denoising data with multi-echo epi · • this technique allows teasing apart data with multi -echo...
TRANSCRIPT
![Page 1: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/1.jpg)
DENOISING DATA WITH MULTI-ECHO EPI(…Multi-echo EPI, Multi-echo EPI, Multi-echo EPI, Multi-echo EPI, Multi-Echo EPI, Multi-Echo EPI……)
Martin M Monti, PhDUCLA Department of Psychology
http://montilab.psych.ucla.edu
![Page 2: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/2.jpg)
THE PROBLEM
“[…] data from standard (i.e. single-echo) fMRI pulse sequences is limited by the fundamental problem that in such experiments, Blood Oxygen Level Dependent (BOLD) signal arising from spontaneous neuronal activity is not differentiable from fluctuations arising from cardiac and respiratory physiology, motion, and many other sources.” Kundu et al., 2012
0
0.5
1
1.5
2
2.5
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
-0.5
0
0.5
1
1.5
2
2.5
1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97
trend slow quadratic HF Motion
![Page 3: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/3.jpg)
THE PROBLEM
• Task-based fMRI (GLM-based analysis):• a-priori hypothesis of the signal of interest. If noise is
correlated with the task-related activity it can produce false activations/deactivations/etc.
• Resting state fMRI: • NO a-priori hypothesis about the signal of interest: any
correlation with noise will produce false positives
Slide from L. Griffanti
![Page 4: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/4.jpg)
APPROACH #1: THE STANDARD
• Motion correction
![Page 5: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/5.jpg)
APPROACH #1: THE STANDARD
• Motion correction• But residual motion, intra-TR
motion, spin history effects remain
• Outliers/Scrubbing (Power et al., 2012)
• Might lose a lot of data• Global Mean Signal regression
• Anti-correlations issue (Murphy et al., 2009)
• WM/CSF Regression• Doesn’t do that much (Power
et al., 2012)• Physiological recording
(Glover et al, 2000)
![Page 6: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/6.jpg)
APPROACH #1: THE STANDARD
• Spatial filtering
![Page 7: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/7.jpg)
APPROACH #1: THE STANDARD
• Motion correction• Spatial filtering
• Low Freq Noise
![Page 8: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/8.jpg)
APPROACH #1: THE STANDARD
• Motion correction• Spatial filtering
• Low Freq Noise• High Freq Noise
![Page 9: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/9.jpg)
APPROACH #2: DATA DRIVEN STRUCTURED NOISE REMOVAL
• Use data driven method to find noise and remove it from the data
• Still have to run standard preprocessing including motion correction, high-pass filtering
• Identify bad components• Subjective (if done manually)
• Remove bad components from signal
![Page 10: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/10.jpg)
APPROACH #2: DATA DRIVEN STRUCTURED NOISE REMOVAL
• Use data driven method to find noise and remove it from the data
• Still have to run standard preprocessing including motion correction, high-pass filtering
• Identify bad components• Subjective (if done manually)
• Remove bad components from signal
![Page 11: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/11.jpg)
APPROACH #2: DATA DRIVEN STRUCTURED NOISE REMOVAL
• Use data driven method to find noise and remove it from the data
• Still have to run standard preprocessing including motion correction, high-pass filtering
• Identify bad components• Subjective (if done manually)
• Remove bad components from signal
![Page 12: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/12.jpg)
APPROACH #2: DATA DRIVEN STRUCTURED NOISE REMOVAL
• Use data driven method to find noise and remove it from the data
• Still have to run standard preprocessing including motion correction, high-pass filtering
• Identify bad components• Subjective (if done manually)
• Remove bad components from signal
• Today, automated tools exist (FIX)
![Page 13: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/13.jpg)
APPROACH #3: MULTI-ECHO EPI
![Page 14: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/14.jpg)
APPROACH #3: MULTI-ECHO EPI
• “We introduce a new method that employs multi-echo acquisition and a TE-dependence test to remove artefactual fluctuations more effectively than these previous approaches by cleanly separating BOLD and non-BOLD signal components of resting state data.” Kundu et al., 2012
![Page 15: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/15.jpg)
BACK TO BASICS
![Page 16: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/16.jpg)
BACK TO BASICS
![Page 17: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/17.jpg)
APPROACH #3: MULTI-ECHO EPI
• “We introduce a new method that employs multi-echo acquisition and a TE-dependence test to remove artefactual fluctuations more effectively than these previous approaches by cleanly separating BOLD and non-BOLD signal components of resting state data.” Kundu et al., 2012
![Page 18: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/18.jpg)
APPROACH #3: MULTI-ECHO EPI
![Page 19: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/19.jpg)
APPROACH #3: MULTI-ECHO EPI
• “We introduce a new method that employs multi-echo acquisition and a TE-dependence test to remove artefactual fluctuations more effectively than these previous approaches by cleanly separating BOLD and non-BOLD signal components of resting state data.” Kundu et al., 2012
![Page 20: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/20.jpg)
APPROACH #3: MULTI-ECHO EPI
• “We introduce a new method that employs multi-echo acquisition and a TE-dependence test to remove artefactual fluctuations more effectively than these previous approaches by cleanly separating BOLD and non-BOLD signal components of resting state data.” Kundu et al., 2012
NON-BOLD model BOLD model
![Page 21: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/21.jpg)
KUNDU ET AL 2012Task & Rest
Non-BOLDchanges
BOLDchanges
% SignalChange
Task - Rest
Good signals haveto fit this model
(goodness of fit F)!
IF a signal fits this model(goodness of fit F), then it’s not
of BOLD origin!
(κ)
(ρ)
![Page 22: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/22.jpg)
KUNDU ET AL 2012Task & Rest
Non-BOLDchanges
BOLDchanges
% SignalChange
Task - Rest
The key idea is that, when expressed in terms of percent
signal change, we know how the BOLD signal should behave
as TE increases.
![Page 23: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/23.jpg)
KUNDU ET AL 2012
![Page 24: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/24.jpg)
Slide from P. Bandettini
![Page 25: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/25.jpg)
RESULTS
%SC
![Page 26: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/26.jpg)
RESULTS
![Page 27: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/27.jpg)
RESULTS
![Page 28: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/28.jpg)
RESULTS
![Page 29: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/29.jpg)
RESULTS
• What happens nearby the κelbow?
• A component with a near-threshold κ score could reflect ΔR2* modulation from respiratory variation or related BOLD-like effects of no interest.
![Page 30: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/30.jpg)
DATA DRIVENANALYSIS
If interested in data driven analysis, reject low κcomponents and keep only high κ components above the elbow.
![Page 31: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/31.jpg)
SEED-BASEDANALYSIS
If interested in seed-based connectivity analysis, then filter out low κ and high ρcomponents from the data (e.g., as in FIX) prior to analysis.
![Page 32: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/32.jpg)
SEED-BASEDANALYSIS
If interested in seed-based connectivity analysis, then filter out low κ and high ρcomponents from the data (e.g., as in FIX) prior to analysis.
![Page 33: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/33.jpg)
SEED-BASEDANALYSIS
If interested in seed-based connectivity analysis, then filter out low κ and high ρcomponents from the data (e.g., as in FIX) prior to analysis.
At the same threshold standard approach shows no sig. results
![Page 34: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/34.jpg)
SUMMARIZING
• This technique allows teasing apart data with multi-echo EPI, identifying BOLD-like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components directly from the data, and using these non BOLD-like components to obtain nuisance regressors.
• PROs: • Based on the characteristic properties of BOLD T2* signal (i.e., transverse
susceptibility-weighted relaxation rate).• Takes advantage of what ICA does best• Does not require external physiologic measures, temporal noise models, or
anatomical templates• Is fully automated
• CONs:• Multi-echo data (multi-echo data, multi-echo data)
Kundu et al., 2012 NI
![Page 35: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/35.jpg)
COMPARING APPROACHES
Kundu et al., 2009, PNAS; 2017 NI
artifact
signal
![Page 36: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/36.jpg)
ME-ICA AND MOTION
Kundu et al., 2017 NI
Low k (i.e., non-BOLD components)High k (i.e., BOLD components)
![Page 37: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/37.jpg)
ME-ICA AND TASK-BASED (BLOCK DESIGN) FMRI: MENTALIZING
• Task based approach w/block design• Task A: Self/Others (reflective judgments about themselves/HMQ,
mentalistic/physical)• Task B: Listen to stories (mentalistic/social/physical contents)
• Analyses:• ME-ICA denoising• Standard approach• GLMDenoise (Kay et al 2013)
![Page 38: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/38.jpg)
ME-ICA AND TASK-BASED
(BLOCK DESIGN)
FMRI: MENTALIZING
![Page 39: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/39.jpg)
ME-ICA AND TASK-BASED
(BLOCK DESIGN)
FMRI: MENTALIZING
![Page 40: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/40.jpg)
ME-ICA AND TASK-BASED
(BLOCK DESIGN)
FMRI: MENTALIZING
![Page 41: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/41.jpg)
ME-ICA AND TASK-BASED (BLOCK DESIGN) FMRI: MENTALIZING
![Page 42: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/42.jpg)
IMPROVING FMRI AT 7TConventional single EPI
Simple sum of multi-echo EPI Kundu et al., 2009 NI
![Page 43: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/43.jpg)
IMPROVING FMRI AT 7T
Kundu et al., 2017 NI
![Page 44: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/44.jpg)
IMPROVING FMRI AT 7T
![Page 45: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/45.jpg)
IMPROVING FMRI AT 7T
Kundu et al., 2009 NI
![Page 46: Denoising data with multi-echo EPI · • This technique allows teasing apart data with multi -echo EPI, identifying BOLD- like (high κ, low ρ) non-BOLD-like (low κ, high ρ) components](https://reader033.vdocuments.mx/reader033/viewer/2022041711/5e485eadfd47da691c1b47e1/html5/thumbnails/46.jpg)
REFERENCES• Kundu, P., Inati, S. J., Evans, J. W., Luh, W. M., & Bandettini, P. A. (2012). Differentiating
BOLD and non-BOLD signals in fMRI time series using multi-echo EPI. Neuroimage, 60(3), 1759-1770.
• Kundu, P., Voon, V., Balchandani, P., Lombardo, M. V., Poser, B. A., & Bandettini, P. (2017). Multi-Echo fMRI: A Review of Applications in fMRI Denoising and Analysis of BOLD Signals. NeuroImage.
• Lombardo, M. V., Auyeung, B., Holt, R. J., Waldman, J., Ruigrok, A. N., Mooney, N., ... & Kundu, P. (2016). Improving effect size estimation and statistical power with multi-echo fMRI and its impact on understanding the neural systems supporting mentalizing. NeuroImage, 142, 55-66.
• Kundu, P., Brenowitz, N. D., Voon, V., Worbe, Y., Vértes, P. E., Inati, S. J., ... & Bullmore, E. T. (2013). Integrated strategy for improving functional connectivity mapping using multiecho fMRI. Proceedings of the National Academy of Sciences, 110(40), 16187-16192.