Optimising 7T-fMRI for imaging regions of magnetic susceptibility

31 The temporal signal-to-noise ratio (tSNR) of functional magnetic resonance imaging (fMRI) is 32 particularly poor in ventral anterior temporal and orbitofrontal regions because of magnetic 33 field inhomogeneity, a problem that is exacerbated at higher field strengths. In this 7T-fMRI 34 study we compared three methods of improving sensitivity in these areas: parallel transmit, 35 which uses multiple transmit elements, controlled independently, to homogenise the flip 36 angle experienced by the tissue; multi-echo, which entails collection of multiple volumes at 37 different echo times following a single radiofrequency pulse; and multiband, in which 38 multiple slices are acquired simultaneously. We found that parallel transmit and multi-echo 39 increased the magnitude of the BOLD signal change, but only multi-echo increased BOLD 40 magnitude in areas prone to susceptibility artefacts. Multiband and denoising of multi-echo 41 data with independent components analysis (ICA) both improved precision of GLM fit. 42 Exploratory results suggested that multi-echo and ICA denoising can both benefit 43 multivariate analyses. In conclusion, a multi-echo, multiband sequence improved fMRI 44 quality in areas prone to susceptibility artefacts while maintaining sensitivity across the 45 whole brain. We recommend this approach for studies investigating the functional roles of 46 ventral temporal and orbitofrontal regions with 7T fMRI. 47 48

7T-fMRI, parallel transmit, multi-echo, multiband, temporal lobe, orbitofrontal 49 cortex 50 51 52 53 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 3 1. Introduction 54 The temporal signal-to-noise ratio (tSNR) of functional magnetic resonance imaging 55 (fMRI) varies across the brain. The ventral anterior temporal cortex and orbitofrontal cortex, 56 for example, are located next to air-filled sinuses and so are affected by magnetic field 57 inhomogeneity that causes signal dropout and geometric distortions (Devlin et al., 2000; 58 Halai et al., 2014, 2015, 2024). This makes it challenging to use fMRI to investigate the roles 59 that these regions may play in a myriad of cognitive processes – including vision (Devereux 60 et al., 2018), language (Borghesani et al., 2016), multimodal semantic cognition (Lambon 61 Ralph et al., 2017), emotion (Fernandez et al., 2017), social cognition (Binney et al., 2016; 62 Zahn et al., 2007), theory of mind (DuPre et al., 2016), and executive function (Duncan, 63 2010). Signal dropout and geometric distortions are more severe at higher field strengths, 64 implying that it may be especially difficult to measure task-related activity in susceptible 65 regions with ultra-high-field fMRI (e.g., 7T-fMRI). However, 7T-fMRI also has many 66 advantages in regions unaffected by magnetic susceptibility artefacts: 7T-fMRI offers 67 improved tSNR relative to 3T-fMRI (Morris et al., 2019), which can be used to reduce voxel 68 size and enable applications such as laminar fMRI (Koopmans et al., 2011) or to reduce 69 acquisition times and enable shorter scan times for special populations such as patients with 70 neurodegenerative diseases (Cope et al., 2023). 7T-fMRI also benefits from improved spatial 71 specificity relative to 3T-fMRI because the signal from cortical microvasculature is enhanced 72 while the signal from large veins is reduced (Marques & Norris, 2018). Improving signal 73 homogeneity would allow researchers studying the whole brain (or focusing on regions 74 prone to susceptibility artefacts) to take full advantage of 7T-fMRI; therefore, in this study 75 we compared three methods of doing so. 76 One possible method for counteracting signal dropout is parallel transmit (pTx), 77 which uses multiple transmit elements, controlled independently, to homogenise the flip 78 angle pattern experienced by the tissue (Deniz et al., 2019). A recent study used pTx to 79 counteract signal dropout in ventral anterior temporal regions for echo-planar imaging (EPI) 80 7T fMRI (Ding et al., 2022). pTx improved tSNR across the brain compared to a standard 81 sequence, particularly in the temporal lobes. However, there was no improvement in 82 functional contrast during a semantic association task that is known to recruit the anterior 83 temporal lobes in 3T-fMRI studies (Jung et al., 2017). 84 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 4 A second method of recovering signal in these regions is multi-echo (ME) imaging 85 (Kundu et al., 2017; Poser et al., 2006; Posse, 2012). T2* is known to vary across the brain 86 (Hagberg et al., 2002); in areas prone to magnetic susceptibility artefacts, T2* is particularly 87 short due to increased intravoxel dephasing. A single echo provides sensitivity to a narrow 88 range of T2* values; the echo time (TE) is therefore selected to provide the best compromise 89 of sensitivity to T2* across the whole brain. Combining data from multiple echoes increases 90 the range of T2* that can be imaged with high fidelity. ME has been shown to improve 91 functional contrast (Poser & Norris, 2009) and spatial specificity (Boyacioğlu et al., 2015) at 92 7T and 3T (Fernandez et al., 2017; Halai et al., 2024; Kirilina et al., 2016; Lynch et al., 2020). 93 Having multiple echoes also facilitates the separation of signal and noise because signals 94 decay in a well-characterised way across echoes, whereas noise does not. This principle 95 underpins multi-echo independent components analysis (ME-ICA), via which ICA 96 components that are TE-independent, and thus are likely to be noise rather than blood-97 oxygen-level-dependent (BOLD) signal, can be removed (Dipasquale et al., 2017; Kundu et 98 al., 2011, 2013, 2015, 2017). This method may enhance signal detection in areas prone to 99 susceptibility artefacts on top of the advantage offered by ME alone (e.g. Lombardo et al., 100 2016). ME sequences have some potential disadvantages. For example, multi-echo can 101 lengthen repetition time (TR). In-plane acceleration is frequently needed to achieve a 102 sufficiently short first TE, which reduces tSNR (Yun & Shah, 2017). In turn, a short first TE, 103 combined with hardware constraints, often limits the minimum voxel size (Koopmans et al., 104 2011). Critically, in many previous studies examining the benefits of ME sequences 105 compared to single-echo (SE), the “single-echo” data were extracted from the ME dataset 106 (Amemiya et al., 2019; Bhavsar et al., 2014; Caballero-Gaudes et al., 2019; Cohen et al., 107 2017, 2017, 2018; Dipasquale et al., 2017; Evans et al., 2015; Fernandez et al., 2017; Gilmore 108 et al., 2022; Kovářová et al., 2022). This means that the “single-echo” data will inherit 109 suboptimal parameters that are ME-specific, making the comparison unfair. 110 A third strategy for improving acquisition is multiband (MB) imaging, also known as 111 simultaneous multi-slice, in which multiple slices are acquired simultaneously (Barth et al., 112 2016; Moeller et al., 2010; Setsompop et al., 2012). Typically MB has been applied to 113 increase temporal resolution – multiple studies have shown that MB can reduce noise 114 aliasing, increase statistical power and counteract the increase in TR associated with ME 115 (Feinberg et al., 2010; Griffanti et al., 2014; Halai et al., 2024; Puckett et al., 2018; Smith et 116 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 5 al., 2013). Benefits specific to task fMRI have been less clear (Demetriou et al., 2018; Todd 117 et al., 2016) since some noise features are unlikely to be correlated with task regressors and 118 can be removed with high-pass filtering. Other possible disadvantages of MB include a 119 reduction in tSNR due to increases in g-factor effects (Demetriou et al., 2018; Risk et al., 120 2021; Setsompop et al., 2012) and leakage of signal into the simultaneously-excited slices 121 (Todd et al., 2016). 122 There is therefore a need to evaluate 7T-fMRI sequences to determine which 123 parameters are important for counteracting signal dropout and improving sensitivity in 124 regions prone to susceptibility artefacts without compromising signal quality elsewhere. We 125 tested five possible sequences. These consisted of pTx, plus a 2 x 2 factorial design varying 126 number of echoes and multiband factor: single-echo single band (SESB), single-echo 127 multiband (SEMB), multi-echo single band (MESB), and multi-echo multiband (MEMB). We 128 used a semantic judgment task that is known to evoke activity across the semantic network, 129 including areas severely affected and those relatively unaffected by susceptibility artefacts 130 (Jung et al., 2017). 131 We had two univariate effects of interest – activation magnitude and activation 132 precision (Halai et al., 2024). Activation magnitude is the magnitude of the BOLD signal 133 change, operationalised as the 1st-level beta values extracted from each voxel. We 134 hypothesised that both pTx and ME would recover signal and hence increase activation 135 magnitude relative to the SESB sequence (which we used as a baseline). Activation precision 136 is the reliability of the BOLD signal change, analogous to the functional contrast-to-noise 137 ratio (fCNR) and operationalised as the 1st-level t-values extracted from each voxel. We 138 hypothesised that MB sequences, with greater effective degrees of freedom, would increase 139 activation precision relative to single band (SB) sequences. We also had two secondary 140 hypotheses: first, since ME-ICA denoising removes TE-independent noise, greater activation 141 precision would be observed in multi-echo denoised data (MEdn) relative to ME data 142 without denoising; second, that any MB advantage would be due to the increase in the 143 number of volumes. 144 Our study focused primarily on univariate effects. However, multivariate analysis 145 techniques, which exploit variance and covariance between voxels and can accommodate 146 participant-specific differences in activation patterns (Coutanche, 2013; Davis et al., 2014; 147 Davis & Poldrack, 2013) are rapidly gaining popularity. These methods frequently rely on the 148 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 6 assumption of good-quality signal across the whole brain (Frisby et al., 2023) and so we 149 conducted an exploratory analysis, following the method of Haxby et al. (2001), to decode 150 task condition from the data acquired with each sequence. Finally, we tested for the 151 presence of slice leakage artefacts in our MB data. 152 To summarise, our study aimed to compare pTx, ME and MB as methods for 153 improving sensitivity in ventral temporal and orbitofrontal regions while maintaining image 154 quality across the rest of the brain. 155 156 2. Methods 157 2.1. Participants 158 20 healthy native speakers of British English (age range 18-50, mean age 33.45 years, 12 159 female, 8 male) participated in the study. All were right-handed, had normal or corrected-160 to-normal vision, and had no neurological or sensory disorders. All participants gave written 161 informed consent. The research was approved by a local National Health Service (NHS) 162 ethics committee (04/Q105/66). 163 164 2.2. Stimuli and task 165 All participants performed a semantic association task and a visual pattern matching task 166 (hereafter called the “control task”) adapted from a previous study (Jung et al., 2017). Each 167 stimulus consisted of three pictures presented simultaneously (Figure 1). Some pictures 168 were line drawings taken from the Pyramids and Palm Trees Test (Howard & Patterson, 169 1992) and some were colour cartoons or photographs taken from the Camel and Cactus Test 170 (Bozeat et al., 2000). In the semantic task, participants were instructed to indicate which of 171 the two pictures at the bottom of the screen had the closest semantic relationship to the 172 picture at the top (hereafter the “probe picture”). In the control task, participants were 173 instructed to indicate which of two scrambled pictures (generated from the pictures used in 174 the semantic task) at the bottom of the screen was identical to a scrambled probe. There 175 were 248 unique picture triplets, so some stimuli were repeated between runs, but no 176 stimulus was repeated within a run. E-Prime software (Psychology Software Tools Inc., 177 Pittsburgh, USA) was used to display stimuli and record responses. Stimuli were rear-178 projected onto a screen at the back of the MRI scanner, and observers viewed stimuli 179 through a mirror mounted to the head coil directly above the eyes. 180 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 7 181 Figure 1: One block of the semantic association task (left of the arrow) and one block of the 182 control task (right of the arrow). Each trial consisted of a fixation cross (500 ms) followed by 183 three pictures presented simultaneously. Participants had to select which of the two 184 pictures at the bottom was more semantically similar to (in the semantic task) or visually 185 identical to (in the control task) the picture at the top (the probe picture). Each block lasted 186 16 s in total. Reprinted with permission from (Halai et al., 2024). 187 188 The study had a block design with three types of block: semantic, control and rest. 189 Each task block consisted of four trials. Each trial consisted of a fixation cross presented for 190 500 ms followed by a stimulus presented for 3500 ms. Each rest block consisted of a fixation 191 cross presented for 16 s. Each run began 16 s after the start of the MR sequence and then 192 consisted of 30 blocks presented in the order: semantic, control, semantic, control, rest. 193 There were five runs per participant, collected in a single session. Accuracy and reaction 194 time were measured for each trial. Since reaction time is not normally distributed, both 195 metrics were compared across tasks using Wilcoxon’s signed-rank tests and across 196 sequences using Friedman’s nonparametric ANOVA. 197 198 2.3. Image acquisition 199 All images were acquired on a whole-body MAGNETOM Terra 7T MRI (Siemens 200 Healthcare, Germany). An 8Tx32Rx head coil (Nova Medical, USA) was used to run all 201 structural and functional imaging sequences. 202 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 8 An MP2RAGE anatomical scan was acquired with the following parameters: 224 203 sagittal slices (interleaved acquisition), FOV 240 x 225.12 x 240 mm3, voxel size 0.75 mm 204 isotropic, TR 4300 ms, inversion times 840 and 2370 ms, TE 1.99 ms, nominal flip angles 5° 205 and 6°, GRAPPA acceleration factor 3, and duration 8 minutes 50 seconds. 206 Next, manual B0-shimming was performed over the volume to be used for EPI. The 207 aim was to reduce the water linewidth, defined as full width of the spectrum at half height, 208 to below 40 Hz. However, for some participants the adjustments proved time-consuming 209 and so adjustment time was capped at 30 minutes from the start of scanning after which 210 the best shim parameters were adopted. The actual average water linewidth was 42 Hz 211 (standard deviation = 9 Hz; missing data for 4 participants). Then a dummy pTx-EPI scan of 212 one volume was acquired to trigger the acquisition of subject-specific B0 and per-channel B1+ 213 field maps. The brain was divided into 5 slabs along the slice direction and slab-specific 2-214 spoke pTx excitation pulses were designed offline. Variable-rate selective excitation (VERSE; 215 Hargreaves et al., 2004) was applied to reduce specific absorption rate (SAR) for the pTx 216 sequence only. 217 There were five functional runs of EPI, one run of each sequence. Key parameters 218 are given in Table 1. The order of sequences was counterbalanced across participants. The 219 following parameters were held constant across sequences: 48 axial slices (interleaved 220 acquisition), FOV 210 x 210 x 210 mm3 to cover the whole brain in most participants (visual 221 inspection ensured that the ventral anterior temporal lobe was included in the FOV for all 222 participants and the FOV was tilted up at the nose to avoid ghosting of the eyes into the 223 temporal lobe), voxel size 2.5 mm isotropic (no gap), and A-P phase encoding direction. 224 After each run, 5 further volumes were acquired with the phase encoding direction changed 225 to P-A to facilitate distortion correction during preprocessing. 226 227 SESB SEMB MESB MEMB pTx No. echoes 1 1 3 3 1 Multiband factor 1 2 1 2 1 TR (ms) 3020 1510 3020 1510 3000 TE1 (ms) 25.00 25.00 11.80 11.80 25.00 TE2 (ms) - - 27.05 27.05 - TE3 (ms) - - 42.30 42.30 - iPAT type Off Off GRAPPA GRAPPA GRAPPA iPAT factor Off Off 3 3 2 Phase partial Fourier 7/8 7/8 7/8 7/8 Off .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 9 Nominal flip angle (°) 50 50 78 63 40 Bandwidth (Hz/Px) 2204 2204 2204 2204 1804 Number of volumes acquired 171 340 171 340 172 Number of pulses 1 1 1 1 5 (VERSE 2- spoke pTx pulses) 228 Table 1: parameters of each sequence. 229 230 2.4. Data analysis 231 All analysis code is available at https://github.com/slfrisby/7TOptimisation/. 232 233 2.4.1. Preprocessing 234 For ease of data sharing we converted all DICOMs to BIDS format (Gorgolewski et al., 235 2016) using heudiconv v1.0.0 (Halchenko et al., 2024). 236 Standard reproducible preprocessing pipelines designed for 3T-fMRI, such as 237 fMRIprep (Esteban et al., 2019; Gorgolewski et al., 2011; Markiewicz et al., 2024) perform 238 poorly on 7T-fMRI EPI data. Therefore, the analysis pipeline was split into stages using 239 different software packages. 240 Since it is notoriously difficult to perform a good-quality brain extraction on 241 MP2RAGE data (because of salt-and-pepper noise in the background and cavities), two 242 pipelines were used. The MP2RAGE T1w (combined) image first had its background noise 243 removed using O’Brien regularisation (O’Brien et al., 2014) and was then submitted to the 244 CAT12 pipeline for segmentation (Gaser et al., 2023; in SPM12; 245 https://www.fil.ion.ucl.ac.uk/spm/). The bias- and global-intensity corrected T1w image 246 produced was provided to as input to the anatomy pipeline in fMRIPrep 21.0.1. TheT1w 247 image was skull-stripped with a Nipype implementation of the antsBrainExtraction.sh 248 workflow (ANTs 2.3.3; Avants et al., 2009, 2011; https://github.com/ANTsX/ANTs/). Volume-249 based spatial normalisation to standard space (MNI152NLin2009cAsym) was performed 250 through nonlinear registration with antsRegistration.sh (ANTs). 251 Functional preprocessing was performed using in-house code, composed of 252 functions from AFNI (v.18.3.03; Cox, 1996; Cox & Hyde, 1997; https://afni.nimh.nih.gov/), 253 FSL (v.5.0; Andersson et al., 2003; Jenkinson et al., 2012; Smith, 2002; Smith et al., 2004; 254 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 10 https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/), tedana (v.23.0.1; DuPre et al., 2021; Kundu et al., 255 2011, 2013; The tedana Community et al., 2023; 256 https://tedana.readthedocs.io/en/stable/index.html) and ANTS (v. 2.2.0; Avants et al., 2009, 257 2011; https://github.com/ANTsX/ANTs/). EPIs were despiked using 3dDespike (AFNI), slice 258 timing was corrected to the middle slice using 3dTshift (AFNI), motion was corrected with 259 3dvolreg and 3dAllineate (AFNI) using the first volume of each run as a reference (for ME 260 datasets, the TE1 image was aligned and the resulting transform was applied to the TE2 and 261 TE3 images), and skull-stripped using BET (FSL) to create a participant-specific brain mask. 262 For ME datasets only, tedana was used to create two timeseries – one in which the 263 echoes were optimally combined based on T2* weighting (Posse, 2012), and one in which 264 the T2* optimally-combined data were denoised using ICA. tedana conducts denoising by 265 decomposing data using PCA and ICA, classifying components according to whether the 266 signal scales linearly with TE (as the BOLD signal does), and reconstructing the data using 267 only BOLD-like components. The brain mask created with BET was used as the mask for this 268 stage. 269 Next, for all datasets, unwarping was conducted using topup and applytopup (FSL). 270 Field displacement maps were calculated using ten volumes (five with A-P phase encoding 271 direction, extracted from the start of each functional run, and five with P-A phase encoding 272 direction, collected separately after each run) and the resulting correction was applied to all 273 images. Finally, the mean EPI for each run was coregistered to the skull-stripped native 274 structural image using a rigid-body registration with AntsRegistrationSyN.sh (ANTS). EPIs 275 were then transformed into standard space (MNI152NLin2009cAsym) by combining the 276 transforms from native EPI to native T1 and the transforms from native T1 to standard space 277 and applying those transforms to the EPIs using antsApplyTransforms (ANTS). Images were 278 smoothed with a 6 mm FWHM Gaussian filter in SPM12 (https://www.fil.ion.ucl.ac.uk/spm/) 279 for GLM analysis. 280 A separate functional preprocessing pipeline was used to create images for slice 281 leakage analysis. This pipeline differed from the main pipeline in the following ways. 282 Despiking was omitted to avoid the removal of noise. For ME datasets only, rather than run 283 the full tedana workflow, we conducted optimal combination of data from multiple echoes 284 (but no denoising) using the t2smap command using all voxels in the volume (i.e. no brain 285 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 11 mask). All coregistration steps were omitted (the images remained in native EPI space) and 286 no smoothing was applied. 287 288 2.4.2. 1st-level (within-participant) GLM 289 Data were analysed using the general linear model (GLM) approach implemented in 290 SPM12 in MATLAB r2019a. We had 5 timeseries of primary interest – standard single-echo 291 single band (SESB), parallel transmit (pTx), single-echo multiband (SEMB), multi-echo single 292 band (MESB) and multi-echo multiband (MEMB). For the latter two sequences, data from all 293 echoes were optimally combined but were not ICA-denoised. We also generated two 294 timeseries with ME-ICA denoising - multi-echo single band with ICA denoising (MESBdn) and 295 multi-echo multiband with ICA denoising (MEMBdn) - and two downsampled MB timeseries 296 created by extracting odd-numbered volumes to match the number of volumes in the single 297 band timeseries - odd-numbered volumes of single-echo multiband (SEMBodd) and odd-298 volumes of multi-echo multiband (MEMBodd). 299 At the individual subject level, each block of the semantic and control task was 300 modelled as a boxcar function (resting blocks were modelled implicitly) and these boxcar 301 functions were subsequently convolved with SPM’s difference of gammas haemodynamic 302 response function. The six motion parameters extracted during preprocessing were used as 303 regressors of no interest. The micro-time resolution was set as the number of slices (n = 48), 304 the micro-time onset was set as the reference slice for slice-time correction (n = 24), and the 305 high-pass filter cutoff was 128 seconds. The same MNI template used during preprocessing 306 was used as a mask for the analysis. The parameter estimation method was restricted 307 maximum likelihood estimation (ReML) and serial correlations were accounted for using an 308 autoregressive AR(1) model during estimation. For univariate analyses the contrast of 309 interest was greater activation for the semantic task than the control task (S > C). For 310 exploratory multivariate pattern analysis (MVPA) each block (12 semantic and 12 control) 311 was modelled individually in order to obtain one beta image per block. 312 Finally, for the slice leakage analysis, the modelling was rerun on the minimally-313 preprocessed timeseries without any brain mask. We obtained both univariate contrasts of 314 interest (S > C) and beta images per block (for MVPA). 315 316 2.4.3. 2nd-level (across-participant) GLM 317 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 12 2.4.3.1. Region-of-interest (ROI) analysis 318 Regions of interest (ROIs) were defined based on a large-scale distortion-corrected 319 3T-fMRI study of the semantic network (Humphreys et al., 2015). For each comparison, ROIs 320 were analysed only if they overlapped by at least one voxel with the whole-brain contrast of 321 interest (S > C) summed over all sequences in the comparison. 322 323 2.4.3.1.1. Univariate analysis 324 Activation magnitude (1st-level beta values) and activation precision (1st-level t-325 values) were extracted from each ROI using a publicly-available script, roi_extract.m 326 (https://github.com/MRC-CBU/riksneurotools/blob/master/Util/roi_extract.m). There were 327 two planned t-tests for activation magnitude (pTx > SESB, ME > SE) and four planned t-tests 328 for activation precision (MB > SB, MEdn > ME, MBodd > SB, SB > MBodd). For each planned 329 t-test, results were Bonferroni-corrected for the number of ROIs included. 330 331 2.4.3.1.2. Exploratory multivariate pattern analysis (MVPA) 332 The input to this analysis was the activation magnitude values extracted from each 333 block individually (12 semantic and 12 control). For each block and each ROI, a vector of 334 beta values within that ROI was created. The cosine dissimilarity between every possible 335 pair of blocks was calculated. MVPA performance was operationalised as the mean 336 between-task dissimilarity minus the mean within-task dissimilarity and paired t-tests were 337 used to compare the metric across sequences (all planned t-tests described in the univariate 338 ROI analysis were conducted; Haxby et al., 2001). 339 340 2.4.3.2. Whole-brain analysis 341 All the above contrasts were assessed at the whole-brain level using t-tests (for 342 comparing pTx and SESB) or using random-effects ANOVAs with one-sample t-tests on the 343 summary statistic (for the three factorial designs: varying echo and band (SESB, SEMB, MESB 344 and MEMB); varying denoising and band (MESBdn, MEMBdn, MESB and MEMB); and 345 varying echo and downsampled band (SEMBodd, MEMBodd, SEMB and MEMB)). The 346 ANOVAs were conducted using a publicly-available script, batch_spm_anova.m 347 (https://github.com/MRC-CBU/riksneurotools/blob/master/SPM/batch_spm_anova.m). The 348 group t-maps were assessed for significance by using a voxel-height threshold of p < 0.001 349 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 13 to define clusters and then a cluster-defining family-wise-error corrected threshold of p C) in standard space was 354 inspected and the coordinates of the peak t-values of the top 5 clusters were identified. 355 These coordinates were back-projected to obtain 5 sets of corresponding coordinates in 356 each participant’s native EPI space (hereafter “seeds”, labelled A). Next, voxels to which 357 signal might be warped were identified as artefact locations based on phase shift (FOV/2, 358 labelled B) and, in the MEMB data, GRAPPA (labelled Ag) and phase shift plus GRAPPA 359 (labelled Bg). A spherical ROI, 4 voxels in radius, was defined around each seed location and 360 possible artefact location using a modified version of the scripts developed for McNabb et 361 al. (2020; https://github.com/DrMichaelLindner/MAP4SL/; our version available at 362 https://github.com/slfrisby/7TOptimisation/). 363 Both univariate (McNabb et al., 2020) and multivariate (Halai et al., 2024) slice 364 leakage tests were conducted. Activation magnitude was extracted from minimally-365 preprocessed data, and, for the multivariate analysis, the difference between mean within-366 task similarity and mean between-task similarity was calculated as in the ROI analysis. 367 Paired t-tests were conducted for each seed and artefact location between each sequence 368 of interest (SEMB and MEMB) and a control sequence. For artefact locations based on phase 369 shift (B), the control sequence was the corresponding SB sequence (SESB for SEMB and 370 MESB for MEMB). For artefact locations based on GRAPPA, the control sequence was the 371 corresponding SE sequence, because SE sequences were collected without GRAPPA (SEMB 372 for MEMB). Results were Bonferroni-corrected for the number of peaks (n = 5) and the 373 number of possible artefact regions (n = 1 for SEMB and n = 3 for MEMB). 374 375 3. Results 376 3.1. Excluded participants 377 Two participants were excluded because of excessive head motion (this was defined 378 by calculating, for each participant, the percentage of volumes per run with absolute 379 translation values of over 2 mm or absolute rotation values over 1°, averaging these 380 percentages over runs, and excluding any participant whose mean percentage was greater 381 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 14 than 2 standard deviations above the mean percentage across participants). One participant 382 was excluded because of technical problems during data acquisition which meant that the 383 pTx run failed. All subsequent analyses were conducted on the remaining 17 participants. 384 385 3.2. Behavioural results 386 The 17 participants had good performance on both tasks and, importantly, there 387 were no significant differences between the 5 sequences in terms of accuracy (Friedman’s χ2 388 = 5.26, p = 0.26) or reaction time (Friedman’s χ2 = 7.20, p = 0.16). The semantic and control 389 tasks did not differ reliably in accuracy (Wilcoxon’s z = 0; p = 0.06) or reaction time 390 (Wilcoxon’s z = 7; p = 1.00). 391 392 3.3. Region-of-interest analysis 393 ROIs that overlapped with the whole-brain contrast of interest (S > C) for at least 394 one of the five sequences are shown in Figure 2. 395 396 Figure 2: Regions of interest, taken from a meta-analysis of semantic tasks by (Humphreys et 397 al., 2015). All spheres are 8 mm in radius. Regions of interest are overlaid on the 398 MNI152NLin2009cAsym template. LTP = left temporal pole; LvATL = left ventral anterior 399 temporal lobe; RITG = right inferior temporal gyrus; LFP = left frontal pole; LmMTG = left 400 medial middle temporal gyrus; LpMTG = left posterior middle temporal gyrus; LIFGpt = left 401 inferior temporal gyrus pars triangularis. 402 403 404 LTP LvATL RITG LFP LmMTG LpMTG LIFGpt Activation magnitude pTx > SESB - 0.2488 0.0240* - - 0.0042** 0.4114 ME > SE 0.0502 0.0001** 0.0333* 0.2951 0.0092* 0.0003* 0.4155 Activation precision MB > SB 0.4569 0.0007** 0.0019** 0.0989 0.2187 0.0266* 0.0016** MEdn > ME 0.3117 <0.0001** <0.0001** 0.0014** 0.1363 <0.0001** SB 0.4225 0.1057 0.0481* - 0.5697 0.1879 0.3300 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 15 MVPA pTx > SESB - 0.8569 0.3280 - - 0.1529 0.9179 ME > SE 0.0096* 0.0063** 0.1612 0.0655 0.0465* 0.0024** 0.2508 MB > SB 0.2455 0.1417 0.0667 0.3021 0.3949 0.2693 0.5207 MEdn > ME 0.0027** <0.0001** MB 0.7416 0.4570 0.7222 - 0.3803 0.6716 0.3242 MBodd > SB 0.2584 0.5430 0.2778 - 0.6197 0.3284 0.6758 405 Table 2: p-values for all t-tests within regions of interest (ROIs) based on the semantic 406 network. * = p < 0.05, ** = p C) 408 summed over all sequences in the comparison, LTP = left temporal pole, LvATL = left ventral 409 anterior temporal lobe, RITG = right inferior temporal gyrus, LFP = left frontal pole, LmMTG 410 = left medial middle temporal gyrus, LpMTG = left posterior middle temporal gyrus, LIFGpt = 411 left inferior temporal gyrus pars triangularis. 412 413 3.3.1. Univariate analysis 414 Table 2 shows the p-values for all planned t-contrasts. pTx provided significantly 415 better activation magnitude than SESB in the left posterior middle temporal gyrus (LpMTG). 416 ME sequences provided significantly better activation magnitude than SE sequences in the 417 left ventral anterior temporal lobe (LvATL). 418 MB sequences offered significantly better activation precision than SB sequences in 419 the LvATL, right inferior temporal gyrus (RITG) and left inferior frontal gyrus pars triangularis 420 (LIFGpt). MEdn sequences offered significantly better activation precision than ME in the 421 LvATL, RITG, LpMTG, LIFGpt and left frontal pole (LFP). There was no significant difference in 422 either direction between MBodd sequences and SB sequences (see Supplementary Table 1 423 for detailed results, plus reverse contrasts). We also found no significant interaction 424 between multi-echo and multi-band. 425 426 3.3.2. Exploratory MVPA 427 Table 2 also shows the p-values for t-tests comparing our MVPA dissimilarity metric 428 between sequences. The ME sequences produced significantly better performance than SE 429 sequences in the LvATL and LpMTG. MEdn sequences produced better performance than 430 ME sequences in every ROI. All other comparisons failed to reach significance. 431 432 3.4. Whole-brain analysis 433 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 16 Figure 3 shows the results for activation magnitude. Figure 3A shows a single cluster, 434 within the right fusiform gyrus and lateral inferior occipital cortex, that showed greater 435 activation magnitude with pTx than with SESB (pTx > SESB). Figure 3B shows the main effect 436 of ME over SE, featuring clusters in the fusiform and inferior temporal gyri bilaterally plus 437 the left orbitofrontal cortex (cluster and peak information is provided in Supplementary 438 Table 2). 439 440 441 Figure 3: Effects on activation magnitude: (A) effect of parallel transmit (pTx > SESB); (B) 442 effect of echo (ME > SE). Results are cluster-corrected at p < 0.05 based on an uncorrected 443 voxel threshold of p < 0.001 and are overlaid on the MNI152NLin2009cAsym template. 444 445 Figure 4 shows the results for activation precision. Both the main effect of MB over 446 SB and the main effect of ME-ICA denoising over ME without ICA denoising extended down 447 the temporal lobe and included frontal regions (Supplementary Table 2). There was no 448 significant difference between downsampled MB sequences and SB sequences in either 449 direction. 450 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 17 451 Figure 4: Effects on activation precision: (A) effect of multiband (MB > SB); (B) effect of ME-452 ICA denoising (MEdn > ME). Results are cluster-corrected at p < 0.05 based on an 453 uncorrected voxel threshold of p < 0.001 and are overlaid on the MNI152NLin2009cAsym 454 template. 455 456 To summarise, these results aligned with results from our ROI analyses – ME 457 increased activation magnitude and both MB and ME-ICA denoising increased activation 458 precision (see Supplementary Figures 1-6 for detailed results, plus reverse contrasts). 459 460 461 3.5. Slice leakage analysis 462 Seed and possible artefact locations for an example participant are shown in Figure 463 5A. Figure 5B shows activation magnitude for an example seed and its corresponding 464 possible artefact locations for all participants in each MB sequence and its corresponding 465 sequences. Figure 5C shows MVPA performance. Violin plots for all other peaks are shown 466 in Supplementary Figure 7. t-tests were conducted at each location, but no comparison 467 reached statistical significance (p-values are shown in Supplementary Table 3). We therefore 468 concluded that there was no evidence for slice leakage in either of our MB datasets. 469 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 18 470 Figure 5: Slice leakage analysis. (A) Seed and possible artefact locations for a single 471 participant. The ROIs (radius 4 voxels) indicate the seed location (blue), possible artefact 472 location based on phase shift (green), possible artefact location based on GRAPPA (red), and 473 possible artefact location based on phase shift and GRAPPA (yellow) in native EPI space. 474 Informed consent was obtained from the participant for this image to be published. (B) 475 Mean activation magnitude within each sphere for each MB sequence and the 476 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 19 corresponding control sequences for the first seed location (plots for other seeds are shown 477 in Supplementary Figure 7; statistics for all seeds are shown in Supplementary Table 3). (C) 478 Mean MVPA dissimilarity within each sphere for each MB sequence and the corresponding 479 control sequences. 480 481 482 4. Discussion 483 The tSNR of fMRI varies across the brain. This is especially evident in ventral anterior 484 temporal and orbitofrontal regions, which are located next to the air-filled sinuses and are 485 therefore affected by signal dropout and geometric distortions (Devlin et al., 2000; Halai et 486 al., 2014, 2015, 2024). This study is the first 7T-fMRI study systematically comparing pTx, ME 487 and MB as methods for improving sensitivity in these regions while maintaining sensitivity 488 across the brain. We found that pTx improved activation magnitude in posterior temporal 489 and occipital regions. ME, however, resulted in improved activation magnitude extending 490 down the temporal lobe and including inferior frontal regions. Both MB and ME-ICA 491 denoising resulted in improved activation precision in the same areas. In an exploratory 492 analysis we found that ME and ME-ICA improved MVPA performance but MB did not. No 493 slice leakage artefacts were associated with our multiband sequences. 494 Although parallel transmit produced better activation magnitude in posterior 495 temporal and occipital regions than the baseline (SESB) sequence without compromising 496 activation magnitude elsewhere, activation magnitude in anterior temporal areas was 497 comparable for both sequences. These results replicated those of Ding et al. (2022) who 498 failed to find improved task contrast in anterior temporal regions with pTx in spite of 499 improved tSNR in the resting state. Note that, despite visible signal dropout on the EPIs, 500 even the baseline sequence was able to identify semantic activity extending into ventral 501 anterior temporal regions (Figure 6). It is possible that this finding reflects the increased 502 sensitivity of 7T-fMRI compared to 3T-fMRI, where semantic activity is rarely observed 503 without using a method such as multi-echo (Halai et al., 2014, 2015, 2024) or spin-echo 504 (Binney et al., 2010; Embleton et al., 2010). That a standard sequence can detect signal in 505 these regions might come as a surprise to many neuroimaging researchers as 7T-fMRI is not 506 only associated with increased sensitivity, but also with exacerbated magnetic susceptibility 507 artefacts. This finding should therefore embolden researchers to use 7T fMRI for 508 experimental designs that are difficult to conduct at a lower field strength - for example, 509 studies of special populations, who may benefit from shorter scan times (Cope et al., 2023), 510 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 20 or studies requiring high spatial specificity (Marques & Norris, 2018). We do acknowledge 511 that the way we have utilised 7T fMRI in this study will not be suitable for, for example, 512 studies that require ultra-high resolution (in our case, scanner hardware constrained voxel 513 size when combined with a short first echo time). 514 515 516 517 Figure 6: Contrast despite signal dropout and geometric distortions. (A) Mean EPI across all 518 subjects for the baseline sequence (SESB) registered to MNI152NLin2009cAsym space. (B) 519 Selected regions of interest overlaid on the mean EPI image. (C) Main effect of activation 520 magnitude for the contrast of interest (S > C) for the baseline sequence. (D) Main effect of 521 activation magnitude for the contrast of interest (S > C) for the MEMB sequence. Results are 522 cluster-corrected at p < 0.05 based on an uncorrected voxel threshold of p < 0.001. 523 524 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 21 The factorial design enabled us to disentangle the effects of ME and MB. As 525 hypothesised, ME improved activation magnitude and these effects were localised to 526 inferior temporal and orbitofrontal areas. Adverse effects on activation magnitude were 527 negligible - although ME reduced activation precision in two very small clusters 528 (Supplementary Figure 4), these were not within areas that the semantic task is known to 529 recruit (Jung et al., 2017). These results demonstrated that having one very short echo is a 530 successful method of increasing activation magnitude in areas where T2* is particularly short 531 (Halai et al., 2014, 2015, 2024; Jung et al., 2017). ME also opens up the opportunity for 532 sophisticated denoising such as tedana (DuPre et al., 2021; Kundu et al., 2011, 2013; The 533 tedana Community et al., 2023), which improved activation precision in agreement with 534 previous findings (Amemiya et al., 2019; Gonzalez-Castillo et al., 2016). Small clusters in 535 which ME-ICA increased activation magnitude (Supplementary Figure 5) or decreased 536 activation precision (Supplementary Figure 6) were not within regions recruited by the task 537 (Jung et al., 2017). 538 As hypothesised, multiband improved activation precision in areas of inhomogeneity 539 and in agreement with other work (Puckett et al., 2018). Supplementary Figure 4 shows two 540 very small clusters in which multiband reduced activation precision, but which lie outside 541 regions recruited by the task (Jung et al., 2017). After downsampling the MB timeseries to 542 match the number of volumes in the SB timeseries, there was no longer any difference 543 between the SB and MB sequences; this suggests that it is the increase in the number of 544 volumes that accounts for the benefits of multiband. Additionally, there was no evidence for 545 signal leakage into the simultaneously-acquired slice location (McNabb et al., 2020; Todd et 546 al., 2016), at least for our modest multiband factor of 2. We note that MB and ME seem to 547 have independent effects as the interaction was not significant within any ROI. However, 548 ME seems to offer larger activation magnitude (Figure 6) and MB offers better activation 549 precision. MB can help reduce the longer TR associated with ME and may decrease noise 550 aliasing in studies using short TRs (~1s). 551 Although this study focused primarily on univariate effects, we also identified 552 improvements to multivariate metrics that will be of interest to researchers using 553 increasingly-popular MVPA tools (Frisby et al., 2023). ME was found to improve MVPA 554 performance compared to SE sequences and, in turn, MEdn sequences performed 555 significantly better than ME sequences without denoising in every region of interest (with 556 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 22 no detrimental effects). Future studies employing multivariate methods should strongly 557 consider using both ME acquisitions and ME-ICA preprocessing. 558 This study provides evidence supporting use of ME and/or MB sequences to detect 559 signal in regions prone to susceptibility artefacts. However, it is important to note that 560 individual 7T MRI scanners have their own hardware and software limitations; this means 561 that each site must optimise and test feasible parameters locally. For example, our system 562 did not allow us to take advantage of the higher spatial resolution offered by 7T-fMRI while 563 retaining an adequately short first echo for our multi-echo sequences. Puckett et al. (2018) 564 used comparable voxel sizes but had a shorter first TE than us (9.9 ms vs our 11.8 ms); 565 Miletić et al. (2020) had higher resolution (1.6 mm vs our 2.5 mm isotropic) and a shorter 566 first echo of 9.66 ms. Additionally, our 7T Terra pTx system imposes very conservative SAR 567 limits (1W per channel, 8W total) for pTx scans but permits 20W total (i.e. 2.5W per 568 channel) for circularly polarised mode (CP-mode or “TrueForm”) scans (equivalent to single 569 transmit). We therefore had to use VERSE-modified pTx pulses which are known to be more 570 sensitive to B0 inhomogeneity. We also acknowledge that pTx can be combined with multi-571 echo or multiband (Ding et al., 2023; Wu et al., 2016). Although we used the pTx head coil 572 for all scans, the combination of sequences was not implemented on our scanner at the 573 time of running this study and therefore needs to be tested empirically in future. Thus, 574 rather than identifying the limits of what is possible, we offer an accessible framework that 575 will enable researchers to leverage 7T-fMRI for investigating regions affected by 576 susceptibility artefacts despite the hardware and software constraints of individual 577 scanners. 578 579 5. Conclusion 580 In this study we compared pTx, ME and MB as methods of improving sensitivity in 581 temporal and frontal areas prone to magnetic susceptibility artefacts. Both pTx and ME 582 improved activation magnitude, but only ME showed improvements in artefact-prone 583 regions. MB and ME-ICA improved activation precision in these areas. Exploratory results 584 suggested that both ME and ME-ICA may also benefit MVPA. We demonstrated that a multi-585 echo, multiband sequence running on the 8Tx32Rx pTx head coil (in CP mode) can detect 586 signal in susceptible regions while maintaining sensitivity across the whole brain and is 587 therefore a versatile choice for future studies using high field strength and investigating the 588 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 23 functional roles of ventral temporal and/or orbitofrontal cortex (Binney et al., 2016; 589 Borghesani et al., 2016; Devereux et al., 2018; Duncan, 2010; DuPre et al., 2016; Fernandez 590 et al., 2017; Lambon Ralph et al., 2017; Zahn et al., 2007). 591 592 Data and code availability: Data will be made publicly available upon peer review and 593 acceptance. Code is publicly available at: https://github.com/slfrisby/7TOptimisation/. 594 595 CRediT statement: Saskia L. Frisby: Conceptualisation, Methodology, Investigation, Formal 596 Analysis, Writing – Original Draft, Writing – Review & Editing, Visualisation; Marta M. 597 Correia: Conceptualisation, Methodology, Writing – Review & Editing, Funding Acquisition; 598 Minghao Zhang: Methodology, Writing – Review & Editing; Christopher T. Rodgers: 599 Methodology, Writing – Review & Editing, Supervision; Timothy T. Rogers: Writing – Review 600 & Editing, Supervision; Matthew A. Lambon Ralph: Writing – Review & Editing, Supervision, 601 Funding Acquisition; Ajay D. Halai: Conceptualisation, Methodology, Formal Analysis, 602 Writing – Review & Editing, Supervision, Funding Acquisition. 603 604 Funding: This work was supported by an MRC Unit grant (SUAG/019 G116768) to M.M.C., 605 an MRC PhD studentship (MR N013433-1) to M.Z., an MRC programme grant 606 (MR/R023883/1) and intramural funding (MC_UU_00005/18) to M.A.L.R, and an MRC 607 Career Development Award (MR/V031481/1) to A.D.H. This work was also supported by the 608 NIHR Cambridge Biomedical Research Centre (NIHR203312) and an MRC Clinical Research 609 Infrastructure Award for 7T (MR/M008983/1). The views expressed are those of the authors 610 and not necessarily those of the NIHR or the Department of Health and Social Care. 611 612 Competing interests: C.T.R. receives research support from Siemens. 613 614

We thank the participants and the Wolfson Brain Imaging Centre 615 radiographers for their assistance. 616 617 618

619 620 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 24 Amemiya, S., Yamashita, H., Takao, H., & Abe, O. (2019). Integrated multi-echo denoising 621 strategy improves identification of inherent language laterality. Magnetic Resonance 622 in Medicine, 81(5), 3262–3271. https://doi.org/10.1002/mrm.27620 623 Andersson, J. L. R., Skare, S., & Ashburner, J. (2003). How to correct susceptibility distortions 624 in spin-echo echo-planar images: Application to diffusion tensor imaging. 625 NeuroImage, 20(2), 870–888. https://doi.org/10.1016/S1053-8119(03)00336-7 626 Avants, B. B., Tustison, N. J., Song, G., Cook, P. A., Klein, A., & Gee, J. C. (2011). A 627 reproducible evaluation of ANTs similarity metric performance in brain image 628 registration. NeuroImage, 54(3), 2033–2044. 629 https://doi.org/10.1016/j.neuroimage.2010.09.025 630 Avants, B. B., Tustison, N., & Johnson, H. (2009). Advanced Normalization Tools (ANTS). 631 Barth, M., Breuer, F., Koopmans, P. J., Norris, D. G., & Poser, B. A. (2016). Simultaneous 632 multislice (SMS) imaging techniques. Magnetic Resonance in Medicine, 75(1), 63–81. 633 https://doi.org/10.1002/mrm.25897 634 Bhavsar, S., Zvyagintsev, M., & Mathiak, K. (2014). BOLD sensitivity and SNR characteristics 635 of parallel imaging-accelerated single-shot multi-echo EPI for fMRI. NeuroImage, 84, 636 65–75. https://doi.org/10.1016/j.neuroimage.2013.08.007 637 Binney, R. J., Embleton, K. V., Jefferies, E., Parker, G. J. M., & Lambon Ralph, M. A. (2010). 638 The ventral and inferolateral aspects of the anterior temporal lobe are crucial in 639 semantic memory: Evidence from a novel direct comparison of distortion-corrected 640 fMRI, rTMS, and semantic dementia. Cerebral Cortex, 20(11), 2728–2738. 641 https://doi.org/10.1093/cercor/bhq019 642 Binney, R. J., Hoffman, P., & Lambon Ralph, M. A. (2016). Mapping the Multiple Graded 643 Contributions of the Anterior Temporal Lobe Representational Hub to Abstract and 644 Social Concepts: Evidence from Distortion-corrected fMRI. Cerebral Cortex, 26(11), 645 4227–4241. https://doi.org/10.1093/cercor/bhw260 646 Borghesani, V., Pedregosa, F., Buiatti, M., Amadon, A., Eger, E., & Piazza, M. (2016). Word 647 meaning in the ventral visual path: A perceptual to conceptual gradient of semantic 648 coding. NeuroImage, 143, 128–140. 649 https://doi.org/10.1016/j.neuroimage.2016.08.068 650 Boyacioğlu, R., Schulz, J., Koopmans, P. J., Barth, M., & Norris, D. G. (2015). Improved 651 sensitivity and specificity for resting state and task fMRI with multiband multi-echo 652 EPI compared to multi-echo EPI at 7 T. NeuroImage, 119, 352–361. 653 https://doi.org/10.1016/j.neuroimage.2015.06.089 654 Bozeat, S., Lambon Ralph, M. A., Patterson, K., Garrard, P., & Hodges, J. R. (2000). Non-655 verbal semantic impairment in semantic dementia. Neuropsychologia, 38(9), 1207–656 1215. https://doi.org/10.1016/S0028-3932(00)00034-8 657 Caballero-Gaudes, C., Moia, S., Panwar, P., Bandettini, P. A., & Gonzalez-Castillo, J. (2019). A 658 deconvolution algorithm for multi-echo functional MRI: Multi-echo Sparse Paradigm 659 Free Mapping. NeuroImage, 202, 116081. 660 https://doi.org/10.1016/j.neuroimage.2019.116081 661 Cohen, A. D., Nencka, A. S., Lebel, R. M., & Wang, Y. (2017). Multiband multi-echo imaging 662 of simultaneous oxygenation and flow timeseries for resting state connectivity. PLOS 663 ONE, 12(3), e0169253. https://doi.org/10.1371/journal.pone.0169253 664 Cohen, A. D., Nencka, A. S., & Wang, Y. (2018). Multiband multi-echo simultaneous 665 ASL/BOLD for task-induced functional MRI. PLOS ONE, 13(2), e0190427. 666 https://doi.org/10.1371/journal.pone.0190427 667 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 25 Cope, T. E., Sohoglu, E., Peterson, K. A., Jones, P. S., Rua, C., Passamonti, L., Sedley, W., Post, 668 B., Coebergh, J., Butler, C. R., Garrard, P., Abdel-Aziz, K., Husain, M., Griffiths, T. D., 669 Patterson, K., Davis, M. H., & Rowe, J. B. (2023). Temporal lobe perceptual 670 predictions for speech are instantiated in motor cortex and reconciled by inferior 671 frontal cortex. Cell Reports, 42(5). https://doi.org/10.1016/j.celrep.2023.112422 672 Coutanche, M. N. (2013). Distinguishing multi-voxel patterns and mean activation: Why, 673 how, and what does it tell us? Cognitive, Affective, & Behavioral Neuroscience, 13(3), 674 667–673. https://doi.org/10.3758/s13415-013-0186-2 675 Cox, R. W. (1996). AFNI: Software for analysis and visualization of functional magnetic 676 resonance neuroimages. Computers and Biomedical Research, 29(3), 162–173. 677 https://doi.org/10.1006/cbmr.1996.0014 678 Cox, R. W., & Hyde, J. S. (1997). Software tools for analysis and visualization of fMRI data. 679 NMR in Biomedicine, 10(4–5), 171–178. https://doi.org/10.1002/(SICI)1099-680 1492(199706/08)10:4/53.0.CO;2-L 681 Davis, T., LaRocque, K. F., Mumford, J. A., Norman, K. A., Wagner, A. D., & Poldrack, R. A. 682 (2014). What do differences between multi-voxel and univariate analysis mean? How 683 subject-, voxel-, and trial-level variance impact fMRI analysis. NeuroImage, 97, 271–684 283. https://doi.org/10.1016/j.neuroimage.2014.04.037 685 Davis, T., & Poldrack, R. A. (2013). Measuring neural representations with fMRI: Practices 686 and pitfalls. Annals of the New York Academy of Sciences, 1296(1), 108–134. 687 https://doi.org/10.1111/nyas.12156 688 Demetriou, L., Kowalczyk, O. S., Tyson, G., Bello, T., Newbould, R. D., & Wall, M. B. (2018). A 689 comprehensive evaluation of increasing temporal resolution with multiband-690 accelerated protocols and effects on statistical outcome measures in fMRI. 691 NeuroImage, 176, 404–416. https://doi.org/10.1016/j.neuroimage.2018.05.011 692 Deniz, F., Nunez-Elizalde, A. O., Huth, A. G., & Gallant, J. L. (2019). The Representation of 693 Semantic Information Across Human Cerebral Cortex During Listening Versus 694 Reading Is Invariant to Stimulus Modality. The Journal of Neuroscience, 39(39), 7722–695 7736. https://doi.org/10.1523/JNEUROSCI.0675-19.2019 696 Devereux, B. J., Clarke, A., & Tyler, L. K. (2018). Integrated deep visual and semantic 697 attractor neural networks predict fMRI pattern-information along the ventral object 698 processing pathway. Scientific Reports, 8(1), 10636. https://doi.org/10.1038/s41598-699 018-28865-1 700 Devlin, J. T., Russell, R. P., Davis, M. H., Price, C. J., Wilson, J., Moss, H. E., Matthews, P. M., & 701 Tyler, L. K. (2000). Susceptibility-induced loss of signal: Comparing PET and fMRI on a 702 semantic task. NeuroImage, 11(6), 589–600. 703 https://doi.org/10.1006/nimg.2000.0595 704 Ding, B., Dragonu, I., Rua, C., Carlin, J. D., Halai, A. D., Liebig, P., Heidemann, R., Correia, M. 705 M., & Rodgers, C. T. (2022). Parallel transmit (pTx) with online pulse design for task-706 based fMRI at 7T. Magnetic Resonance Imaging, 93, 163–174. 707 https://doi.org/10.1016/j.mri.2022.07.003 708 Ding, B., Williams, S. N., Zhang, M., Herrler, J., Liebig, P., Dragonu, I., Neji, R., Rodgers, C. T., 709 & Porter, D. A. (2023). Optimisation of pulse-specific phase schedule reduces peak RF 710 amplitude in multiband parallel-transmit pulses. ISMRM & ISMRT Annual Meeting & 711 Exhibition, Toronto, Canada. 712 Dipasquale, O., Sethi, A., Laganà, M. M., Baglio, F., Baselli, G., Kundu, P., Harrison, N. A., & 713 Cercignani, M. (2017). Comparing resting state fMRI de-noising approaches using 714 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 26 multi- and single-echo acquisitions. PLOS ONE, 12(3), e0173289. 715 https://doi.org/10.1371/journal.pone.0173289 716 Duncan, J. (2010). The multiple-demand (MD) system of the primate brain: Mental programs 717 for intelligent behaviour. Trends in Cognitive Sciences, 14(4), 172–179. 718 https://doi.org/10.1016/j.tics.2010.01.004 719 DuPre, E., Luh, W.-M., & Spreng, R. N. (2016). Multi-echo fMRI replication sample of 720 autobiographical memory, prospection and theory of mind reasoning tasks. Scientific 721 Data, 3(1), 160116. https://doi.org/10.1038/sdata.2016.116 722 DuPre, E., Salo, T., Ahmed, Z., Bandettini, P. A., Bottenhorn, K. L., Caballero-Gaudes, C., 723 Dowdle, L. T., Gonzalez-Castillo, J., Heunis, S., Kundu, P., Laird, A. R., Markello, R., 724 Markiewicz, C. J., Moia, S., Staden, I., Teves, J. B., Uruñuela, E., Vaziri-Pashkam, M., 725 Whitaker, K., & Handwerker, D. A. (2021). TE-dependent analysis of multi-echo fMRI 726 with tedana. Journal of Open Source Software, 6(66), 3669. 727 https://doi.org/10.21105/joss.03669 728 Embleton, K. V., Haroon, H. A., Morris, D. M., Lambon Ralph, M. A. L., & Parker, G. J. M. 729 (2010). Distortion correction for diffusion-weighted MRI tractography and fMRI in 730 the temporal lobes. Human Brain Mapping, 31(10), 1570–1587. 731 https://doi.org/10.1002/hbm.20959 732 Esteban, O., Markiewicz, C. J., Blair, R. W., Moodie, C. A., Isik, A. I., Erramuzpe, A., Kent, J. D., 733 Goncalves, M., DuPre, E., Snyder, M., Oya, H., Ghosh, S. S., Wright, J., Durnez, J., 734 Poldrack, R. A., & Gorgolewski, K. J. (2019). fMRIPrep: A robust preprocessing 735 pipeline for functional MRI. Nature Methods, 16(1), 111–116. 736 https://doi.org/10.1038/s41592-018-0235-4 737 Evans, J. W., Kundu, P., Horovitz, S. G., & Bandettini, P. A. (2015). Separating slow BOLD 738 from non-BOLD baseline drifts using multi-echo fMRI. NeuroImage, 105, 189–197. 739 https://doi.org/10.1016/j.neuroimage.2014.10.051 740 Feinberg, D. A., Moeller, S., Smith, S. M., Auerbach, E., Ramanna, S., Glasser, M. F., Miller, K. 741 L., Ugurbil, K., & Yacoub, E. (2010). Multiplexed Echo Planar Imaging for Sub-Second 742 Whole Brain FMRI and Fast Diffusion Imaging. PLOS ONE, 5(12), e15710. 743 https://doi.org/10.1371/journal.pone.0015710 744 Fernandez, B., Leuchs, L., Sämann, P. G., Czisch, M., & Spoormaker, V. I. (2017). Multi-echo 745 EPI of human fear conditioning reveals improved BOLD detection in ventromedial 746 prefrontal cortex. NeuroImage, 156, 65–77. 747 https://doi.org/10.1016/j.neuroimage.2017.05.005 748 Frisby, S. L., Halai, A. D., Cox, C. R., Lambon Ralph, M. A., & Rogers, T. T. (2023). Decoding 749 semantic representations in mind and brain. Trends in Cognitive Sciences, 27(3), 750 258–281. https://doi.org/10.1016/j.tics.2022.12.006 751 Gaser, C., Dahnke, R., Thompson, P. M., Kurth, F., Luders, E., & Alzheimer’s Disease 752 Neuroimaging Initiative. (2023). CAT – A computational anatomy toolbox for the 753 analysis of structural MRI data (p. 2022.06.11.495736). bioRxiv. 754 https://doi.org/10.1101/2022.06.11.495736 755 Gilmore, A. W., Agron, A. M., González-Araya, E. I., Gotts, S. J., & Martin, A. (2022). A 756 comparison of single- and multi-echo processing of functional MRI data during overt 757 autobiographical recall. Frontiers in Neuroscience, 16, 854387. 758 https://doi.org/10.3389/fnins.2022.854387 759 Gonzalez-Castillo, J., Panwar, P., Buchanan, L. C., Caballero-Gaudes, C., Handwerker, D. A., 760 Jangraw, D. C., Zachariou, V., Inati, S., Roopchansingh, V., Derbyshire, J. A., & 761 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 27 Bandettini, P. A. (2016). Evaluation of multi-echo ICA denoising for task based fMRI 762 studies: Block designs, rapid event-related designs, and cardiac-gated fMRI. 763 NeuroImage, 141, 452–468. https://doi.org/10.1016/j.neuroimage.2016.07.049 764 Gorgolewski, K. J., Auer, T., Calhoun, V. D., Craddock, R. C., Das, S., Duff, E. P., Flandin, G., 765 Ghosh, S. S., Glatard, T., Halchenko, Y. O., Handwerker, D. A., Hanke, M., Keator, D., 766 Li, X., Michael, Z., Maumet, C., Nichols, B. N., Nichols, T. E., Pellman, J., … Poldrack, R. 767 A. (2016). The brain imaging data structure, a format for organizing and describing 768 outputs of neuroimaging experiments. Scientific Data, 3(1), 160044. 769 https://doi.org/10.1038/sdata.2016.44 770 Gorgolewski, K. J., Burns, C. D., Madison, C., Clark, D., Halchenko, Y. O., Waskom, M. L., & 771 Ghosh, S. S. (2011). Nipype: A flexible, lightweight and extensible neuroimaging data 772 processing framework in python. Frontiers in Neuroinformatics, 5. 773 https://doi.org/10.3389/fninf.2011.00013 774 Griffanti, L., Salimi-Khorshidi, G., Beckmann, C. F., Auerbach, E. J., Douaud, G., Sexton, C. E., 775 Zsoldos, E., Ebmeier, K. P., Filippini, N., Mackay, C. E., Moeller, S., Xu, J., Yacoub, E., 776 Baselli, G., Ugurbil, K., Miller, K. L., & Smith, S. M. (2014). ICA-based artefact removal 777 and accelerated fMRI acquisition for improved resting state network imaging. 778 NeuroImage, 95, 232–247. https://doi.org/10.1016/j.neuroimage.2014.03.034 779 Hagberg, G. E., Indovina, I., Sanes, J. N., & Posse, S. (2002). Real-time quantification of T2* 780 changes using multiecho planar imaging and numerical methods. Magnetic 781 Resonance in Medicine, 48(5), 877–882. https://doi.org/10.1002/mrm.10283 782 Halai, A. D., Henson, R. N., Finoia, P., & Correia, M. M. (2024). Comparing the effect of multi 783 gradient echo and multi band fMRI during a semantic task (p. 2024.03.20.585909). 784 bioRxiv. https://doi.org/10.1101/2024.03.20.585909 785 Halai, A. D., Parkes, L. M., & Welbourne, S. R. (2015). Dual-echo fMRI can detect activations 786 in inferior temporal lobe during intelligible speech comprehension. NeuroImage, 787 122, 214–221. https://doi.org/10.1016/j.neuroimage.2015.05.067 788 Halai, A. D., Welbourne, S. R., Embleton, K., & Parkes, L. M. (2014). A comparison of dual 789 gradient-echo and spin-echo fMRI of the inferior temporal lobe. Human Brain 790 Mapping, 35(8), 4118–4128. https://doi.org/10.1002/hbm.22463 791 Halchenko, Y. O., Goncalves, M., Ghosh, S., Velasco, P., Castello, M. V. di O., Salo, T., 792 Wodder, J. T., Hanke, M., Sadil, P., Gorgolewski, K. J., Ioanas, H.-I., Rorden, C., 793 Hendrickson, T. J., Dayan, M., Houlihan, S. D., Kent, J., Strauss, T., Lee, J., To, I., … 794 Kennedy, D. N. (2024). HeuDiConv—Flexible DICOM conversion into structured 795 directory layouts. Journal of Open Source Software, 9(99), 5839. 796 https://doi.org/10.21105/joss.05839 797 Hargreaves, B. A., Cunningham, C. H., Nishimura, D. G., & Conolly, S. M. (2004). Variable-rate 798 selective excitation for rapid MRI sequences. Magnetic Resonance in Medicine, 52(3), 799 590–597. https://doi.org/10.1002/mrm.20168 800 Haxby, J. V., Gobbini, M. I., Furey, M. L., Ishai, A., Schouten, J. L., & Pietrini, P. (2001). 801 Distributed and overlapping representations of faces and objects in ventral temporal 802 cortex. Science, 293(5539), 2425–2430. https://doi.org/10.1126/science.1063736 803 Howard, D., & Patterson, K. E. (1992). The pyramids and palm trees test. 804 Https://Eprints.Ncl.Ac.Uk. https://eprints.ncl.ac.uk 805 Humphreys, G. F., Hoffman, P., Visser, M., Binney, R. J., & Lambon Ralph, M. A. (2015). 806 Establishing task- and modality-dependent dissociations between the semantic and 807 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 28 default mode networks. Proceedings of the National Academy of Sciences, 112(25), 808 7857–7862. https://doi.org/10.1073/pnas.1422760112 809 Jenkinson, M., Beckmann, C. F., Behrens, T. E. J., Woolrich, M. W., & Smith, S. M. (2012). FSL. 810 NeuroImage, 62(2), 782–790. https://doi.org/10.1016/j.neuroimage.2011.09.015 811 Jung, J., Williams, S. R., Sanaei Nezhad, F., & Lambon Ralph, M. A. (2017). GABA 812 concentrations in the anterior temporal lobe predict human semantic processing. 813 Scientific Reports, 7(1), 15748. https://doi.org/10.1038/s41598-017-15981-7 814 Kirilina, E., Lutti, A., Poser, B. A., Blankenburg, F., & Weiskopf, N. (2016). The quest for the 815 best: The impact of different EPI sequences on the sensitivity of random effect fMRI 816 group analyses. NeuroImage, 126, 49–59. 817 https://doi.org/10.1016/j.neuroimage.2015.10.071 818 Koopmans, P. J., Barth, M., Orzada, S., & Norris, D. G. (2011). Multi-echo fMRI of the cortical 819 laminae in humans at 7T. NeuroImage, 56(3), 1276–1285. 820 https://doi.org/10.1016/j.neuroimage.2011.02.042 821 Kovářová, A., Gajdoš, M., Rektor, I., & Mikl, M. (2022). Contribution of the multi-echo 822 approach in accelerated functional magnetic resonance imaging multiband 823 acquisition. Human Brain Mapping, 43(3), 955–973. 824 https://doi.org/10.1002/hbm.25698 825 Kundu, P., Benson, B. E., Baldwin, K. L., Rosen, D., Luh, W.-M., Bandettini, P. A., Pine, D. S., & 826 Ernst, M. (2015). Robust resting state fMRI processing for studies on typical brain 827 development based on multi-echo EPI acquisition. Brain Imaging and Behavior, 9(1), 828 56–73. https://doi.org/10.1007/s11682-014-9346-4 829 Kundu, P., Brenowitz, N. D., Voon, V., Worbe, Y., Vértes, P. E., Inati, S. J., Saad, Z. S., 830 Bandettini, P. A., & Bullmore, E. T. (2013). Integrated strategy for improving 831 functional connectivity mapping using multiecho fMRI. Proceedings of the National 832 Academy of Sciences, 110(40), 16187–16192. 833 https://doi.org/10.1073/pnas.1301725110 834 Kundu, P., Inati, S. J., Evans, J. W., Luh, W.-M., & Bandettini, P. A. (2011). Differentiating 835 BOLD and non-BOLD signals in fMRI time series using multi-echo EPI. NeuroImage, 836 60(3), 1759–1770. https://doi.org/10.1016/j.neuroimage.2011.12.028 837 Kundu, P., Voon, V., Balchandani, P., Lombardo, M. V., Poser, B. A., & Bandettini, P. A. 838 (2017). Multi-echo fMRI: A review of applications in fMRI denoising and analysis of 839 BOLD signals. NeuroImage, 154, 59–80. 840 https://doi.org/10.1016/j.neuroimage.2017.03.033 841 Lambon Ralph, M. A., Jefferies, E., Patterson, K., & Rogers, T. T. (2017). The neural and 842 computational bases of semantic cognition. Nature Reviews Neuroscience, 18(1), 42–843 55. https://doi.org/10.1038/nrn.2016.150 844 Lombardo, M. V., Auyeung, B., Holt, R. J., Waldman, J., Ruigrok, A. N. V., Mooney, N., 845 Bullmore, E. T., Baron-Cohen, S., & Kundu, P. (2016). Improving effect size estimation 846 and statistical power with multi-echo fMRI and its impact on understanding the 847 neural systems supporting mentalizing. NeuroImage, 142, 55–66. 848 https://doi.org/10.1016/j.neuroimage.2016.07.022 849 Lynch, C. J., Power, J. D., Scult, M. A., Dubin, M., Gunning, F. M., & Liston, C. (2020). Rapid 850 Precision Functional Mapping of Individuals Using Multi-Echo fMRI. Cell Reports, 851 33(12), 108540. https://doi.org/10.1016/j.celrep.2020.108540 852 Markiewicz, C. J., Esteban, O., Goncalves, M., Provins, C., Salo, T., Kent, J. D., DuPre, E., Ciric, 853 R., Pinsard, B., Blair, R. W., Poldrack, R. A., & Gorgolewski, K. J. (2024). fMRIPrep: A 854 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 29 robust preprocessing pipeline for functional MRI (Version 24.0.1) [Computer 855 software]. Zenodo. https://doi.org/10.5281/zenodo.12774221 856 Marques, J. P., & Norris, D. G. (2018). How to choose the right MR sequence for your 857 research question at 7 T and above? NeuroImage, 168, 119–140. 858 https://doi.org/10.1016/j.neuroimage.2017.04.044 859 McNabb, C. B., Lindner, M., Shen, S., Burgess, L. G., Murayama, K., & Johnstone, T. (2020). 860 Inter-slice leakage and intra-slice aliasing in simultaneous multi-slice echo-planar 861 images. Brain Structure and Function, 225(3), 1153–1158. 862 https://doi.org/10.1007/s00429-020-02053-2 863 Miletić, S., Bazin, P.-L., Weiskopf, N., van der Zwaag, W., Forstmann, B. U., & Trampel, R. 864 (2020). FMRI protocol optimization for simultaneously studying small subcortical and 865 cortical areas at 7 T. NeuroImage, 219, 116992. 866 https://doi.org/10.1016/j.neuroimage.2020.116992 867 Moeller, S., Yacoub, E., Olman, C. A., Auerbach, E., Strupp, J., Harel, N., & Uğurbil, K. (2010). 868 Multiband multislice GE-EPI at 7 Tesla, with 16-fold acceleration using partial parallel 869 imaging with application to high spatial and temporal whole-brain fMRI. Magnetic 870 Resonance in Medicine, 63(5), 1144–1153. https://doi.org/10.1002/mrm.22361 871 Morris, L. S., Kundu, P., Costi, S., Collins, A., Schneider, M., Verma, G., Balchandani, P., & 872 Murrough, J. W. (2019). Ultra-high field MRI reveals mood-related circuit 873 disturbances in depression: A comparison between 3-Tesla and 7-Tesla. Translational 874 Psychiatry, 9(1), 94. https://doi.org/10.1038/s41398-019-0425-6 875 O’Brien, K. R., Kober, T., Hagmann, P., Maeder, P., Marques, J., Lazeyras, F., Krueger, G., & 876 Roche, A. (2014). Robust T1-weighted structural brain imaging and morphometry at 877 7T using MP2RAGE. PLOS ONE, 9(6), e99676. 878 https://doi.org/10.1371/journal.pone.0099676 879 Poser, B. A., & Norris, D. G. (2009). Investigating the benefits of multi-echo EPI for fMRI at 7 880 T. NeuroImage, 45(4), 1162–1172. 881 https://doi.org/10.1016/j.neuroimage.2009.01.007 882 Poser, B. A., Versluis, M. J., Hoogduin, J. M., & Norris, D. G. (2006). BOLD contrast sensitivity 883 enhancement and artifact reduction with multiecho EPI: Parallel-acquired 884 inhomogeneity-desensitized fMRI. Magnetic Resonance in Medicine, 55(6), 1227–885 1235. https://doi.org/10.1002/mrm.20900 886 Posse, S. (2012). Multi-echo acquisition. NeuroImage, 62(2), 665–671. 887 https://doi.org/10.1016/j.neuroimage.2011.10.057 888 Puckett, A. M., Bollmann, S., Poser, B. A., Palmer, J., Barth, M., & Cunnington, R. (2018). 889 Using multi-echo simultaneous multi-slice (SMS) EPI to improve functional MRI of 890 the subcortical nuclei of the basal ganglia at ultra-high field (7T). NeuroImage, 172, 891 886–895. https://doi.org/10.1016/j.neuroimage.2017.12.005 892 Risk, B. B., Murden, R. J., Wu, J., Nebel, M. B., Venkataraman, A., Zhang, Z., & Qiu, D. (2021). 893 Which multiband factor should you choose for your resting-state fMRI study? 894 NeuroImage, 234, 117965. https://doi.org/10.1016/j.neuroimage.2021.117965 895 Setsompop, K., Gagoski, B. A., Polimeni, J. R., Witzel, T., Wedeen, V. J., & Wald, L. L. (2012). 896 Blipped-controlled aliasing in parallel imaging for simultaneous multislice echo 897 planar imaging with reduced g -factor penalty. Magnetic Resonance in Medicine, 898 67(5), 1210–1224. https://doi.org/10.1002/mrm.23097 899 Smith, S. M. (2002). Fast robust automated brain extraction. Human Brain Mapping, 17(3), 900 143–155. https://doi.org/10.1002/hbm.10062 901 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint 30 Smith, S. M., Beckmann, C. F., Andersson, J., Auerbach, E. J., Bijsterbosch, J., Douaud, G., 902 Duff, E., Feinberg, D. A., Griffanti, L., Harms, M. P., Kelly, M., Laumann, T., Miller, K. 903 L., Moeller, S., Petersen, S., Power, J., Salimi-Khorshidi, G., Snyder, A. Z., Vu, A. T., … 904 Glasser, M. F. (2013). Resting-state fMRI in the Human Connectome Project. 905 NeuroImage, 80, 144–168. https://doi.org/10.1016/j.neuroimage.2013.05.039 906 Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E. J., Johansen-907 Berg, H., Bannister, P. R., De Luca, M., Drobnjak, I., Flitney, D. E., Niazy, R. K., 908 Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J. M., & Matthews, P. M. 909 (2004). Advances in functional and structural MR image analysis and implementation 910 as FSL. NeuroImage, 23, S208–S219. 911 https://doi.org/10.1016/j.neuroimage.2004.07.051 912 The tedana Community, Ahmed, Z., Bandettini, P. A., Bottenhorn, K. L., Caballero-Gaudes, 913 C., Dowdle, L. T., DuPre, E., Gonzalez-Castillo, J., Handwerker, D., Heunis, S., Kundu, 914 P., Laird, A. R., Markello, R., Markiewicz, C. J., Maullin-Sapey, T., Moia, S., Molfese, 915 P., Salo, T., Staden, I., … Whitaker, K. (2023). ME-ICA/tedana: 23.0.1 (Version 23.0.1) 916 [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.7926293 917 Todd, N., Moeller, S., Auerbach, E. J., Yacoub, E., Flandin, G., & Weiskopf, N. (2016). 918 Evaluation of 2D multiband EPI imaging for high-resolution, whole-brain, task-based 919 fMRI studies at 3T: Sensitivity and slice leakage artifacts. NeuroImage, 124, 32–42. 920 https://doi.org/10.1016/j.neuroimage.2015.08.056 921 Wu, X., Schmitter, S., Auerbach, E. J., Uğurbil, K., & Van de Moortele, P.-F. (2016). A 922 generalized slab-wise framework for parallel transmit multiband RF pulse design. 923 Magnetic Resonance in Medicine, 75(4), 1444–1456. 924 https://doi.org/10.1002/mrm.25689 925 Yun, S. D., & Shah, N. J. (2017). Whole-brain high in-plane resolution fMRI using accelerated 926 EPIK for enhanced characterisation of functional areas at 3T. PLOS ONE, 12(9), 927 e0184759. https://doi.org/10.1371/journal.pone.0184759 928 Zahn, R., Moll, J., Krueger, F., Huey, E. D., Garrido, G., & Grafman, J. (2007). Social concepts 929 are represented in the superior anterior temporal cortex. Proceedings of the 930 National Academy of Sciences, 104(15), 6430–6435. 931 https://doi.org/10.1073/pnas.0607061104 932 933 .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted March 18, 2025. ; https://doi.org/10.1101/2025.03.17.643748doi: bioRxiv preprint