fMRI studies require high BOLD sensitivity (BS), however susceptibility-related magnetic field inhomogeneities lead to signal dropouts and thus to BS loss in basal brain areas. There are a number of techniques for maximising BS in selected brain areas: e.g., z-shimming (1), inverting the phase encoding (PE) gradient polarity (2), optimizing the slice tilt (2, 3), increasing the spatial resolution (4) or optimizing the echo time TE (5). Previous optimization methods have been based on atlases derived from multiple EPI acquisitions (3) thus requiring resource and time and limiting the parameter range over which optimization can be performed. Here, we present an automated optimization method that can be employed for a large parameter space. It is based on numerical simulations informed by a large database of magnetic field (B₀) maps. In contrast to previous experimental approaches, it saves time and expensive measurements and allows for optimizing arbitrary EPI protocols including variable resolution, TE and slice orientation. The parameter optimization was compared to earlier experimental approaches and verified by BS measurements in healthy volunteers.

All scans were acquired on a 3T whole body MR scanner (Magnetom TIM Trio, Siemens). B₀ field maps (double echo GE) and anatomical data (3D FLASH) from 138 healthy volunteers were acquired as part of a whole brain quantitative MPM protocol (6). Field maps were calculated from the GE data and normalised to MNI space using the individual anatomical data. BS maps were calculated for each subject, by accounting for through-plane dephasing (1), local echo time/k-space shifts and signal loss due to susceptibility-induced in-plane gradients in the PE (1, 2) and readout (7) direction, and then averaged across volunteers. For the voxel-wise optimization results, shown in figures 2-4, the following EPI parameters were fixed: TE=30ms, resolution=3x3x3mm³, echo spacing 0.5ms and a 64x64 matrix. The following parameters were optimized for oblique transverse, sagittal and coronal slices with PE gradient directions pointing from anterior to posterior for transverse and sagittal slices and foot to head for coronal slices (see figure 1): slice tilt from -45° to 45° in steps of 5°, z-shim gradient pointing in slice direction with a moment from -5 to 5mT/m*ms in steps of 0.5mT/m*ms and a positive or negative polarity of the PE gradient. For comparison with earlier experimental approaches parameter optimization was performed using the same parameter space, orientation and resolution as used in (3). Additionally, for further validation EPI data was acquired for 36 different EPI protocols and four volunteers and the BS calculated from the complex raw data was then compared to the predictions of the numerical simulations.

In the case of a transverse acquisition, the optimal choice of z-shim gradient moment (fig. 2) was found to be negative in the orbitofrontal cortex but positive in the temporal lobes. For the sagittal acquisition, a left-right asymmetric distribution of z-shim values was found in the orbitofrontal cortex and near zero values in the temporal lobe, while for the coronal acquisition positive and negative z-shim gradients were found close to each other in the orbitofrontal cortex and the temporal lobe. For transverse/sagittal acquisitions, optimal slice angulations/rotations (fig. 3) were found to be positive in the orbitofrontal cortex and negative in the temporal lobes. Areas with positive and negative slice angulations could be found close to each other in the orbitofrontal cortex and temporal lobes in case of coronal acquisition, as was the case for the optimized z-shim gradient moments. The respective maps of the achieved BS gain is shown in fig. 4. The table in fig. 5 compares the results of the simulation based BS optimization with literature results (3) in different ROIs. The resulting optimized parameters were in good agreement for most of the regions. The comparison between simulated and experimental BS gain showed a good agreement with Gaussian distributed deviations around 5% pooled over the brain mask and subjects for each protocol.

The advantage of the proposed method is that it allows for automated optimization of EPI protocols by simulation, thus avoiding time and resource consuming measurements, allowing a larger parameter space to be optimised as well as easy adaptation if the basic protocol is changed. This framework also promises to provide improved optimization for group studies, since the typical distribution of field inhomogeneities in the population is better captured. The results of the optimization by simulations are in good agreement with earlier experimental optimization outcomes (3) and the expected BS increases are in line with the experimental BS results.