Maximising BOLD sensitivity through automated EPI protocol optimisation

Abstract

Gradient echo echo-planar imaging (GE EPI) is used for most fMRI studies but can suffer substantially from image distortions and BOLD sensitivity (BS) loss due to susceptibility-induced magnetic field inhomogeneities. While there are various post-processing methods for correcting image distortions, signal dropouts cannot be recovered and therefore need to be addressed at the data acquisition stage. Common approaches for reducing susceptibility-related BS loss in selected brain areas are: z-shimming, inverting the phase encoding (PE) gradient polarity, optimizing the slice tilt and increasing spatial resolution. The optimization of these parameters can be based on atlases derived from multiple echo-planar imaging (EPI) acquisitions. However, this requires resource and time, which imposes a practical limitation on the range over which parameters can be optimised meaning that the chosen settings may still be sub-optimal. To address this issue, we have developed an automated method that can be used to optimize across a large parameter space. It is based on numerical signal simulations of the BS loss predicted by physical models informed by a large database of magnetic field (B₀) maps acquired on a broad cohort of participants. The advantage of our simulation-based approach compared to previous methods is that it saves time and expensive measurements and allows for optimizing EPI protocols by incorporating a broad range of factors, including different resolutions, echo times or slice orientations. To verify the numerical optimisation, results are compared to those from an earlier study and to experimental BS measurements carried out in six healthy volunteers.

Fig. 1

Definition of coordinate axis for phase encoding (PE), readout (RO) and slice direction (SL) for the main slice orientations transverse (left), sagittal (middle) and coronal (right) as used in the experiments. For illustration of slice orientations, slices are overimposed over a sagittal view of the human head. The directions for slice angulations and in-plane rotation are denoted with the red arrows respectively.

Fig. 2

Maps of susceptibility-induced B₀ field gradients obtained from the field maps for all three directions.

Fig. 3

Maps of the voxel-wise optimized parameters for 16 slices in case of oblique transverse acquisitions with an in-plane resolution of 3 × 3mm², a matrix size of 64 × 64 and a slice thickness of 3 mm. Optimal shim gradient moment (a), optimal slice angulation (b) and BS gain achieved with the optimal parameter set compared to standard EPI with no shim gradient and slice angulation (c). In each case the optimized parameters are shown for a positive PE gradient (top row) and for a negative PE gradient (bottom row). A mask has been applied to show only optimized parameters with a BS gain of at least 20%.

Fig. 4

Maps of the voxel-wise optimized parameters for 16 slices in case of sagittal acquisitions. Optimal shim gradient moment (a), optimal in-plane rotation (b) and BS gain achieved with the optimal parameter set compared to standard EPI with no shim gradient and slice rotation (c). In each case the optimized parameters are shown for a positive PE gradient (top row) and for a negative PE gradient (bottom row).

Fig. 5

Maps of the voxel-wise optimized parameters for 16 slices in case of coronal acquisitions. Optimal shim gradient moment (a), optimal slice angulation (b) and BS gain achieved with the optimal parameter set compared to standard EPI with no shim gradient and rotation (c). In each case the optimized parameters are shown for a positive PE gradient (top row) and for a negative PE gradient (bottom row).

Fig. 6

Optimization of the transverse standard resolution protocol for different ROIs: OFC = mOFC + rACC, ITL = Inferior temporal lobes, TL = Temporal lobes, Amy = Amygdala, and Hippo = Hippocampus + Parahippocampus. For the optimization based on group-average field maps the BS gain depending on slice tilt and z-shim assuming either a negative PE direction (a) or a positive phase encoding direction (b) are shown. To convey how well the optimal parameters derived from group-average fieldmaps translate to single subjects, the histograms in (c) and (d) show for how many subjects a particular value of slice tilt and z-shim would result in the maximal BS gain based on individual field maps. The number of subjects for the optimal slice tilt are displayed for each PE direction separately. (e) shows how the optimization for one ROI affects the BS in the other ROIs. As additional ROIs the Whole Brain (WB) and Well Shimmed areas (WS) are shown.

Fig. 7

Optimization of the sagittal standard resolution protocol for different ROIs: OFC = mOFC + rACC, ITL = Inferior temporal lobes, TL = Temporal lobes, Amy = Amygdala, and Hippo = Hippocampus + Parahippocampus. For the optimization based on group-average field maps the BS gain depending on slice tilt and z-shim assuming either a negative PE direction (a) or a positive phase encoding direction (b) are shown. To convey how well the optimal parameters derived from group-average fieldmaps translate to single subjects, the histograms in (c) and (d) show for how many subjects a particular value of slice tilt and z-shim would result in the maximal BS gain based on individual field maps. The number of subjects for the optimal slice tilt are displayed for each PE direction separately. (e) shows how the optimization for one ROI affects the BS in the other ROIs. As additional ROIs the Whole Brain (WB) and Well Shimmed areas (WS) are shown.

Fig. 8

Optimization of the coronal standard resolution protocol for different ROIs: OFC = mOFC + rACC, ITL = Inferior temporal lobes, TL = Temporal lobes, Amy = Amygdala, and Hippo = Hippocampus + Parahippocampus. For the optimization based on group-average field maps the BS gain depending on slice tilt and z-shim assuming either a negative PE direction (a) or a positive phase encoding direction (b) are shown. To convey how well the optimal parameters derived from group-average fieldmaps to single subjects, the histograms in (c) and (d) show for how many subjects a particular value of slice tilt and z-shim would result in the maximal BS gain based on individual field maps. The number of subjects for the optimal slice tilt are displayed for each PE direction separately. (e) shows how the optimization for one ROI affects the BS in the other ROIs. As additional ROIs the Whole Brain (WB) and Well Shimmed areas (WS) are shown.

Fig. 9

Histograms of the percent deviations between simulated and experimentally measured BS gains pooled across the brain mask (for each of the 36 protocols and six subjects). The optimization results for the transverse protocol are shown for the negative and positive PE direction in a) and b), respectively. The optimization results for the sagittal protocol with negative and positive PE direction are shown in c) and d). The protocol with negative PE and no tilt and z-shim is not shown, since it was the reference protocol. The blue histogram represents the experimental data, while the red curve is a Gaussian fit with the respective estimated mean and standard deviation.