Introduction

The entorhinal cortex (EC) is a small brain structure located in the medial temporal lobe (Fig.1) below the hippocampus. It interfaces between the hippocampus and cortex playing an important role in memory formation and consolidation^1,2. Performing cortical depth-dependent (laminar) fMRI in the EC has potential to provide valuable insights into the EC-hippocampus-cortex network, e.g. in visual perception. However, the EC is plagued by high levels of artefact, particularly signal loss. We investigated the underlying causes of dropout in the EC and how this is affected by excitation (specifically binomial pulse order).

Theory

Signal, $$$S= S_0 e^{-t⁄T_2^*}$$$, on gradient-echo 3DEPI images is low in the EC (Fig.1). Mitigation strategies depend on the underlying reason for low signal. We therefore explored the following potential factors:

1. Intra-voxel B₀ inhomogeneity: shortens T₂^' and therefore T₂^*
2. B₀ inhomogeneity interacting with the EPI readout: results in spatially varying TE and complete dropout when the echo forms outside the k-space acquisition window³
3. Transmit field (B₁⁺) inhomogeneity: results in non-ideal excitation (lowers S₀)
4. B₀ offset: lowers excitation efficiency (lowers S₀) when using frequency-selective (“water excitation”) binomial pulses

Spatial resolution and angulation could be manipulated to mitigate points 1,2 above, but are typically dictated by the research question, especially when studying multiple regions (e.g. EC-hippocampus-V1), and by temporal resolution requirements. Changes to the excitation (mitigating points 3,4) can be implemented relatively easily. We typically use a binomial 1-3-3-1 pulse train for efficient water-selective excitation and fat-suppression. Here we explored lowering the binomial pulse order (to 1-2-1 or 1-1) to increase signal in the EC by increasing the width of the excitation’s pass-band at the cost of narrowing the stop-band (Fig.3).

Methods

One healthy volunteer was scanned at 7T. We acquired three 226-second runs of 0.92mm 3DEPI (acceleration=4x2, in-plane segmentation⁴ factor=2 to reduce distortions, Partial Fourier(early) = 6/8, TE/TR=18/40ms, flip angle=13°, echo spacing=1.2ms, matrix size=208x208x88, posterior-anterior phase-encoding direction). The acquisitions used either 1-3-3-1, 1-2-1 or 1-1 binomial excitation pulses with an inter-pulse spacing of 1500μs. Data were also collected to provide an anatomical reference (0.65mm MP2RAGE) and map B₁⁺ (Bloch-Siegert 4mm), T₂^* and B₀ (multi-echo FLASH: 0.9mm, 8 echoes, TE₁=2.52ms, ΔTE=2.52ms). Combining the B₀ map and readout parameters, we calculated a local TE map³ and determined if the echo is sampled.

From the 3DEPI time-series mean functional images and tSNR were calculated. We evaluated B₀, T₂^*, local TE and B₁⁺ in EC, V1, Hippocampus and Orbitofrontal cortex (OFC) ROIs.

The excitation efficiency of binomial pulses was simulated (as a series of hard pulses) over a range of frequencies and inter-pulse periods. For comparisons with experimental data noise (SNR≈30) was added to the simulated excitation efficiency. Following frequency-specific smoothing, maps of signal gain were computed as the ratio of the signal from 1-2-1 or 1-1 excitations to that of 1-3-3-1 and compared to B₀.

Results

The EC contains a broad range of B₀ offsets (Fig.2, Median frequency=-116Hz, IQR=189Hz) encompassing both positive and negative frequencies. The off-resonance is larger compared to other regions such as V1, Hippocampus and even the notoriously difficult-to-image OFC (Median=62Hz, IQR=154Hz). T₂^* mapping indicated that parts of the EC suffer from short T₂^*: 22% of voxels have T₂^*<10ms and can be considered completely dephased at TE=18ms. The strong local field gradients lead to variable local TE, with positive and negative shifts. 27% of voxels in the EC suffered complete dropout because the echo was outside the acquisition window. 36% of voxels were lost to the combination of these two factors (the overlap is due to both being driven by the local field gradients). The B₁⁺ in the EC is homogeneous and close to target efficiency and therefore not considered a driving factor for low EC signal levels.

Binomial pulse order dictates the frequency response of the excitation (Fig.3). Whole-brain analysis showed good qualitative correspondence between simulation and experiment as pulse order changes (Fig.4). As predicted, signal gains occurred as off-resonance increases, especially in the range 100-300Hz, and subsequently decrease.

In the EC we observed the same pattern with improvements for off-resonant voxels (Fig.5). The improved excitation translated into improved tSNR for the 1-1 excitation (median tSNR=10.0, IQR=8.2) compared to 1-3-3-1 (median tSNR=7.0, IQR=8.2).

Discussion

The large and geometrically complex local distortion of the B₀ field makes fMRI of the EC very challenging. Reducing the binomial pulse order to 1-2-1 or even 1-1 is a simple and effective strategy to boost signal in off-resonance regions but comes at the cost of reduced fat suppression. No fat artefact was visible in these data but was occasionally seen in other volunteers and protocols (data not shown).

Even after improvement with the 1-1 pulse the tSNR in the EC remains low, with a substantial number of voxels (almost) completely dephased. The large spread of B₀ values (including both positive and negative frequency offsets/gradients) makes mitigation challenging. For many strategies, such as shortening nominal TE, we would expect to gain in some areas of the EC but lose in others. Further development is needed to make EC fMRI at 7T robust. Due to the lower field inhomogeneity, 3T fMRI may compete with 7T fMRI for the EC, especially if combined with an SNR-enhancing technique, e.g. NORDIC⁵.

Acknowledgements

The Wellcome Centre for Human Neuroimaging is supported by core funding from the Wellcome [203147/Z/16/Z].

References

1. Staresina, B.P., Reber, T.P., Niediek, J. et al. Recollection in the human hippocampal-entorhinal cell circuitry. Nat Commun 10, 1503 (2019)
2. Koster, Raphael et al. Big-Loop Recurrence within the Hippocampal System Supports Integration of Information across Episodes. Neuron vol. 99,6 (2018): 1342-1354.e6.
3. Dymerska, Barbara et al. The Impact of Echo Time Shifts and Temporal Signal Fluctuations on BOLD Sensitivity in Presurgical Planning at 7 T. Investigative radiology vol. 54,6 (2019): 340-348.
4. Stirnberg, R, Stöcker, T. Segmented K-space blipped-controlled aliasing in parallel imaging for high spatiotemporal resolution EPI. Magn Reson Med. 2020; 85: 1540-1551.
5. Moeller, Steen et al. NOise reduction with DIstribution Corrected (NORDIC) PCA in dMRI with complex-valued parameter-free locally low-rank processing. NeuroImage. 2021; 226: 1053– 8119.