3873
Slice Timing Effects on Spectral Dynamic Causal Modeling1University of Wisconsin - Madison, Madison, WI, United States, 2Medical College of Wisconsin, Milwaukee, WI, United States
Synopsis
Functional MRI may not be slice time corrected when time differences are less than one second. However, these differences may be important for causal neuronal relationships. We investigated slice timing correction in spectral dynamic causal modeling using multiband resting-state data with TR = 802 ms from controls and temporal lobe epilepsy patients. Differences investigated were numeric value, parameter interpretation, and group level models. No significant difference in the first two were found. Group level modeling differences were found. Thus, slice timing correction appears necessary when applying spectral dynamic causal modeling even with slice timing differences of less than one second.
Introduction
Great debate still exists for processing steps for functional MRI (fMRI) data such as global signal regression in resting state data1. Previously, slice timing corrections were found not to be needed for TRs of less than 1s due to the blurring of the hemodynamic response function for the original task-based dynamic causal model (DCM)2. However, slice timing correction has not been investigated in recent DCMs which incorporate stochastic elements in modeling unlike the previously used deterministic task models. Given the stochastic nature of these models, the effects of preprocessing are likely to differ from their deterministic counterparts. Thus, we investigated the effects of slice timing correction on the stochastic cross-spectral density DCM (spDCM)3.Methods
Resting state fMRI data were acquired using 3T GE 750 scanners with whole-brain simultaneous multi-slice imaging (8 bands, 72 slices, TR 802 ms, voxel size 2mm isotropic). 17 healthy controls and 17 left temporal lobe epilepsy patients enrolled in the NIH-sponsored Epilepsy Connectome Project were used in analysis. Two preprocessing pipelines were used: one with slice timing correction initially applied and one without application of slice timing correction. The rest of the pipeline was consistent with the HCP minimal preprocessing pipeline4 and included smoothing of fMRI data using a 6mm Gaussian kernel, time series extraction of white matter and CSF masks, nuisance regression of fMRI data using white matter and CSF time series and the 6 motion parameters and their derivatives as derived during HCP minimal preprocessing pipeline. Time series for the default mode network ROIs from Van Dijk et al (2010)5 were then extracted. Fully connected spDCMs were then constructed using the default mode ROIs. Analysis of the differences between slice time corrected data and uncorrected data proceeded in three ways: (1) numeric stability using t-tests, (2) parameter interpretation differences occurring greater than a 50% using binomial tests, and (3) group level network differences using parametric empirical Bayes (PEB) modeling and Bayesian model reduction and averaging6.Results
No significant differences were found following multiple testing for either the t-tests or the binomial tests. However, the PEB model using slice time corrected data had a higher free energy both in the full and reduced models. Following thresholding to strong evidence (posterior probability of greater than 95%), the reduced slice time corrected PEB model had 21 parameters for the group mean and four parameters for the group difference while the reduced non-slice time corrected PEB model had 25 group mean parameters and 1 group difference parameter. The group difference parameters were unique for each model (Figures 1 and 2).Discussion
Though none of the t-tests or binomial tests were significant following multiple testing correction, the thresholded reduced PEB models were drastically different from one another. Without slice timing, the only group difference found was a self-connection. With slice timing correction, no self-connections, only extrinsic connections between regions differed between groups. This suggests that, unlike the initial task-based DCM, spDCM is not robust to slice timing differences even when a TR less than one second is used.Conclusion
Though slice timing correction has been shown to be unnecessary in many cases of fMRI analysis, this analysis demonstrates that slice timing correction may still be important even with a TR of less than 1s when using spDCM. Further investigation using simulated data is needed.Acknowledgements
The authors would like to thank all the participants and their families. Additionally, the authors would like to thank all those involved in the Epilepsy Connectome Project. Funding for healthy control subjects' data acquisition was provided in part by the Department of Radiology at the University of Wisconsin - Madison. This study was supported by grant number U01NS093650 from the National Health Institute (J Binder and M Meyerand Co-PIs).References
1. Murphy, K. & Fox, M. D. Towards a consensus regarding global signal regression for resting state functional connectivity MRI. Neuroimage 154, 169–173 (2017).
2.Friston, K. J., Harrison, L. & Penny, W. Dynamic causal modelling. Neuroimage 19, 1273–1302 (2003).
3. Friston, K. J., Kahan, J., Biswal, B. & Razi, A. A DCM for resting state fMRI. Neuroimage 94, 396–407 (2014).
4. Glasser, M. F. et al. The minimal preprocessing pipelines for the Human Connectome Project. Neuroimage 80, 105–124 (2013).
5. Van Dijk, K. R. A. et al. Intrinsic Functional Connectivity As a Tool For Human Connectomics: Theory, Properties, and Optimization. J. Neurophysiol. 103, 297–321 (2010).
6. Friston, K. J. et al. Bayesian model reduction and empirical Bayes for group (DCM) studies. Neuroimage 128, 413–431 (2016).
Figures
