2401

Functional connectivity sensitivity to image acceleration and orientation in simultaneous PET/MRI
Alessandro Palombit1,2, Marco Castellaro1,2, Erica Silvestri1,2, Enrico De Vita3, Diego Cecchin2,4, and Alessandra Bertoldo1,2

1Department of Information Engineering, University of Padova, Padova, Italy, 2Padova Neuroscience Center, University of Padova, Padova, Italy, 3Department of Biomedical Engineering, King's College London, London, United Kingdom, 4Department of Diagnostic Medical Sciences, University of Padova, Padova, Italy

Synopsis

Resting state fMRI (rs-fMRI) permits in-vivo characterization of brain’s functional connectivity (FC). Multi-Band accelerated EPI allows to improve the temporal resolution of rs-fMRI data and, potentially, to achieve a better characterization of the brain network correlations. However, the impact of image acceleration and orientation on FC structure has not been quantified. In this work we investigated FC changes related to image acceleration effects in a test/retest rs-fMRI protocol. We found FC differences involving relevant networks, confirmed even by graph analysis of the FC maps. Our findings explore the lower bound of single-subject FC reliability and network-dependent acceleration sensitivity.

Purpose

Resting state fMRI (rs-fMRI) allows in-vivo characterization of brain’s functional connectivity (FC) and its decomposition in resting networks (RSNs). However, many confounds can bias rs-fMRI derived FC estimations. Multi-Band EPI (MB-EPI) sequences can help in this by improving both temporal noise characterization and statistical power but the impact of acceleration on FC underpinning structures have been poorly investigated. In this work, we characterized FC structural changes caused by acceleration and orientation in a test/retest rs-fMRI protocol by varying MB and/or Parallel Imaging (iPAT) factors.

Materials and Methods

Subjects: five healthy participants (25.2 ± 4.3 y) scanned on a Siemens Biograph mMR 3T (12-channel PET-transparent head coil).

Acquisition Protocol: Six rs-fMRI runs (MB-EPI1 R014, 600 volumes each), 3 along opposed phase encoding directions (PEdir), additional parameters in Table 1. Geometrically matched spin-echo pair for distortion correction and T1w-MPRAGE (1 mm isotropic) for co-registration and brain parcellation purposes.

Image Processing: standardly pre-processed rs-fMRI data2 was sampled (FreeSurfer3, Connectome Workbench4) over a surface-based functional parcellation5. Parcel signals were Pearson cross-correlated6 to obtain FC matrices (Figure 1), characterised by node Degree (DEG), Strength (STR), Local Efficiency (LE) and Betweeness Centrality (BC) graph metrics7,8. By definition of Pearson correlation, the temporal covariance of two signals is divided by the product of their temporal standard deviations, thus any experimental setup able to modify temporal variance or covariance could affect FC estimation. To measure changes in signal variances related to noise propagation with increasing accelerations, we calculated temporal Signal to Noise Ratio (tSNR) and Physiological Contribution in Spontaneous Oscillations9 (PICSO) at parcel level. FC values were linearly regressed to tSNR and PICSO (multiplied for the involved areas) to describe noise propagation effects over three rs-fMRI settings, separately adding a categorical covariate to describe slicing orientation effects. A similar regression was carried out over FC-derived graph metrics. Statistical FC differences were investigated with ANOVA-1w test (setting as factor), FDR correction (0.05 level) and paired t-test as post-hocs.

Results

Significant FC differences were found between different settings (Figure 2-A), describing a monotonically decreasing FC trend with the total acceleration. These differences structurally affected sensory-motor (SMM, SMH, AUD), cingulo-opercular (CON), ventral (VAN), default mode (DMN) and subcortical (SUB) networks (Figure 2-B,C). A significant linear association between FC amplitude and both tSNR and PICSO products was respectively found in 75.3%, 67.2% of setting-affected FC edges. However, considering the slicing orientation covariate, this association respectively held only in 17.4% and 24.0% of setting-affected edges. Overall, noise metrics were not associated to FC amplitude when evaluated intra-setting (associated in < 3% of the edges). Observed FC differences were confirmed by FC graph metrics (Table 2). In most of the parcels, FC graph metrics differences were again linearly associated to noise effects along different sequence settings but not intra-setting.

Discussion

Increasing acceleration factors in rs-fMRI FC studies can improve characterization and removal of physiological confounds (motion, cardiac, respiratory, etc.) which could bias FC evaluation enhancing its statistical significance. However, we found it also lowers FC amplitudes of many relevant RSNs (Figure 2). Observed FC differences were mostly related to an enhanced noise propagation level due to increased accelerations over three sequence settings used. We have been able to pinpoint this effect by virtue of the equi-parameterized (other than acceleration) and equi-processed rs-fMRI data, a condition expected to provide same average signal and confound level in all sequence settings. In this condition, tSNR and PICSO are maximally sensitive to temporal variance differences related to noise propagation rather than subject-level effects. Their observed linear association to FC amplitude suggests that acceleration-derived noise increase plays a fundamental role over the observed FC alterations at single-subject level. However, as showed in Figure 5, this relation does not always hold at edge-level as the FC amplitude decrease is not accompanied by a lower tSNR (thus an increased temporal signal variance) from AxR1MB2 to CorR1MB3, though the overall trend is maintained. This inconsistence can be explained in terms of different receive head coil performance along different slicing orientations that alter noise propagation probed by tSNR. Taken together, these results point toward PICSO metric to consistently describe FC structural alterations in single-subject FC studies.

Conclusion

Brain’s FC can be structurally altered at single-subject level by the image acceleration, differentially affecting RSNs connectivity. Noise arguments were able to predict these differences and, even if ineffective to describe inter-subject FC variability, they describe a non trivial interplay between temporal covariance and signal variance over FC estimation.

Acknowledgements

We acknowledge that multi band EPI sequence was made available from University of Minnesota through the C2P Siemens sharing mechanism.

References

1. S. Moeller, E. Yacoub, C. A. Olman, E. Auerbach, J. Strupp, N. Harel, and K. Uğurbil, “Multiband multislice GE-EPI at 7 tesla, with 16-fold acceleration using partial parallel imaging with application to high spatial and temporal whole-brain fMRI,” Magn. Reson. Med., vol. 63, no. 5, pp. 1144–1153, May 2010.

2. Y. Behzadi, K. Restom, J. Liau, and T. T. Liu, “A component based noise correction method (CompCor) for BOLD and perfusion based fMRI,” Neuroimage, vol. 37, no. 1, pp. 90–101, Aug. 2007.

3. B. Fischl, M. I. Sereno, and A. M. Dale, “Cortical Surface-Based Analysis,” Neuroimage, vol. 9, no. 2, pp. 195–207, Feb. 1999.

4. M. F. Glasser, S. N. Sotiropoulos, J. A. Wilson, T. S. Coalson, B. Fischl, J. L. Andersson, J. Xu, S. Jbabdi, M. Webster, J. R. Polimeni, D. C. Van Essen, and M. Jenkinson, “The minimal preprocessing pipelines for the Human Connectome Project,” Neuroimage, vol. 80, pp. 105–124, Oct. 2013.

5. E. M. Gordon, T. O. Laumann, B. Adeyemo, J. F. Huckins, W. M. Kelley, and S. E. Petersen, “Generation and Evaluation of a Cortical Area Parcellation from Resting-State Correlations,” Cereb. Cortex, vol. 26, no. 1, pp. 288–303, Jan. 2016.

6. L. Geerligs, M. Rubinov, Cam-CAN, and R. N. Henson, “State and Trait Components of Functional Connectivity: Individual Differences Vary with Mental State,” J. Neurosci., vol. 35, no. 41, pp. 13949–13961, Oct. 2015.

7. M. Rubinov and O. Sporns, “Complex network measures of brain connectivity: Uses and interpretations,” Neuroimage, vol. 52, no. 3, pp. 1059–1069, Sep. 2010.

8. E. Bullmore and O. Sporns, “Complex brain networks: graph theoretical analysis of structural and functional systems,” Nat. Rev. Neurosci., vol. 10, no. 4, pp. 312–312, 2009.

9. A.-L. Hsu, K.-H. Chou, Y.-P. Chao, H.-Y. Fan, C. W. Wu, and J.-H. Chen, “Physiological Contribution in Spontaneous Oscillations: An Approximate Quality-Assurance Index for Resting-State fMRI Signals,” PLoS One, vol. 11, no. 2, p. e0148393, Feb. 2016.

Figures

Figure 1. Normalized and z-Fisher transformed FC maps for each rs-fMRI run (AxR1MB2, CorR1MB3, CorR2MB3 grouped by color). Same FC intensity range (r=-0.2 to 0.7) has been used to scale all 12 (runs) x 5 (subjects) maps. Slicing factor (AXIAL/CORONAL) report the prescribed slicing direction, MB the out-of-plane acceleration factor, iPAT the in-plane acceleration, PEdir the phase encoding direction, scan the specific day-to-day session for each rs-fMRI run.

Table 1. Main sequence parameters for the three defined rs-fMRI settings, each repeated along opposed PEdir and consecutive days obtaining 12 runs per subject. Common factors to all settings were: flip angle (set to Ernst angle), in-plane acquisition matrix of 68x68, 3 mm voxel size and TE=30 ms. The usage of different slicing orientation was here employed to overcome acceleration limitations posed by the receive head coil cylindrical element geometry which hinder z-axis MB acceleration factors to 2 with adequate image quality.

Figure 2. (A) Map of significantly different FC edges along setting factor (reported as 1-pvalue; blue for not significant, yellow for significant) where color-coded and referenced by acronym are the RSNs grouping. (B) Graphical representation (sagittal on left, axial on right) of network nodes (333 nodes, green unaffected, red affected) and significantly different edges (red). For visualization purposes a more aggressive FDR correction was applied (10-5 level) which preserves the pattern of setting-affected connections. (C) Spatial representation over the cortical surface (left) and subcortical slices (right) of the fraction of setting-affected edges concerning each node [0-100%].

Table 2. For each FC metric: Affected nodes report the fraction of nodes (over the total) significantly affected by setting (Kruskal-Wallis test at node level, FDR at 0.05 level), subsequently specifying involved networks in Involved RSNs (affected nodes relative to network size in brackets); tSNR and PICSO interaction report the fraction of nodes (over the total or the significantly associated ones to the setting in brackets) where the FC metric is significantly associated (FDR, 0.05 level) to the noise metric specified by the column header.

Figure 3. FC amplitude differences (upper row) and noise descriptive metrics (tSNR - central row, PICSO - bottom row) along three sequence settings (AxR1MB2, CorR1MB3, CorR2MB3) of two randomly selected edges (see column header). Note that, as the total acceleration increases (from left to right of each plot), the noise propagation not necessarily follow same trend of the FC amplitude though overall correlates non trivially depending on the prescribed slicing orientation employed.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
2401