To find the optimal echo time (TE_opt) for highly accelerated fMRI acquisitions.

A simple calculation shows that the largest BOLD signal is obtained at TE≈T₂^* of gray matter(1). Many fMRI experiments, however, employ a shorter TE to improve temporal sampling density, the BOLD contrast-to-noise ratio (CNR_BOLD), or to reduce through-slice dephasing. Here, we simulated CNR_BOLD as a function of TR, TE, field strength and number of slice excitations N_slices, taking into account the BOLD signal amplitude (which peaks at TE≈T₂^*_GM ), the increase in CNR_BOLD with increased temporal sampling density as well as the signal loss due to T₁-weighting, always using the Ernst angle for excitation. TR was assumed to be >>T₂. BOLD changes in a venous compartment were also simulated to estimate BOLD specificity.

BOLD detection power was modelled taking into account TR, sampling time and expected T₂^*_BOLD contrast: $$CNR_{BOLD} = \Delta R_2^* \times TE\times exp^{\frac{-TE}{T_2^*}} \frac{S_{\alpha_{Ernst}} \sqrt{ETL}}{\sqrt{TR}}$$ whereby $$$S_{\alpha_{Ernst}} = sin(\alpha_{Ernst})\frac{1-exp^{\frac{-TR}{T_{1}}}}{1-cos(\alpha_{Ernst})exp^{\frac{-TR}{T_{1}}}}$$$. The minimum TR was assumed to be $$$TR = (TE+\frac{ETL}{2})N_{slices}$$$, with the centre of k-space sampled in the middle of the echo train, where ETL is the echo train length and N_slices the number of slice excitations. A 0.65mm³ 7T field map (in rad/s, scaled appropriately for lower field strengths) was used to calculate signal loss from through-slice dephasing assuming an axial slice orientation and $$$CNR_{BOLD} =CNR_{BOLD}sinc(\gamma \frac{\delta \Delta B_{0}}{\delta z}\Delta z TE)$$$. Simulations were performed for a range of ETLs, B₀, number slice excitations and thicknesses. For brevity only results for 100/60/30/15 slices, ETL=30ms, B₀ =1.5/3/7T and a slice thickness of 2 mm are presented. Table 1 lists the relevant relaxation parameters used in the simulations.

Table 2 shows TE_opt values for protocols with different N_Slices. It can be seen that, even in the absence of through-slice dephasing, TE_opt was generally found to be < T₂^* and closer to the typically used TE values at lower field strength (Table 2). However, TE_opt tends to increase with decreasing N_Slices. The variation is larger at lower field strengths where the shorter T1 values make large N_Slices (and increased TRs) suboptimal. Figure 1 compares BOLD sensitivity at the different fields for N_Slices=100 and N_Slices=15 (corresponding to whole brain coverage with a through-slice acceleration factor of ~6 for 2mm slices). There, the increase of TE_opt can be seen, but also the fact that lower fields benefit more from multiband technology (Figure 1) with a CNR_BOLD improvement of 90, 40 and 18% at 1.5, 3 and 7T respectively (ignoring g-noise penalties).

Relative venous signal contributions (Figure 1) increase with B0 and reduce with TE in all cases.

When taking into account through-slice dephasing due to B₀ field inhomogeneities, shorter optimal TE values are found above the frontal sinuses at 3T (Figure 2). At the optimal echo time, this indeed results in lower CNR_BOLD in the affected regions (Figure 3, top row). However, if a shorter TE (for illustration: 35ms@1.5T, 30ms@3T, 25ms@7T, Figure 3 bottom row) is used to prevent BOLD signal losses over the sinuses, the BOLD contrast in the rest of the brain is significantly lowered, while only small increases in CNR_BOLD are found in the affected regions.

For example, TE=35ms at 1.5T leads to homogeneous BOLD CNR levels throughout the example slice, but this tSNR level is 10-20% lower than the slower, TE=74ms acquisition (Figure 3, top left). For TE=30ms at 3T versus the 44ms TE protocol (Figure 3, middle column), there is a 4% increase in tSNR in the frontal region, accompanied by a larger, 6%, loss of CNR_BOLD in the cortex. For 7T the choice for a shorter TE might be more beneficial, as here the through-slice dephasing in a 2-mm slab is more significant.

The use of TE values shorter than T₂^* has become commonplace to allow for the acquisition of a large number of slices in a limited TR and to reduce dropout. However, the application of parallel imaging acceleration in the slice-encoding direction means that optimal BOLD contrast is achieved at TE values close(r) to T₂^*, anywhere in the brain at 1.5T and 3T, and also in well-shimmed tissue at 7T.

Considerations on the importance of removing cardiac noise, or studying temporal characteristics of the hrf, have not been taken into account here, but could justify sacrificing BOLD sensitivity for temporal resolution.