INS NYC 2024 Program

Paper	Paper Session 15 Program Schedule 02/16/2024 03:30 pm - 04:55 pm Room: West Side Ballroom - Salon 3 Paper Session 15: Neuroanatomy, Neuroimaging, and Neurophysiology Final Abstract #5 Understanding Test-Retest Reliability of Within-Network Connectivity in a Clinical Population: A “Mini” Multiverse Approach Hollie Mullin, Penn State University, State College, United States Catherine Carpenter, Penn State University, State College, United States Andrew Cwiek, Penn State University, State College, United States Gloria Lan, Penn State University, State College, United States Emily Carter, Penn State University, State College, United States Samantha Vervoordt, Penn State University, State College, United States Amanda Rabinowitz, Moss Rehabilitation Research Institute, Elkins Park, United States Umesh Venkatesan, Moss Rehabilitation Research Institute, Elkins Park, United States Frank Hillary, Penn State University, State College, United States Category: Neuroimaging Keyword 1: traumatic brain injury Keyword 2: neuroimaging: functional connectivity Keyword 3: brain injury Objective: Resting-state functional magnetic resonance imaging (rsfMRI) has the ability to provide information about brain functioning. However, it is difficult to interpret conclusions about rsfMRI data due to questions about the reliability of resting-state functional connectivity (RSFC). One issue is that there are many data processing pipelines to choose from, including how to handle head motion artifacts, and choices made during this stage impact network reproducibility. This study investigated the test-retest reliability of RSFC in men and women with moderate-to-severe traumatic brain injury (TBI) using back-to-back rsfMRI scans. Results were compared between different pipelines. This is an understudied area that can improve our understanding of RSFC and its potential use in clinical populations. Participants and Methods: 46 (13 female) individuals with moderate-to-severe TBI and 41 (17 female) healthy controls (HCs) received consecutive, 10-minute rsfMRI scans. XCP-D was used to post-process data using three pipelines: a 36-nuisance parameter model with motion scrubbing, a 36-nuisance parameter model without motion scrubbing, and a model using neither nuisance regressors nor motion scrubbing. If included, XCP-D performed scrubbing using a framewise displacement (FD) threshold of 0.5. XCP-D divided the brain into 400 regions based on the Schaefer atlas, resulting in a 400 x 400 Fisher r-to-z transformed, weighted, connectivity matrix per subject. The regions were grouped into 17 networks based on the Yeo atlas. Intraclass correlation coefficients (ICCs) were calculated to examine the reliability of within-network connectivity (strength of a region’s connection to other regions within the same network) across each subject’s back-to-back rsfMRI scans. Results: The 36-parameter model with motion scrubbing resulted in the lowest ICCs across the brain networks for both the TBI group (ICCs = 0.00-0.61) and HCs (ICCs = 0.11-0.77). ICCs for the men (ICCs = 0.00-0.65) and women (ICCs = 0.00-0.84) reinforced this trend. Scrubbing influenced the TBI group more, perhaps due to greater motion in the TBI sample (FD = 0.31) compared to HCs (FD = 0.23). Using the 36-parameter model without motion scrubbing, ICCs increased for the TBI group (ICCs = 0.55-0.84) and the HCs (ICCs = 0.55-0.83). ICCs for the men (ICCs = 0.58-0.85) and women (ICCs = 0.38-0.87) demonstrated this as well. The model using neither nuisance regressors nor motion scrubbing resulted in noisy data, but ICCs remained reliable for the TBI group (ICCs = 0.65-0.85) and HCs (0.54-0.86). This finding was also consistent for men (ICCs = 0.57-0.86) and women (ICCs = 0.64-0.90). For the TBI and HC groups, the default mode (ICCs = 0.25-0.82) and salience networks (ICCs = 0.44-0.85) were relatively stable and demonstrated greater reliability compared to other networks, regardless of the pipeline. Conclusions: This study emphasizes that data processing pipelines, including motion scrubbing, impact RSFC reliability. In the study of brain injury, reproducible canonical networks (e.g., default mode network) can serve as benchmarks to assess neuroplastic changes due to brain trauma. These data reveal that RSFC can be used to identify reproducible findings after TBI that may serve as biomarkers for clinical outcomes.

Paper

02/16/2024
03:30 pm - 04:55 pm
Room: West Side Ballroom - Salon 3

Paper Session 15: Neuroanatomy, Neuroimaging, and Neurophysiology

Final Abstract #5

Understanding Test-Retest Reliability of Within-Network Connectivity in a Clinical Population: A “Mini” Multiverse Approach

Hollie Mullin, Penn State University, State College, United States
Catherine Carpenter, Penn State University, State College, United States
Andrew Cwiek, Penn State University, State College, United States
Gloria Lan, Penn State University, State College, United States
Emily Carter, Penn State University, State College, United States
Samantha Vervoordt, Penn State University, State College, United States
Amanda Rabinowitz, Moss Rehabilitation Research Institute, Elkins Park, United States
Umesh Venkatesan, Moss Rehabilitation Research Institute, Elkins Park, United States
Frank Hillary, Penn State University, State College, United States

Category: Neuroimaging

Keyword 1: traumatic brain injury
Keyword 2: neuroimaging: functional connectivity
Keyword 3: brain injury

Objective:

Resting-state functional magnetic resonance imaging (rsfMRI) has the ability to provide information about brain functioning. However, it is difficult to interpret conclusions about rsfMRI data due to questions about the reliability of resting-state functional connectivity (RSFC). One issue is that there are many data processing pipelines to choose from, including how to handle head motion artifacts, and choices made during this stage impact network reproducibility. This study investigated the test-retest reliability of RSFC in men and women with moderate-to-severe traumatic brain injury (TBI) using back-to-back rsfMRI scans. Results were compared between different pipelines. This is an understudied area that can improve our understanding of RSFC and its potential use in clinical populations.

Participants and Methods:

46 (13 female) individuals with moderate-to-severe TBI and 41 (17 female) healthy controls (HCs) received consecutive, 10-minute rsfMRI scans. XCP-D was used to post-process data using three pipelines: a 36-nuisance parameter model with motion scrubbing, a 36-nuisance parameter model without motion scrubbing, and a model using neither nuisance regressors nor motion scrubbing. If included, XCP-D performed scrubbing using a framewise displacement (FD) threshold of 0.5. XCP-D divided the brain into 400 regions based on the Schaefer atlas, resulting in a 400 x 400 Fisher r-to-z transformed, weighted, connectivity matrix per subject. The regions were grouped into 17 networks based on the Yeo atlas. Intraclass correlation coefficients (ICCs) were calculated to examine the reliability of within-network connectivity (strength of a region’s connection to other regions within the same network) across each subject’s back-to-back rsfMRI scans.

Results:

The 36-parameter model with motion scrubbing resulted in the lowest ICCs across the brain networks for both the TBI group (ICCs = 0.00-0.61) and HCs (ICCs = 0.11-0.77). ICCs for the men (ICCs = 0.00-0.65) and women (ICCs = 0.00-0.84) reinforced this trend. Scrubbing influenced the TBI group more, perhaps due to greater motion in the TBI sample (FD = 0.31) compared to HCs (FD = 0.23). Using the 36-parameter model without motion scrubbing, ICCs increased for the TBI group (ICCs = 0.55-0.84) and the HCs (ICCs = 0.55-0.83). ICCs for the men (ICCs = 0.58-0.85) and women (ICCs = 0.38-0.87) demonstrated this as well. The model using neither nuisance regressors nor motion scrubbing resulted in noisy data, but ICCs remained reliable for the TBI group (ICCs = 0.65-0.85) and HCs (0.54-0.86). This finding was also consistent for men (ICCs = 0.57-0.86) and women (ICCs = 0.64-0.90). For the TBI and HC groups, the default mode (ICCs = 0.25-0.82) and salience networks (ICCs = 0.44-0.85) were relatively stable and demonstrated greater reliability compared to other networks, regardless of the pipeline.

Conclusions:

This study emphasizes that data processing pipelines, including motion scrubbing, impact RSFC reliability. In the study of brain injury, reproducible canonical networks (e.g., default mode network) can serve as benchmarks to assess neuroplastic changes due to brain trauma. These data reveal that RSFC can be used to identify reproducible findings after TBI that may serve as biomarkers for clinical outcomes.