Fuzzy ripple artifact in high resolution fMRI: identification, cause, and mitigation

Abstract

Purpose: High resolution fMRI is a rapidly growing research field focused on capturing functional signal changes across cortical layers. However, the data acquisition is limited by low spatial frequency EPI artifacts; termed here as Fuzzy Ripples. These artifacts limit the practical applicability of acquisition protocols with higher spatial resolution, faster acquisition speed, and they challenge imaging in lower brain areas.

Methods: We characterize Fuzzy Ripple artifacts across commonly used sequences and distinguish them from conventional EPI Nyquist ghosts, off-resonance effects, and GRAPPA artifacts. To investigate their origin, we employ dual polarity readouts.

Results: Our findings indicate that Fuzzy Ripples are primarily caused by readout-specific imperfections in k-space trajectories, which can be exacerbated by inductive coupling between third-order shims and readout gradients. We also find that these artifacts can be mitigated through complex-valued averaging of dual polarity EPI or by disconnecting the third-order shim coils.

Conclusion: The proposed mitigation strategies allow overcoming current limitations in layer-fMRI protocols: (1)Achieving resolutions beyond 0.8mm is feasible, and even at 3T, we achieved 0.53mm voxel functional connectivity mapping.(2)Sub-millimeter sampling acceleration can be increased to allow sub-second TRs and laminar whole brain protocols with up to GRAPPA 8.(3)Sub-millimeter fMRI is achievable in lower brain areas, including the cerebellum.

Keywords: 7T acquisition; fuzzy ripples; layer-fMRI; ventral brain.

Fig 1:. Fuzzy Ripples are the reason why layer-fMRI is confined to conventional protocols.

Fuzzy Ripples are the primary reason why layer-fMRI is restricted to conventional protocols. Standard layer-fMRI protocols are generally limited to 0.8mm resolution with TRs of several seconds, focusing on upper cortical brain areas. These limitations cannot be surpassed because, with more ambitious acquisition protocols. Fuzzy Ripple artifacts become too strong and too frequent. Panel A) exemplifies issues of pushing TR. Panel B) exemplifies issues of pushing resolutions beyond 0.8mm. Panel C) exemplifies issues of pushing protocols towards lower brain areas. Images from Panels A and B are taken from Koiso 2023 and Huber 2023, respectively. They refer to 3D-EPI readouts with planar EPI trajectories.

Fig 2:. Fuzzy Ripples and their impact on layer-fMRI research Panel

A). The widespread effect of fuzzy ripples: Representative EPI images from among top 10 layer-fMRI labs (based on number of publications on www.layerfmri.com/papers): Maastricht, Nijmegen/Essen, CMRR, NIH, MGH, Amsterdam, Leipzig, Cambridge. Note Utrecht/Tübingen are excluded, as despite being among the top 10 layer-fMRI labs, none of their papers include publicly available layer-fMRI EPI data. Panel B). The Necessity of High-Resolution fMRI in Lower Brain Areas. High-resolution fMRI in lower brain areas is crucial for addressing key open questions about human brain circuitry. While most layer-fMRI studies have focused on upper brain regions, lower brain areas are equally important for fundamental neuroscience research. Examples of relevant research topics include: feedforward vs. feedback processing in layers of ventral cortical areas like the FFA/PPA, differential processing in mesoscale subnuclei of the amygdala, laminar differentiation of memory encoding and retrieval in the hippocampus and entorhinal cortex, multi-modal sensory integration across the colliculi, and mesoscale representations of body parts in the fine-scale lobules of the cerebellum. The underlay images in Fig. 2 were generated using the Brain Tutor app for Android by Brain Innovations (Rainer Goebel).

Fig 3:. Concept of Fuzzy ripples as EPI odd-even delays with ramp sampling with readout (here k_x)-specific phase errors.

A) Gradient imperfections in high resolution EPI are most prominent at the corners of trapezoids,including both the rising and falling edges. The nominal EPI trajectory is derived from SIEMENS IDEA simulations, while the measured trajectory is obtained using SKOPE on a standard SIEMENS 7T MAGNETOM with third-order shimming. B) Despite phase correction in the image reconstruction process, residual imperfections persist in the part of k-space that encodes lower spatial frequencies. These residual errors, shown as deviations in k_x, manifest as irregular k-space grids for odd and even lines. Such odd-even errors are expected to produce EPI ghosting artifacts in low spatial frequencies. B-C) This study introduces a strategy to address Fuzzy Ripple artifacts by employing a dual-polarity EPI approach that alternates the read direction every other TR (see methods section). EPI images with opposite read directions are anticipated to produce ghosts with opposite phases. Option 2 refers to an alternative future implementation that is referred to in the discussion section.

Fig 4:. Interaction of Fuzzy Ripples with other common EPI artifacts: GRAPPA ghosts and static off-resonance ghosts.

This figure illustrates EPI acquisitions with different combinations of ramp sampling, poor B0 shim, and GRAPPA. The unusually large field of view (FOV) was purposefully chosen to detect peripheral ghost artifacts. Signal differences between reverse EPI polarity images are shown to emphasize spatial ghost patterns that might be too subtle to observe with conventional image intensity windowing. The read direction in left-right, phase encoding direction is anterior-posterior. A) Without ramp sampling, imaging data are acquired only during the flat top of the gradient waveform. This minimizes the impact of large gradient errors, resulting in relatively weak Fuzzy Ripples in the EPI images. B) With ramp sampling enabled, EPI becomes more sensitive to the largest gradient errors, causing Fuzzy Ripples to intensify. These ripples appear as aliasing of low spatial frequencies, with no sharp edges evident in the phase encoding direction. C) GRAPPA, which relies on a known aliasing pattern, is affected by erroneous Fuzzy Ripple ghosts and thus amplifies their impact. D) This differs from static off-resonance effects. For instance, with suboptimal shimming (deliberately altered in this case), the off-resonance effects do not amplify the low-spatial frequency Fuzzy Ripples. Instead, they introduce edge ghosts at high spatial frequencies, which differ from Fuzzy Ripples in their appearance. E) The dual-polarity averaging approach effectively mitigates both sources of artifacts. The resulting images are nearly artifact-free. Acquisition parameters of data presented here are mentioned in methods section 3.3. See supplementary Figure S1 for the reproducibility of these results in another participant and on another scanner.

Fig 5:. Impact of 3rd order shim-induced Fuzzy Ripples on fMRI activation detectability.

This figure demonstrates how Fuzzy Ripples, induced by the 3rd order shim, can affect the detectability of fMRI activation. These data refer to 2D-EPI with block designed auditory activation with NORDIC denoising. When the 3rd order shim is connected, the Fuzzy Ripples can be so pronounced that they mask parts of the auditory activation, preventing it from reaching the detection threshold. White arrows indicate areas where the Fuzzy Ripples are more intense with the 3rd order shim engaged. Although Fuzzy Ripples are still present when the 3rd order shim is disconnected, they are less severe. Acquisition parameters of data presented here are mentioned in methods section 3.4. See Supplementary Fig. S2 for a replication of these findings in a different participant.

Fig 6:. 3^rd-order shim-induced Fuzzy Ripples as a function of echo spacing, dual-polarity averaging, 3rd order shim.

A) The Fuzzy Ripple artifact varies with the echo spacing of the EPI readout. Consequently, the strength of this artifact can be reduced by adjusting the readout protocol, although such adjustments may compromise TE and readout efficiency. B) The Fuzzy Ripples induced by the third-order shim can be mitigated by disconnecting its circuit. Opening this circuit reduces the inductive coupling between the third-order shim and the gradient and reduces Fuzzy Ripple artifacts. C) As indicated by Figures 3 and 4, dual polarity averaging can counteract the Fuzzy Ripples induced by the third-order shim. This approach can mitigate fuzzy ripples, even for the most problematic echo spacings with the third-order shim still connected. Though, faint residual fuzzy ripples remain. Acquisition parameters of data presented here are mentioned in methods section 3.5. Supplementary Figure S3 presents a reproduction of the results shown here.

Fig 7:. Dual polarity averaging with respect to other popular sequences.

All sequences are used with the same resolution, echo spacing, and acceleration parameters. A) This panel shows the CMRR multiband sequence with these protocols, where off-resonance effects and Fuzzy Ripple artifacts are clearly visible. B) This panel displays the MGH simultaneous multi-slice sequence with the option of dual polarity GRAPPA. Whileoff-resonance effects are mitigated, Fuzzy Ripple artifacts, though reduced, remain visible. C) This panel illustrates the same protocols using 3D-EPI. Due to its different Mz steady-state behavior, 3D-EPIinherently has a higher signal-to-noise ratio (SNR). Additionally, off-resonance effects are less noticeable, as they are smeared and partially averaged out. However, 3D-EPI still suffers from Fuzzy Ripples. D) This panel depicts 3D-EPI with dual polarity averaging. It is visible that Fuzzy Ripples are effectively mitigated. Acquisition parameters of data presented here are mentioned in methods section 3.6. Supplementary Figure S4 presents a reproduction of the results shown here.

Fig. 8:. Examples of high resolution protocols pushing the limits of conventional protocols.

The individual panels illustrate that achieving high spatial resolution, rapid sampling (through acceleration), and imaging of lower brain areas is challenged by Fuzzy Ripples (as demonstrated in Figure 1). However, dual polarity averaging can mitigate these challenges, enabling the extension beyond the current limitations of conventional layer-fMRI protocols. A) 3T Prisma, 3D-EPI with GRAPPA 3, 15-minute movie-watching paradigm, resolution of 0.53 mm. Acquisition parameters of the data shown here are described in methods section 3.2. B) 7T Terra, 3D-EPI, GRAPPA 3, three times 14 min checkerboard, resolution 0.82mm, TR_vol=0.98s for 14 slices. C) 7T Terra, 3D-EPI, GRAPPA 3, three averages of a 12-minute finger-tapping experiment, resolution of 0.82 mm. Acquisition parameters of the data shown here are described in methods section 3.7.