low and high voltage results, respectively. Moreover, medium voltage results in the region of the brain between the center and the cortex are improved compared with either of the other two. If a single voltage is to be used, it is clear that the medium voltage is preferable.

Figure 4 shows the same plots for data processed with either 2 averages of the medium voltage or combined low and high voltage data. Coherence in the single flip data is slightly lower across the brain in medium voltage data than the dual flip, while dispersions are similar. Orientation results of the single and dual flip strategies (Figure 5) show that in both cortex and the center of the brain that while metrics to determine which approach is optimal, the actual differences are quite subtle.

This work presents a comparison of two acquisition strategies to overcome B1 inhomogeneities at 7T. Analysis of the three different transmit voltages demonstrates that we can exploit B1 inhomogeneity to tailor the flip angles, thereby allowing for targeting regions of the brain between the center and the cortex. Results indicate that the latter, simpler approach is if anything slightly better in terms of orientation estimates. However, separate work has begun to appreciate the potential to fit more sophisticated signal models (e.g. biexponential decay or restriction effects) to DW-SSFP with multiple flip angles, potentially allowing us to achieve microstructural estimates at little to no cost to orientation estimates.