Multi-echo Functional Connectivity as an evaluation metric for denoising
Poster No:
1375
Submission Type:
Abstract Submission
Authors:
Javier Gonzalez-Castillo1,2, Cesar Caballero-Gaudes2,3, Daniel Handwerker1, Peter Bandettini1,4
Institutions:
1Section on Functional Imaging Methods, National Institute of Mental Health, Bethesda, MD, 2Basque Center on Cognition, Brain and Language, San Sebastian, Spain, 3Ikerbasque, Basque Foundation for Science, Bilbao, Spain, 4FMRI Core, National Institute of Mental Health, Bethesda, MD
First Author:
Javier Gonzalez-Castillo
Section on Functional Imaging Methods, National Institute of Mental Health|Basque Center on Cognition, Brain and Language
Bethesda, MD|San Sebastian, Spain
Section on Functional Imaging Methods, National Institute of Mental Health|Basque Center on Cognition, Brain and Language
Bethesda, MD|San Sebastian, Spain
Co-Author(s):
Cesar Caballero-Gaudes
Basque Center on Cognition, Brain and Language|Ikerbasque, Basque Foundation for Science
San Sebastian, Spain|Bilbao, Spain
Basque Center on Cognition, Brain and Language|Ikerbasque, Basque Foundation for Science
San Sebastian, Spain|Bilbao, Spain
Daniel Handwerker
Section on Functional Imaging Methods, National Institute of Mental Health
Bethesda, MD
Section on Functional Imaging Methods, National Institute of Mental Health
Bethesda, MD
Peter Bandettini
Section on Functional Imaging Methods, National Institute of Mental Health|FMRI Core, National Institute of Mental Health
Bethesda, MD|Bethesda, MD
Section on Functional Imaging Methods, National Institute of Mental Health|FMRI Core, National Institute of Mental Health
Bethesda, MD|Bethesda, MD
Introduction:
Evaluating denoising for fMRI is challenging because we lack gold standards of how fully denoised data should look. While we can rely on basic heuristics (e.g., anatomical viability, presence of canonical networks, test-retest reliability, etc.) to discard clearly problematic data; lacking substantial data issues, no experienced neuroimager can unequivocally say which of the Default Network maps or FC matrices shown in Fig. 1.A and B are better, and therefore what denoising approach should be selected.
Multi-echo (ME) fMRI can help alleviate this issue as one can mathematically derive expectations for how FC should vary across echoes when data is dominated by only one type of fluctuations: BOLD or non-BOLD (Fig 1.C). Pearson's correlation-based FC (FC-R) should be TE-independent when BOLD fluctuations dominate (Fig 1.D). Unfortunately, the same is true for non-BOLD. Yet, Covariance-based FC (FC-C) is TE-independent only when data is dominated by non-BOLD fluctuations and shows TE-dependence when BOLD fluctuations dominate (Fig 1.E).
Here we apply these ideas to 434 ME (TE=13.7/30/47ms) scans from the Neurocognitive Aging dataset (Spreng et al., 2022) and quantitatively examine how effective tedana is at removing non-BOLD fluctuations.
Multi-echo (ME) fMRI can help alleviate this issue as one can mathematically derive expectations for how FC should vary across echoes when data is dominated by only one type of fluctuations: BOLD or non-BOLD (Fig 1.C). Pearson's correlation-based FC (FC-R) should be TE-independent when BOLD fluctuations dominate (Fig 1.D). Unfortunately, the same is true for non-BOLD. Yet, Covariance-based FC (FC-C) is TE-independent only when data is dominated by non-BOLD fluctuations and shows TE-dependence when BOLD fluctuations dominate (Fig 1.E).
Here we apply these ideas to 434 ME (TE=13.7/30/47ms) scans from the Neurocognitive Aging dataset (Spreng et al., 2022) and quantitatively examine how effective tedana is at removing non-BOLD fluctuations.
Methods:
Data was pre-processed in two ways using afni_proc (Reynolds et al., 2024) :
1) Basic Denoising: Polynomials, Motion and CompCor regression.
2) Advanced Denoising: Basic Denoising and non-BOLD tedana components.
After pre-processing, FC-R and FC-C matrices were computed for all 9 echo pairs (e.g., (TE1,TE1), (TE1,TE2), etc.) using the Power Atlas (Power et al., 2011). Next, for each scan, the slope and intercept of the linear regression between all possible FC pairs (e.g., (TE1,TE1) vs. (TE1,TE2)) was estimated.
For FC-R we report average of estimated Slope and Intercept per scan.
For FC-C, because slope for BOLD-dominated data is expected to be TE dependent, we report results in terms of the Euclidean distance between empirically obtained slope/intercept and those expected for BOLD-dominated data using the DBOLD metric described in Fig 1.F.
1) Basic Denoising: Polynomials, Motion and CompCor regression.
2) Advanced Denoising: Basic Denoising and non-BOLD tedana components.
After pre-processing, FC-R and FC-C matrices were computed for all 9 echo pairs (e.g., (TE1,TE1), (TE1,TE2), etc.) using the Power Atlas (Power et al., 2011). Next, for each scan, the slope and intercept of the linear regression between all possible FC pairs (e.g., (TE1,TE1) vs. (TE1,TE2)) was estimated.
For FC-R we report average of estimated Slope and Intercept per scan.
For FC-C, because slope for BOLD-dominated data is expected to be TE dependent, we report results in terms of the Euclidean distance between empirically obtained slope/intercept and those expected for BOLD-dominated data using the DBOLD metric described in Fig 1.F.
Results:
Fig. 2.A-B shows single scan FC-R and FC-C, respectively, for two TE pairs. FC-R matrices look quite similar, and when contrasted, fall near the identity line (Slope=1, Inter=0). Conversely, FC-C matrices look substantially different. When contrasted, data (blue dots and line) falls in between the non-BOLD dominated (red line) and BOLD dominated (green line) expected behaviors; yet, closer to the latter.
Fig 2.C shows how slope and intercept for FC-R changes between Basic and Advanced denoising for the whole sample. Each vector represents one scan. Their color and direction indicate whether following Advanced denoised data became more dominated by a single fluctuation type. For most scans, that is the case (i.e., most vectors are blue and point towards (0,1)).
Fig 2.D shows DBOLD changes between both denoising approaches for all scans (each dot represents a scan). For most scans, DBOLD decreases for Advanced denoising relative to Basic, as indicated by most dots sitting below the red-colored region. This confirms that tedana brought most scans closer to the behavior expected for BOLD-dominated data. A few scans did not behave this way (those in red shaded region), signaling potential issues during denoising. It is also worth noticing, that even in cases of relative improvement, scans still sit far apart from an ideal scenario where all non-BOLD fluctuations have been removed (e.g., green line at the bottom for DBOLD,Advanced=0).
Fig 2.C shows how slope and intercept for FC-R changes between Basic and Advanced denoising for the whole sample. Each vector represents one scan. Their color and direction indicate whether following Advanced denoised data became more dominated by a single fluctuation type. For most scans, that is the case (i.e., most vectors are blue and point towards (0,1)).
Fig 2.D shows DBOLD changes between both denoising approaches for all scans (each dot represents a scan). For most scans, DBOLD decreases for Advanced denoising relative to Basic, as indicated by most dots sitting below the red-colored region. This confirms that tedana brought most scans closer to the behavior expected for BOLD-dominated data. A few scans did not behave this way (those in red shaded region), signaling potential issues during denoising. It is also worth noticing, that even in cases of relative improvement, scans still sit far apart from an ideal scenario where all non-BOLD fluctuations have been removed (e.g., green line at the bottom for DBOLD,Advanced=0).
Conclusions:
We demonstrate how computing FC across echoes can help quantitatively evaluate effectiveness of denoising algorithms and identify problematic scans. Our evaluation confirms that tedana improves data quality over basic denoising for most scans. Yet, it also highlights how, even then, data still includes some residual non-BOLD signal fluctuations.
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Novel Imaging Acquisition Methods:
BOLD fMRI 2
Keywords:
Data analysis
FUNCTIONAL MRI
Other - Multi-echo, functional connectivity
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
DuPre, E., Salo, T., Ahmed, Z., Bandettini, P., Bottenhorn, K., Caballero-Gaudes, C., Dowdle, L., Gonzalez-Castillo, J., Heunis, S., Kundu, P., Laird, A., Markello, R., Markiewicz, C., Moia, S., Staden, I., Teves, J., Uruñuela, E., Vaziri-Pashkam, M., Whitaker, K., Handwerker, D., 2021. TE-dependent analysis of multi-echo fMRI with tedana. J. Open Source Softw. 6, 3669. https://doi.org/10.21105/joss.03669
Power, J.D., Cohen, A.L., Nelson, S.M., Wig, G.S., Barnes, K.A., Church, J.A., Vogel, A.C., Laumann, T.O., Miezin, F.M., Schlaggar, B.L., Petersen, S.E., 2011. Functional Network Organization of the Human Brain. Neuron 72, 665–678. https://doi.org/10.1016/j.neuron.2011.09.006
Reynolds, R.C., Glen, D.R., Chen, G., Saad, Z.S., Cox, R.W., Taylor, P.A., 2024. Processing, evaluating and understanding FMRI data with afni_proc.py. arXiv. https://doi.org/10.48550/arxiv.2406.05248
Spreng, R.N., Setton, R., Alter, U., Cassidy, B.N., Darboh, B., DuPre, E., Kantarovich, K., Lockrow, A.W., Mwilambwe-Tshilobo, L., Luh, W.-M., Kundu, P., Turner, G.R., 2022. Neurocognitive aging data release with behavioral, structural and multi-echo functional MRI measures. Sci. Data 9, 119. https://doi.org/10.1038/s41597-022-01231-7