Comparison of the Relationships Between Body Size and Cardiorespiratory Fitness With High Frequency Head‐Motion Contamination in fMRI Authors: Marco Pipoly, Hayley Chappell, Lyndsey E DuBose, Bryan Madero, Adriana Rivera‐Dompenciel, Matthew Sodoma, Chris Oehler, Will Daniels, Kelsey Baller, Jenna Springer, Matthew Armstrong, Virginia Nuckols, Vincent Magnotta, Jeffrey Long, Gary L Pierce, Michelle W Voss Published: 2026 ABSTRACT ABSTRACTFunctional MRI (fMRI) is widely used to assess brain function, but neural‐derived fMRI signals are susceptible to contamination from in‐scanner head motion. Preprocessing pipelines often regress out and censor head motion artifacts using six rigid‐body motion parameters. However, recent research suggests these head‐motion estimates are susceptible to contamination from other sources of noise, such as respiratory rate, body size, and estimated cardiorespiratory fitness (eCRF), that correspond to apparent head motion at higher frequencies (HF‐motion; > 0.1 Hz). Thus, these health variables are thought to be linked to the artifactual inflation of head motion estimates, introducing additional challenges to appropriate modeling and preprocessing procedures that reduce noise. Whether this artifactual relationship extends to changes in body size, midline abdominal fat (i.e., body composition), CRF measured with gold‐standard methods, and respiratory rate measured during scanning remains unclear. The Brain EXTEND Trial acquired multiple health variables before and after an exercise intervention that changed CRF, offering a means to test the relationship between these health variables and HF‐motion in a longitudinal design. Subjects (55–80 years of age) completed a 6‐month chronic exercise intervention at two intensities. We analyzed baseline and longitudinal (EXTEND, baseline n = 122; longitudinal n = 84) relationships between HF‐motion and health variables that included body mass index (BMI), waist circumference (WC), CRF measured with a maximal exercise test (VO2max), and resting respiratory rate. HF‐motion was examined in each head‐motion parameter as the proportion of HF‐motion above > 0.1 Hz—the typical upper bound of fMRI signals at rest. At baseline, HF‐motion for all translational axes was positively associated with body size (BMI, WC) while CRF was negatively associated with only the roll y‐rotation. Additionally, longitudinally, decreased BMI related to reduced HF‐motion in the z‐translation. Results indicate health changes related to body size associate with the presentation of high‐frequency MRI head‐motion artifacts. We caution researchers against ignoring the presence of HF‐motion in their analyses as their use of contaminated motion traces will negatively affect their modeling of fMRI measures. Results have implications for a wide range of investigations in health neuroscience involving body size, respiratory function, and cardiorespiratory fitness.Keywords: exercise, health behavior, intervention, neuroimaging, older adults Introduction Advanced age is associated with changes in body size, body composition, and cardiorespiratory fitness (CRF) that together impact age‐related changes in brain function (Rapp et al. 2018; Tari et al. 2025; Voss and Jain 2022). Exercise interventions can help to mitigate the adverse effects of advanced age by improving body composition, CRF, brain blood flow, and cognitive function (Jayedi et al. 2024; Stillman et al. 2017; Voss et al. 2019). Functional magnetic resonance imaging (fMRI), particularly when individuals are not performing a task, termed “rest‐state,” is frequently used to evaluate the efficacy of an exercise prescription on brain function (Baniqued et al. 2018; Intzandt et al. 2020; Moore et al. 2022; Voss et al. 2011; Won et al. 2023). However, interpreting rest‐state fMRI outcomes can be complicated by in‐scanner head motion, which obscures the locale of fMRI signals (Power et al. 2012). Such in‐scanner head‐motion artifacts disturb meaningful brain signals by appearing in later processing as structured noise (Power et al. 2014), which can differ systematically by age (Gratton et al. 2020; Satterthwaite et al. 2012), body size (Gratton et al. 2020; Power et al. 2019), and clinical status (Dosenbach et al. 2017). Emerging evidence suggests head motion at higher frequencies (HF‐motion) is primarily due to respiratory processes that are modulated by body size and CRF (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). This raises a key concern for clinical trial research targeting brain function through health‐behavior change, as a change in health may lead to confounding changes in estimates of in‐scanner head motion by the intervention prescribed. Yet, prior research has been cross‐sectional in nature, limiting our understanding of whether health‐related changes in CRF or body size relate to changes in higher frequency in‐scanner head motion. Head motion during an fMRI scan poses a significant technical challenge for the examination of brain activity in both translational and basic science (Ciric et al. 2017; Power et al. 2012; Siegel et al. 2014). FMRI measures neural activity via changes in oxygenation and blood flow associated with neuronal firing demands, known as the blood‐oxygen‐level‐dependent (BOLD) signal. Accurate BOLD measurement at rest requires precise spatial and temporal information to approximate where and when neural‐derived changes in BOLD signal occur. One common method of describing age‐related decline in brain function relies on the BOLD time series to estimate functional coupling of oscillatory BOLD signals between brain regions, known as “functional connectivity” (Brier et al. 2012). Sub‐millimeter changes in head position due to in‐scanner head movement cause temporally and spatially structured shifts in image intensity and signal origin, confounding and inflating functional connectivity (Power et al. 2012). A number of common approaches such as motion “scrubbing,” despiking, and frame censoring have been proposed to deal with fMRI frames contaminated by large, bulk head position shifts (Ciric et al. 2017, 2018; Jones et al. 2022). Frame‐by‐frame shifts in head position along each axis of head movement modeled as a spherical “rigid‐body” can be represented in a summary estimate called framewise displacement (FD; Power et al. 2012). FD estimates bulk head movement by summarizing millimeter (mm) shifts in the x‐, y‐, z‐ translations and x‐ (pitch), y‐ (roll), z‐ (yaw) rotations from head rigid‐body realignment parameters at each fMRI frame (Figure 1B), quantified as the change in position (in units of mm) from the previous frame (t‐1) (Power et al. 2012). The utility of FD comes from its use as a frame based summary score to despike, censor, or exclude frames from analyses, or to exclude whole fMRI runs (see example Pipoly et al. 2024). However, FD is vulnerable to inflation from noise associated with body size and health related constraints on respiration rate that contribute to instances of apparent head motion. Depicts an overview of head movement in a single subject fMRI run from the Brain EXTEND trial alongside a depiction of the experimental design adapted from Voss et al. (2024). (A) The relative orientation of movement in the x‐translation, y‐translation, z‐translation, pitch, roll, and yaw (x‐, y‐, z‐rotations) with respect to a human head. (B) A single subject's change in head motion for the x‐translation (red‐line), y‐translation (blue‐line), z‐translation (grey‐line), pitch (gold‐line), roll (black‐line), and yaw (orange‐line) at each sampling point (akin to repetition time of 2 s). (C) Y‐axis depicts relative power and x‐axis depicts frequency. The black bar depicts the inflection point for density reference > 0.1 Hz, where respiratory HF‐motion is likely to be captured in each motion parameter. Red arrows indicate elevated HF‐motion in the y‐translation and z‐translation in an example EXTEND baseline (pre‐intervention) subject run. (D) Schematic of the administration of the exercise intervention where MRI, cognitive, and health variable measurement (i.e., CRF, BMI) were acquired at baseline and again at 6‐months, post‐intervention, for use in this secondary analyses. Trans = translation; Rot = rotation. The relation of in‐scanner head motion with body size and respiration is complex. During respiration, the periodic expansion of the chest wall can lead to real or apparent displacements in MRI head position (Fair et al. 2020; Power et al. 2019). These apparent displacements in head position, or “apparent head motion,” can be proportional to body size and/or the degree of lung expansion and the effort required for restful breathing (Power et al. 2019). In particular, the onset of inhalation can shift the apparent in‐scanner position of the brain, creating the appearance of head movement despite no actual head movement having taken place (Power et al. 2019). These respiratory linked apparent head motion‐like artifacts are important to disentangle from overt head motion as each has distinct consequences to the BOLD signal (Glasser et al. 2018). For example, actual head motion degrades fMRI BOLD signal quality (Friston et al. 1996; Siegel et al. 2014), whereas apparent respiratory‐related head motion introduces periodic shifts in signal frequency (Fair et al. 2020; Power et al. 2019). Characterizing each artifact requires careful thought and separate strategies in terms of both the data quality assessment used and preprocessing procedure applied. A recent strategy to address the influence of body size and respiration on head motion estimates has included filtering each motion parameter estimate to minimize high frequency signal contamination (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). As others have shown (Gratton et al. 2020), ignoring this issue will lead to inflated estimates of bulk head motion and the unintended removal of data flagged as noise due to other noise sources. A challenge to understanding HF‐motion is that some respiratory‐linked apparent shifts in head motion may be concurrent with true head movement (Glasser et al. 2018; Power et al. 2019). Larger individuals often exhibit lower lung capacity, elevated respiratory rate (Littleton 2012; Luce 1980), and limited torso mobility to accommodate chest and abdominal breathing styles (Littleton 2012; Luce 1980). The physical constraints of a larger body size may alter patterns of chest expansion during breathing by reducing the range of trunk mobility, which can increase the possibility of respiratory locked displacements of the neck and head, resulting in true head movement during fMRI scanning (Power et al. 2019). Although the precise source of each form of noise is difficult to detect, researchers have statistically dissociated features of body size from CRF (Gratton et al. 2020) and respiratory‐linked influences on patterns of HF‐motion (Power et al. 2019). HF‐motion has been described from several different measurement approaches (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019, 2020). One robust and simple approach summarizes the proportion of HF‐motion contamination in each run/scan based on the relative power density spectra above > 0.1 Hz to the Nyquist limit (Figure 1C). This estimate of HF‐motion has been shown to index the amount of high frequency contamination present in each motion parameter trace (Gratton et al. 2020). The utility of identifying respiratory‐linked noise above > 0.1 Hz lies in its relevance to processing resting fMRI signals, which are dominated by lower‐frequency oscillations (Fair et al. 2020; Gratton et al. 2020). Because young adults breathe 12 to 20 times per minute (i.e., 0.2–0.3 Hz), respiratory measures collected alongside higher fMRI sampling rates are better suited to identify and remove high‐frequency respiratory noise in motion estimates > 0.1 Hz (as articulated by Fair et al. 2020). However, this is not always possible in traditional, slower TR single‐band fMRI acquisitions, because higher frequency respiratory noise can alias into lower frequency motion‐trace ranges that overlap with brain signals of interest (Fair et al. 2020). As a result, motion traces estimated with a slower sampling rate may contain respiratory‐related signals both above and below 0.1 Hz (Fair et al. 2020; Gratton et al. 2020). The process by which respiration is embedded into in‐scanner head motion has been well‐described. Displacements associated with HF‐motion are often described within the context of specific rigid‐body parameter estimates (Fair et al. 2020; Power et al. 2019). For example, rhythmic breathing has been shown to mostly affect HF‐motion in the phase encoding direction (i.e., AP, PA, RL, LR; Fair et al. 2020). Practically, this means most studies focus on the y‐translation, the z‐translation, and pitch rotation because traditional sequences often do not use RL or LR acquisitions (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). Restful breathing from an individual laying supine (as typically done during MRI) can affect the y‐ and to lesser extent z‐translational planes because breathing often corresponds to front to back (most common phase encoding direction) and upward‐downward movement (Figure 1A), matching common phase encoding acquisitions. Yet, the spread or “leakage” of high frequency respiratory signals from one axis to another are noted to occur from the transformations associated with rigid‐body alignment, which has led to reports of HF‐motion like noise appearing in other parameter estimates (see appendix A Fair et al. 2020). Differences in fMRI temporal resolution may reduce the clarity of respiratory relations to HF‐motion. However, a study by Gratton et al. (2020), showed that across multiple legacy lifespan datasets with slower single‐band acquisitions at ~2 s sampling rates, respiratory‐like noise was present in the expected y‐translation across a broad range of ages. Their study reported that HF‐motion in the y‐translation, which was their phase encoding direction, increased with age, body mass index (BMI), and lower estimated CRF. Given that respiratory rates often disturb specific motion axes such as the phase encoding y‐translation (Fair et al. 2020), and may vary by CRF level, researchers (Gratton et al. 2020) proposed an estimate of CRF may be useful to understand HF‐motion and respiratory‐like noise. A major limitation to this conclusion acknowledged by researchers was their lack of access to measured respiratory and gold‐standard CRF measurements. Nevertheless, Gratton et al. (2020) study provided compelling evidence that health variables like CRF, in addition to body size and composition, may be useful to understand HF‐motion in slow single‐band fMRI data acquisitions. Therefore, we aimed to replicate and extend their results by testing the cross‐sectional association between baseline health measures of BMI and CRF, and respiration with HF‐motion. Based on these studies, it is also reasonable to hypothesize that changes in health variables, such as BMI and CRF, would be associated with change in scanner head motion and related sources of noise. Accordingly, we tested the prediction that exercise‐induced reductions in body size, alongside increased CRF, would be associated with lower HF‐motion. The Brain EXTEND study (Voss et al. 2024) was a randomized controlled trial that had subjects complete a chronic experimental exposure to one of two different exercise intensity programs over a 6‐month period (Figure 1D). The longitudinal and health‐targeted nature of the Brain EXTEND trial allowed us to test our hypothesis at baseline and after chronic exposure to a six‐month exercise manipulation that was shown to increase CRF (Voss et al. 2024). Materials and Methods Subjects were recruited within a 1‐hour driving radius of the University of Iowa, Iowa City, IA. Outreach methods are detailed elsewhere (Voss et al. 2024), but briefly included culturally inclusive mailed materials, mass email, word of mouth, advertisements, and community presentations. IRB approval was provided for continued secondary analyses. Secondary analyses reported here only utilize data collected from the recently completed randomized controlled Brain EXTEND trial (Voss et al. 2024). Inclusion and exclusion criteria were developed to reach the target goals of the Brain EXTEND trial (NCT03114150), and are described elsewhere (Voss et al. 2024). In brief, this study inclusion criteria consisted of: (1) 55–80 years of age, (2) a completed health history questionnaire that indicated no overt or subclinical cardiovascular disease, (3) completion of baseline maximal exercise test with 12‐lead ECG. Our exclusion criteria consisted of: (1) self‐reported psychiatric or neurological disorders that may impair cerebrovascular or cognitive function (i.e., psychosis, dementia), (2) prior history of brain injury or loss of consciousness > 30 min, (3) self‐reported diagnosis of disease or conditions associated with increased cerebrovascular disease risk, or alterations in cardiovascular functioning that included: heart disease, chronic obstructive pulmonary disease, renal or liver disease, unregulated thyroid disorder (i.e., not on stable dose of medication for at least 3 months before enrollment), (4) smoking or living with someone who has smoked in the past 3 months, (5) self‐reported regular use of steroid‐based medication, psychotropics, and/or chemotherapy, (6) female subjects were postmenopausal and had not taken hormone therapy for at least 6 months, (7) a race and education adjusted score on the Montreal Cognitive Assessment (MoCA) of < 21 out of 30, which would identify people with potential dementia (Nasreddine et al. 2005). Procedure Subjects first completed a study orientation session that included informed consent, cognitive screening, health history/demographic surveys, and a mock MRI. Following the orientation session, a total of 122 subjects completed baseline assessments that included (1) a maximal exercise test to screen for exercise training safety and measure CRF, (2) two neuropsychological testing sessions, and (3) two MRI sessions consisting of a single exercise session and structural and functional MR imaging (Voss et al. 2024). A total of n = 12 subjects met criteria for possible mild cognitive impairment at baseline after race/ethnicity and educational attainment MoCA adjustment (Milani et al. 2018). Following baseline assessments, subjects who met eligibility criteria were then admitted and randomized to a crossover designed acute exercise manipulation, beyond the scope of this study. Subjects who completed the acute manipulation were then randomized into one of two chronic exercise groups over a 6‐month period. At the conclusion of the study, a total of 98 subjects (Light intensity: n = 48; ModVig intensity: n = 50) completed the post‐intervention maximal exercise testing, neuropsychological, health metric, and MRI assessment. As reported elsewhere (Voss et al. 2024), the Brain EXTEND trial exercise intervention was successful at increasing CRF in the higher intensity group (i.e., ModVig) relative to the lower intensity group (i.e., Light). Only data from baseline (pre) and the chronic 6‐month exercise exposure were used in this study (see Table 1). We acknowledge here that due to technical device issues and constraints from the Covid‐19 pandemic, a subset of respiratory, CRF, and waist circumference (WC) measures were not available for analysis. Additionally, an unanticipated scanner model change during data collection led us to conduct a sensitivity analysis to determine whether HF‐motion differed by scanner model. Analysis results suggested HF‐motion may differ by scanner model (see Supplemental Results S1). Therefore, we chose to exclude 14 subjects from the final 98 post‐intervention sample who were either missing health measures and/or were scanned with different scanner models between pre and post. Due to the reduced sample size from the confound of scanner model change, we chose not to model an interaction between intervention group and time for this study. When relevant, sample size differences between analysis families are reported in the results to assist reader interpretation. Baseline demographics did not significantly differ by exercise intensity group as reported by Voss et al. (2024). TABLE 1Brain EXTEND trial demographics.Baseline (Pre—n = 122)Intervention completed: Light (Post—n = 48)Intervention completed: ModVig (Post—n = 50)Change analysis: Sub‐sample (n = 84)Age (years)63.40 ± 5.9063.70 ± 5.5063.10 ± 6.7063.50 ± 6.10Sex (F/M)80/4230/1836/1458/26Education (years)16.50 ± 2.4016.60 ± 2.5016.40 ± 2.1016.50 ± 2.20BMI (kg/m2)29.30 ± 5.5029.30 ± 6.30 (n = 42)28.70 ± 5.20 (n = 48)28.70 ± 5.80WC (cm)98.307 ± 14.10 (n = 113)98.147 ± 14.90 (n = 32)97.399 ± 14.10 (n = 26)96.925 ± 15.22 (n = 46)MoCA27.30 ± 1.80———VO2max (ml/kg/min)20.58 ± 5.4721.85 ± 5.97 (n = 42)22.12 ± 5.50 (n = 48)21.93 ± 5.69RR Median (bpm)15.20 ± 0.90 (n = 116)15.70 ± 0.90 (n = 47)15.70 ± 1.20 (n = 48)15.50 ± 1.10 (n = 82)FD Mean (mm)0.208 ± 0.100.191 ± 0.080.214 ± 0.110.202 ± 0.10X‐Trans HF‐mot0.306 ± 0.050.299 ± 0.050.311 ± 0.050.307 ± 0.05Y‐Trans HF‐mot0.415 ± 0.100.405 ± 0.090.411 ± 0.100.406 ± 0.10Z‐Trans HF‐mot0.374 ± 0.090.364 ± 0.070.373 ± 0.080.366 ± 0.08X‐Rot (Pitch) HF‐mot0.360 ± 0.080.346 ± 0.060.349 ± 0.080.348 ± 0.07Y‐Rot (Roll) HF‐mot0.310 ± 0.050.308 ± 0.050.316 ± 0.050.314 ± 0.05Z‐Rot (Yaw) HF‐mot0.302 ± 0.050.296 ± 0.040.301 ± 0.050.299 ± 0.05RaceCaucasian (n = 114), Asian (n = 4), Black (n = 2), Multi (n = 2)Caucasian (n = 45), Multi (n = 2), Asian (n = 1)Caucasian (n = 47), Black (n = 2), Asian (n = 1)Caucasian (n = 79), Asian (n = 2), Black (n = 2), Multi (n = 1)Ethnicity (Hispanic: No/Yes)121/147/150/083/1Table 1. Demographic and health measures are reported by session (mean ± SD) for relevant baseline (pre), complete intervention (post), and change analysis subject groups. Variable scale: age is in years, sex is self‐reported as at birth, education in years, BMI as body mass index (kg/m2), WC as waist circumference (cm), MoCA (Montreal Cognitive Assessment) as a continuous measure, relative VO2max (CRF) after weight adjustment (ml/kg/min). MRI acquired measures are represented as an average of three runs within each pre‐ and post‐intervention session, which includes derived RR as resting respiratory rate (mean of median run level, units are breaths per minute; bpm), FD as framewise displacement (mm), and HF‐motion for each x‐, y‐, and z‐translations (i.e., Trans) and rotations (i.e., Rot: pitch, roll, yaw) parameter (measured as the proportion of high frequency noise out of 1.0—from a frequency range above > 0.1 Hz to Nyquist limit). We note MoCA was not reassessed post‐intervention (i.e., ‘—’ in table), and available data for RR and WC is smaller at baseline (RR n = 116, WC n = 113) relative to the full sample, while WC, CRF, and RR measures were smaller at 6‐month follow‐up, for both the full post‐intervention (WC n = 58, CRF n = 92, RR n = 90) and the change analysis sub‐samples (WC n = 46, RR n = 82). WC is reported but not used in longitudinal analyses due to too few observations. Full description and range for all Brain EXTEND trial subject data is available elsewhere (Voss et al., 2024). Brain EXTEND trial demographics. Table 1. Demographic and health measures are reported by session (mean ± SD) for relevant baseline (pre), complete intervention (post), and change analysis subject groups. Variable scale: age is in years, sex is self‐reported as at birth, education in years, BMI as body mass index (kg/m2), WC as waist circumference (cm), MoCA (Montreal Cognitive Assessment) as a continuous measure, relative VO2max (CRF) after weight adjustment (ml/kg/min). MRI acquired measures are represented as an average of three runs within each pre‐ and post‐intervention session, which includes derived RR as resting respiratory rate (mean of median run level, units are breaths per minute; bpm), FD as framewise displacement (mm), and HF‐motion for each x‐, y‐, and z‐translations (i.e., Trans) and rotations (i.e., Rot: pitch, roll, yaw) parameter (measured as the proportion of high frequency noise out of 1.0—from a frequency range above > 0.1 Hz to Nyquist limit). We note MoCA was not reassessed post‐intervention (i.e., ‘—’ in table), and available data for RR and WC is smaller at baseline (RR n = 116, WC n = 113) relative to the full sample, while WC, CRF, and RR measures were smaller at 6‐month follow‐up, for both the full post‐intervention (WC n = 58, CRF n = 92, RR n = 90) and the change analysis sub‐samples (WC n = 46, RR n = 82). WC is reported but not used in longitudinal analyses due to too few observations. Full description and range for all Brain EXTEND trial subject data is available elsewhere (Voss et al., 2024). MRI Acquisition and Head‐Motion Processing Briefly, MRI scans were collected with a posterior‐to‐anterior (PA) phase encoding direction on either a 3T GE Discovery 750 W with a 32‐channel coil or a Signa Premier MRI scanner with a 48‐channel coil, respectively, at the University of Iowa Magnetic Resonance Research Facility. Subjects scanned with the Discovery (D) model (n = 52, 42.6%) and Signa Premier (P) model (n = 70, 57.4%) did not differ in age, education, sex, or CRF (all p > 0.05), and training groups were balanced in the number of subjects scanned at baseline on each model and who had the same scanner pre/post (DD, PP) or switched (DP) (χ2(2) = 1.39, p = 0.50) as reported elsewhere (Voss et al. 2024). A covariate for scanner model is included in all baseline (pre) and longitudinal change analyses based on the results of our sensitivity analysis that suggested scanner model affects patterns of HF‐motion (see Supplemental Results S1a). A baseline pre‐ and post‐intervention MRI session acquired multiple runs of cross sectional and longitudinal measures to assess brain function. One high‐resolution anatomical image was collected via a three‐dimensional fast spoiled gradient echo sequence (FSPGR) BRAVO T1w with the following parameters: TI = 900 ms, TE = 3.168 ms, TR = 8.388 ms, flip angle = 8°, Acquisition Matrix = 256 × 256 × 196, FOV = 256 × 256 × 196, 1 mm3 isotropic voxels. All functional images, whose rigid‐body transformation motion parameters were used for these analyses, were acquired with ascending axial slice acquisition with no gap between slices, with an echo‐planar gradient echo sequence: TE = 30 ms, TR = 2000 ms, flip angle = 80°, FOV = 220 × 220, with an acquisition matrix of 64 × 64 mm and 37 slices (voxel size of 3.45 × 3.45 × 4 mm). Prior to scanning, subjects were instructed not to exercise for at least 24 hours all the way up to the hour of scanning. All subjects' data used in analyses were based on the three 8‐min rest‐state fMRI scans (comprised of 240 frames each) at pre as well as the three 8‐min rest‐state scans collected at post to quantify HF‐motion. The total time for each MRI session (pre and post) was 24 min. Resting wakefulness was achieved by instructing subjects to attend their eyes to a central fixation while remaining relaxed and awake throughout the rest‐state scanning session. Processing of motion parameters from fMRI scanning occurred after rigid body alignment. Six rigid‐body head realignment parameters (x, y, z, pitch, roll, yaw) were used to estimate motion within each 8‐min rest‐state scan prior to any spatial or temporal processing (e.g., slice time correction, EPI distortion correction) following established recommendations (Power et al. 2017). Rotation parameters (pitch, roll, yaw) were calculated in radians with FSL's mcflirt (Jenkinson et al. 2002), which we converted to mm based on an assumed 50 mm head radius (Power et al. 2012). FD was calculated for reference in Table 1 as the sum of the absolute value of motion parameter change at each time point in mm for every run (Power et al. 2012). HF‐motion from respiration related artifacts is more likely to appear in acquisitions where higher fMRI sampling rates are employed (Fair et al. 2020). To ensure we observed HF‐motion in our phase encoding (y‐translation) and z‐translation directions, as others have observed, we conducted a visual inspection of HF‐motion. We observed higher frequency noise at higher densities above > 0.1 Hz to the Nyquist limit in our sample of mid to late‐life adults (Figure 2). This pattern was most readily present in the y‐translation, z‐translation, and to a lesser extent the x‐rotation (pitch) alongside the other motion parameters. These patterns precisely mirror previously reported respiratory related HF‐motion noise (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019), justifying our examination of relationships between HF‐motion and health metrics. Depicts the proportion of relative power at each frequency (x‐axis) for each motion parameter arranged per run (n = 3 runs) in run acquisition order at baseline. Each subject's run group is arranged by the overall magnitude of their cross run mean HF‐Motion such that overall higher HF‐motion subject data is clustered at the bottom of each graph while lower HF‐motion data is clustered at the top. Vertical white lines indicate the > 0.1 Hz cutoff for high frequency signals. Y‐axis is subject run count, and x‐axis is frequency in 0.5 Hz increments. Red arrows indicate the presence of significant clusters of relatively higher HF‐motion runs for some parameters above 0.1 Hz. HF‐motion was computed for each run's six motion parameters in the frequency domain following multitaper spectral decomposition of the full, unfiltered, time series. Let Xmit represent the time series of each motion parameter mi={1, …, 6}, that was sampled at a frequency rate of fs=1TR. The power spectra of each motion parameter was calculated with the multitaper power spectral density (PMTM) function in Matlab as Pmif=PMTMXmitNW=8fs, where NW denotes the time half bandwidth product, which we set to 8 (as in Gratton et al. 2020). Vector Pmi represents spectral power estimates over frequencies that range from 0 to the Nyquist limit (fNyq=fs2). Pmi was log‐scaled, z‐scored across frequencies, and re‐scaled into an estimate of “relative power” R(fi) using min‐max normalization, where each frequency fi bin represents a value between 0 and 100. Last, HF‐motion was quantified as the summed proportion of the normalized spectral quantity > 0.1 Hz up to the fNyq, which can be expressed as HF‐motion = ∑fi>0.1Rfi∑iRfi following (code: www.github.com/GrattonLab/Gratton2020_NI_HFmotion/blob/master/mot_FFT.m) previous definitions (Gratton et al. 2020). All analyses reported utilize a session average of HF‐motion (i.e., average of three runs per session). Head motion (Hodgson et al. 2016) and breathing (Power et al. 2019) are considered trait‐like and more similar to the individual than to other subjects (Power et al. 2019). Given that three runs of data were acquired per subject for each session, we chose to average across runs so analyses were representative of average session‐level fluctuations in subject HF‐motion. We recognize prior studies have concatenated available runs to estimate HF‐motion (Fair et al. 2020; Gratton et al. 2020), which raised the possibility that averaging might dilute estimates. We visually inspected (see Figure 3) and performed an additional analysis of variance to confirm between parameter specific differences in HF‐motion were still present after session‐level averaging. A one‐way ANOVA confirmed that a mean difference between motion parameters is present F (5, 1) = 52.31, p < 0.001 at baseline (illustrated in Figure 3). Each subject's session‐level average proportion of HF‐motion is plotted here from pre‐intervention baseline data. % of standardized relative power > 0.1 Hz is expressed as a proportion out of 100 or 1 Hz to the Nyquist limit. Colors depict magnitude of average HF‐motion. Y‐axis represents subject count and x‐axis motion parameter name. Data is arranged by y‐translation HF‐motion so that higher HF‐motion individuals are at the bottom and lower HF‐motion individuals are at the top. Lighter colors mean higher HF‐motion, while darker colors mean lower HF‐motion. Red arrows indicate which parameters had the highest amount of HF‐motion. Trans = translation; Rot = rotation. Health Metric Procedures and Measures Anthropometric health metrics used to describe body size and body composition included BMI and WC. BMI was calculated as weight (kg)/height (m)2. WC was measured in centimeters as the midpoint between the iliac crest and the lowest rib margin (Jacobs et al. 2010). Prior work has shown that BMI is less specific to fat distributions around the midline as opposed to WC (Jacobs et al. 2010; Wier et al. 2006), which is important for disentangling unhealthy rises in midline fat distribution that may restrict trunk movement, increasing the likelihood that smaller shifts in body position will translate into overt head motion (Power et al. 2019). To compare our understanding of these relations within the Brain EXTEND trial data, we ran a parallel analysis in our data with both BMI and WC measures, when available, to evaluate body size contributions to HF‐motion relations. Cardiorespiratory fitness was measured as VO2max at baseline pre‐ and post‐training with a symptom‐limited maximal exercise test to exhaustion, on an upright cycle ergometer with 12‐lead electrocardiography (ECG), and with cardiopulmonary gas analysis (True One, Parvomedics Inc.). A maximal exercise test measures CO2 expired relative to oxygen inspired in response to increasing workload, as well as heart rate to identify maximum heart rate (HR). Each test was completed with a trained specialist and physician's oversight by a board certified cardiologist. A trained research assistant first placed a 12‐lead ECG apparatus on the subject prior to exercise test initiation where an initial supine resting blood pressure measure was taken. A standardized ramp protocol increased resistance every 2 min, with HR, ratings of perceived exertion (RPE), and blood pressure recorded at the end of every stage (Voss et al. 2024). VO2max was determined based on whether subjects satisfied two of the following three criteria: a respiratory exchange ratio value of ≥ 1.10, RPE of 17 or greater (on 6–20 Borg scale) (Borg 1982), maximal heart rate within 10 beats/min based on predicted maximal heart rate (220‐age or attainment of 80% of predicted heart rate max or greater). CRF analyses utilized relative VO2max (mL/kg/min) adjusted by body weight. Adjusting for body mass differences reduces bias and facilitates between subject comparisons of CRF related oxygen demand and consumption (DuBose et al. 2022; Ross et al. 2016). Respiratory rate was measured in the Brain EXTEND trial with an MRI compatible respiratory belt worn by subjects during MRI scanning. A series of analyses suggested half the sample had respiratory rates greater than the Nyquist limit. We therefore limit discussion of respiratory processing and analysis in the main manuscript. Instead, bulk discussion of respiratory rate and respiratory variability across session runs can be found in Supplemental Methods S1 and Supplemental Results S2. Results The relationship between HF‐motion with each motion parameter was tested cross‐sectionally and longitudinally with a subset of subject data from the complete intervention. All hypothesis tests examined the independent relationships of HF‐motion with anthropometric body size (i.e., BMI and WC) and physiological (i.e., CRF) measurements. Respiratory measurements were also analyzed in a series of models with each HF‐motion parameter. Results and limitations of respiratory analyses are discussed in Supplemental Results S2. Every analysis family was corrected for six comparisons at a Bonferroni adjusted α (0.05/6 = 0.0083) due to separate health variable models being run for each of the six motion parameters. HF‐Motion Is Associated With Anthropometric and CRF Baseline Health Metrics Our first analysis established whether baseline HF‐motion associates with BMI, WC, and CRF measured as relative VO2max beyond scanner model, age, and sex. A total of 3 linear regression models were tested for each HF‐motion parameter (totaling 18 models per health measure) following this formula: HF‐motion ~ age + sex + health measure + scanner model. Missing data was not imputed. Differences from the main sample size (n = 122) are noted here for WC (n = 113). First, we tested whether body size and body composition measured by BMI and WC vary with HF‐motion at baseline. Prior work established that BMI correlates with FD (Power et al. 2019) and HF‐motion (Gratton et al. 2020). Table 2 shows that BMI was associated with HF‐motion in the y‐translation, z‐translation, and pitch x‐rotation, such that greater BMI was related to greater HF‐motion (see Figure 4). Unsurprisingly, no significant association was found between BMI and HF‐motion in the x‐translation, y‐ and z‐rotations after multiple comparison correction. TABLE 2Pre‐intervention baseline HF‐motion models.BMI (kg/m2)WC (cm)CRF (ml/kg/min)HF‐motion outcomeUnstd. betaStd. betaSE p Unstd. betaStd. betaSE p Unstd. betaStd. betaSE p X Trans HF‐mot0.0010.1210.0920.1900.0010.1310.0990.187−0.001−0.0770.1180.517Y Trans HF‐mot0.0070.4120.080 <0.001 0.0030.4010.089 <0.001 −0.004−0.2400.1110.032 + Z Trans HF‐mot0.0050.3060.085 <0.001 0.0020.2830.092 0.003 −0.003−0.2170.1120.056Pitch HF‐mot0.0050.4020.085 <0.001 0.0020.3580.094 <0.001 −0.004−0.2630.1150.024 + Roll HF‐mot0.0020.1840.0900.043 + 0.0010.1900.0980.054−0.003−0.3510.112 0.002 Yaw HF‐mot0.0010.1660.0930.0760.0010.1690.1000.094−0.002−0.2140.1180.071Table 2. Cross‐sectional analysis results reported for each health variable is described in separate columns. Abbreviations: SE = standard error, from standardized model fits; Mot = motion; Rot = rotations; Trans = translation. Linear models were adjusted for scanner model, age, and sex. Note: Bold p values survive Bonferroni correction at 0.05. +Denotes marginal effects between α 0.05 and a corrected α of 0.0083 = 0.05/6. p > 0.001 are reported exactly. Pre‐intervention baseline HF‐motion models. Table 2. Cross‐sectional analysis results reported for each health variable is described in separate columns. Abbreviations: SE = standard error, from standardized model fits; Mot = motion; Rot = rotations; Trans = translation. Linear models were adjusted for scanner model, age, and sex. Note: Bold p values survive Bonferroni correction at 0.05. +Denotes marginal effects between α 0.05 and a corrected α of 0.0083 = 0.05/6. p > 0.001 are reported exactly. Depicts standardized beta estimates for hypothesis tests of HF‐motion and health metric relations in baseline data. X‐axis from left to right depicts relative VO2max as our gold‐standard measure of cardiorespiratory fitness (CRF), body mass index (BMI), and Waist Circumference (WC). Y‐axis from top to bottom depicts the x‐translation, y‐translation, z‐translation, pitch as x‐rotation, roll as y‐rotation, and yaw as z‐rotation column rows. Darker colors depict significance after Bonferroni correction (p < 0.0083) while lighter colors depict marginal effects between p < 0.05 and p ≥0.0083. Standardized beta weight cell colors correspond to negative associations when blue, or positive associations when red. Mot = motion; Rot = rotations; Trans = translation. Linear models were adjusted for scanner model, age, and sex. BMI is limited in its description of midline fat distribution. It is suggested that midline fat displaces the distribution of tissue closer to the MRI head‐coil when laying supine (Power et al. 2019). Therefore, we repeated analyses replacing BMI with WC to investigate whether fat around the waist (i.e., body composition) contributed to our BMI HF‐motion association. We replicated our BMI finding to show greater WC was related to greater HF‐motion in the y‐ and z‐ translation, as well as the pitch x‐rotation. Additionally, we also found a near marginal association for WC in the roll y‐rotation, as reported for BMI. This pattern closely resembled our BMI results with similar effect sizes (see Table 2), implicating midline fat deposition as a correlate of HF‐motion (see Figure 4). CRF was related to HF‐motion. Analyses revealed a negative trend between CRF and HF‐motion in the y‐ and z‐translation as well as the pitch x‐ and roll y‐rotations, with only the y‐rotation model surviving Bonferroni correction (Table 2); model results with standardized effect sizes are depicted in the Figure 4 tile plot alongside BMI and WC. Overall, we found HF‐motion is associated with BMI, WC, and CRF. To assess whether these associations also relate to general variation in head motion, we repeated each health variable analysis with FD, a measure of bulk head movement. Results showed greater body size (BMI, WC), in addition to lower CRF, relates to higher in‐scanner bulk head movements (see Supplemental Results S3a). Longitudinal Change Analysis Our goal was to determine whether our established associations between health variables and HF‐motion persisted across individuals who completed a chronic 6‐month exercise regimen. We tested this to understand whether a change in health from exposure to a 6‐month exercise regimen corresponded to a change in HF‐motion. All longitudinal change (∆) measures were quantified as post‐pre for each health and HF‐motion variable in a subset of 84 subjects (see sample subset description in Table 1 and Supplemental Results S1). Thirteen linear models were constructed as follows: ∆HF‐motion ~ ∆health variable + age + sex + group + scanner model. We first sought to establish whether ΔBMI and ΔCRF (measured as relative VO2max) relates to ∆HF‐motion. Results show that reductions in BMI from pre to post (i.e., more negative ΔBMI values) are significantly associated with reductions in z‐translation ΔHF‐motion (i.e., more negative ΔHF‐motion values), as we report in Table 3 and show in Figure 5. This pattern mirrored trends we observed in the association between ΔBMI and ΔFD (see Supplemental Results S3b). No other BMI model reached significance. Unlike baseline analyses, we did not find an association between a ΔCRF and a ΔHF‐motion in any motion parameter. TABLE 3Change‐score HF‐motion models.ΔBMI (kg/m2)ΔCRF (ml/kg/min)ΔHF‐motion outcomeUnstd. beta95% CIStd. betaSE p Unstd. beta95% CIStd. betaSE p X Trans HF‐mot0.002[−0.007, 0.011]0.0500.1170.6720.002[−0.003, 0.007]0.0950.1200.432Y Trans HF‐mot0.012[−0.000, 0.024]0.2290.1160.0510.005[−0.002, 0.012]0.1580.1210.196Z Trans HF‐mot0.020[0.011, 0.029]0.4640.107 <0.001 0.000[−0.006, 0.006]0.0050.1230.967Pitch HF‐mot0.002[−0.009, 0.014]0.0430.1170.7160.001[−0.006, 0.007]0.0250.1210.837Roll HF‐mot0.009[−0.001, 0.018]0.2000.1140.084−0.001[−0.006, 0.006]−0.0040.1200.974Yaw HF‐mot−0.001[−0.010, 0.007]−0.0400.1180.7390.003[−0.002, 0.008]0.1570.1210.199Table 3. Reduced scanner‐matched ‘Change Analysis‘ subjects (N = 84) with complete pre/post BMI and cardiorespiratory fitness (CRF) measured as relative VO2max; change scores for (post‐pre) ΔBMI and ΔCRF scores were modeled separately as independent main effects on ΔHF‐motion. Abbreviations: CI=confidence interval, from unstandardized model fits; SE=standard error, from standardized model fits; Mot=motion; Rot=rotations; Trans=translation. All models were adjusted for scanner model, intervention group, age, and sex. Note: Bold p values survive Bonferroni correction at 0.05. All p > 0.001 are reported exactly. Change‐score HF‐motion models. Table 3. Reduced scanner‐matched ‘Change Analysis‘ subjects (N = 84) with complete pre/post BMI and cardiorespiratory fitness (CRF) measured as relative VO2max; change scores for (post‐pre) ΔBMI and ΔCRF scores were modeled separately as independent main effects on ΔHF‐motion. Abbreviations: CI=confidence interval, from unstandardized model fits; SE=standard error, from standardized model fits; Mot=motion; Rot=rotations; Trans=translation. All models were adjusted for scanner model, intervention group, age, and sex. Note: Bold p values survive Bonferroni correction at 0.05. All p > 0.001 are reported exactly. Shows distinct relations amongst a ∆BMI and ∆HF‐motion outcome. Top row, right‐panel, shows Decreased ∆BMI (pre to post; more negative) relates to reduced ∆HF‐motion. Red circle highlights a subject that underscores the result of analyses. Only the model in the third column, top row panel is significant (p < 0.0083*). Data points jittered for clarity and adjusted by the influence of model covariates scanner model, intervention group, age, and sex. Shading represents a 95% confidence interval. Model results do not change when ∆BMI is restricted to values < 0.4. Trans = translation; Mot = motion. Discussion Regular moderate‐to‐vigorous aerobic exercise can affect body size and improve CRF, which confers protective benefits against the effects of old age on the brain (Voss et al. 2011; Voss and Jain 2022). Rest‐state BOLD fMRI is extensively used to non‐invasively evaluate the protective benefits of exercise on aging brain function (Baniqued et al. 2018; Intzandt et al. 2020; Moore et al. 2022; Voss et al. 2011; Won et al. 2023). However, outcomes derived from rest‐state BOLD fMRI are susceptible to inflation from frame‐by‐frame changes in head position during scanning (Power et al. 2012; Siegel et al. 2014). Elevated respiratory rate, body size and midline fat, and reduced CRF are all proposed to be contributing factors to distinct elevations in high frequency noise that contaminate nuisance variables used to benchmark noise in BOLD fMRI (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). We leveraged data from the Brain EXTEND trial to test whether an exercise‐induced benefit to health metrics related to change in HF‐motion over time, in an exercise intervention context. Results uncovered cross‐sectional and longitudinal relations between health metrics and parameter specific changes in HF‐motion. As predicted, we found that smaller body size (i.e., BMI, WC) and higher CRF (i.e., measured as relative VO2max) were related to lower HF‐motion in multiple HF‐motion axes at baseline. Unexpectedly, as described in Supplemental Section S2, we did not find support for a linear relationship between respiratory rate and HF‐motion for any parameter. Longitudinal analyses showed that reduced BMI related to reductions in HF‐motion of the z‐translation. Against our predictions, we did not find longitudinal evidence for a change in CRF relating to a change in HF‐motion for any parameter. However, these findings do support our prediction that changes in body size are associated with change in the proportion of relative HF‐motion contamination found in rigid‐body realignment estimates. A Question of Health Measurement and HF‐Motion at Baseline Baseline analyses replicated prior findings (Gratton et al. 2020) to show health measures from the Brain EXTEND trial do associate with HF‐motion during fMRI scanning. In our cross‐sectional mid to late‐life sample, body size analyses show high BMI and WC (abdominal adipose fat) relate to greater HF‐motion across the y‐ and z‐translations, and pitch x‐rotation. Additionally, we found lower CRF relates to higher HF‐motion in the roll rotation, with trends in the y‐translation and pitch x‐rotation. Overall, these results support our general prediction that smaller body size (i.e., BMI, WC), and higher CRF are associated with lower HF‐motion beyond the effects of age and sex. It is clear in our data that body size and composition dominantly explain HF‐motion at baseline. We propose two mechanistic explanations for why this may be the case. A larger body size, particularly when abdominal obesity/midline fat is high, can restrict trunk mobility when an individual is laying supine, which can lead to a change in breathing style and chest wall expansion (Dixon and Peters 2018; Power et al. 2019, 2020). Similarly, a restriction in trunk mobility due to higher abdominal fat may increase the likelihood that chest expansion during breathing translates to the neck and head, introducing both real head movement and apparent head movement into motion traces (Dixon and Peters 2018; Power et al. 2019, 2020). One might view the composition of the Brain EXTEND trial subjects at baseline as evidence for both explanations as their body size characteristics show a larger portion of the sample had a BMI above 25 (n = 94/122), that included a smaller number of individuals with a BMI of 30 or higher (n = 53/94), at the onset of the study. The United States Department of Health and Human Services suggests this pattern of BMI qualifies as overweight (> 25 BMI) or obese (> 30 BMI; U.S. Department of Health and Human Services 2023). Prior work has highlighted distinctions in the influence body size and CRF measures have on respiratory processes that underlie HF‐motion (Gratton et al. 2020; Power et al. 2019). Our results did not find support for overlap in BMI and CRF (as relative VO2max) relations to HF‐motion, contrasting some of the findings Gratton et al. (2020) reported with their y‐translation analysis. In our data, we only found a marginal association between CRF and HF‐motion in the y‐translation, whereas body size and composition showed a strong association with y‐translation HF‐motion. This pattern suggests body size and oxygen consumption may modulate different aspects of respiratory‐linked HF‐motion. In theory, respiratory‐linked sources of noise should be most prominent along the axes that correspond to our PA phase encoding direction, particularly the y‐ and z‐translation (Fair et al. 2020; Gratton et al. 2020). However, we found the strongest associations for CRF with HF‐motion of the roll y‐rotation. This finding was unexpected. One possibility is that our PA acquisition left the roll axis more vulnerable to respiratory‐linked noise related to oxygen consumption. Indeed, there is some evidence to support the emergence of HF‐motion in the roll rotation (Fair et al. 2020) for PA acquisitions. Another possibility relates to the proposed spread or “leakage” of signal between motion traces due to artifacts associated with the transformations that are implemented during the rigid‐body realignment (see Appendix A Fair et al. 2020 supplemental). Because of this, the practical application of motion trace correction is recommended to be applied to all axes (Fair et al. 2020). Respiratory rate and to a lesser extent, respiratory‐associated head movements, are the dominant source of HF‐motion across individual rigid‐body realignment parameters (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). Our investigation of respiratory rate did not yield a clear linear association with HF‐motion in any parameter, even when we removed individuals whose respiratory patterns were likely aliased into lower frequency ranges at baseline (see discussion in Supplemental Results S2). Given that our analyses focus on the proportion of noise over a limited frequency range in our relative power spectra, we speculate that it is possible our slow single‐band data is limited to unpack how respiratory processes were modulated by health variables to influence HF‐motion. Cross‐sectional analyses of the Brain EXTEND trial support that there is an association between HF‐motion and body size/composition and CRF. Exploratory analyses were not conclusive over the relation between a respiratory summary measure and the average proportion of HF‐motion at baseline (see Supplemental Results S2). Together, it is evident that even among relatively physically inactive older individuals, HF‐motion is related to multiple health metrics beyond age and sex. Chronic Exercise Induced CRF and BMI Change With HF‐Motion We examined whether a change in CRF and BMI was associated with parameter‐specific changes in HF‐motion. First, we confirmed that a chronic exercise prescription maintained over a 6‐month period successfully changed our health measures of interest as reported in Supplemental Results S1b and elsewhere (Voss et al. 2024). Our prediction that increased CRF and reduced BMI would correspond to a reduction in HF‐motion along our phase encoding axis, particularly in the y‐ and z‐translation, was only partially supported. Indeed, a reduction in overall weight was only associated with a reduction in overall z‐translation HF‐motion while accounting for age, sex, scanner model, and intervention group. No analyses revealed change in CRF relates to change in HF‐motion. Future studies will need to determine whether parameter‐specific changes in HF‐motion incrementally tie to specific features of exercise‐induced health variable change. For example, whether BMI change in central mass reduces variation in chest wall expansion or opens trunk mobility to reduce HF‐motion. Practical Implications of Change Analyses Our results support that an exercise intervention can contribute to alterations in patterns of high frequency noise contamination found in estimates measured from fMRI head reallignment. Researchers should, whenever possible, assess the presence of HF‐motion contamination to determine which strategies are needed to address noise in head motion estimates. We consider noise mitigation strategies and intervention‐based effects below. Lifespan changes in health may elevate in‐scanner HF‐motion, which artifactually inflates head motion estimates (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). Common pre‐processing procedures utilize head motion estimates to exclude and mitigate noise contamination in fMRI BOLD measures. However, the contamination of head motion estimates may lead to an increase in the number of flagged frames and subject‐level runs considered for removal, which is relevant for noise reduction strategies like motion scrubbing. Given the advent of high sampling rate, multi‐band data, and the presence of respiratory signal in some single‐band fMRI motion estimates, researchers should become more familiar with proposed strategies for reducing nuisance variable signal bias in protocol development, BOLD preprocessing, and nuisance measure processing (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). Several strategies have been proposed to address the contamination of rigid‐body realignment estimates of motion to save data and improve signal processing. For example, filtering of FD estimates across motion parameters is one recommended step to minimize the influence of respiratory related noise on motion traces (Fair et al. 2020; Gratton et al. 2020; Power et al. 2019). Specifically, single‐band fMRI data appear to benefit from applying a low‐pass first‐order Butterworth filter at 0.1 Hz across motion estimates (Gratton et al. 2020), which is in contrast to the notch filter recommended by others when data is acquired with a higher sampling rate (Fair et al. 2020). However, it is not clear whether slower sampling rates (i.e., > 3 s) employed by some clinical studies will benefit from parameter filtering. Thus, for studies where respiratory signals are likely to contaminate motion traces (i.e., fast sampling rates), a filtered FD approach alongside rigorous signal processing steps should be considered to alleviate high frequency signals in head motion estimates. Limitations and Future Directions The current study has several strengths that extend previous observations on the relation between CRF, body size and composition, and HF‐motion in single‐band fMRI. These include (1) the Brain EXTEND trial's health intervention design which supported our test of a change in BMI and CRF in relation to a change in HF‐motion, (2) our large quantity (i.e., 24 min) of session‐level resting‐state BOLD fMRI data and (3) our sample of 55–80‐year‐old adults, a typically high motion demographic, that are inactive and representative of mid to late‐life adults. However, there are also several limitations to this study. First, patterns of respiratory related signals in BOLD fMRI were not examined to understand whether a change in CRF, respiration, and BMI relate to differences in BOLD signal noise contamination. This distinction is important as it will likely prove challenging for researchers to decide how to approach noise reduction strategies in BOLD fMRI when health related changes in aged adults theoretically correspond to changes in both the nuisance variable signals and BOLD related brain activity. Future work will need to disentangle how sources of physiological noise influence the BOLD signal in the context of a health intervention. Second, our analyses were limited in their description of specific mechanisms driving HF‐motion. Prior work has proposed that the relation between HF‐motion and both BMI and eCRF are due to a combination of health factors that modulate variation in respiratory rate (Gratton et al. 2020), that include breathing style (i.e., chest or belly), chest cavity size, midline adiposity, and limited trunk mobility. It is unclear from the analyses reported which feature (e.g., breathing style, chest cavity) of these measures directly contributed to our samples pattern of HF‐motion. A problem complicated by our limited temporal resolution and change in scanner model which prevented a clear determination of whether respiratory rate distinctly influenced specific CRF and BMI associated HF‐motion relations. Future studies investigating HF‐motion should target measures of respiratory function alongside health variable change to clarify how breathing style (i.e., chest or belly), spurious breaths, shallow breathing, trunk mobility, and body size attributes (e.g., greater midline adipose fat) affect HF‐motion change in resting fMRI. Last, our study sample was predominantly composed of non‐demented, inactive, highly educated, and white identifying individuals. It is possible that a more diverse lifespan demographic, that varies in body size, shape, breathing style, and age, would strengthen our understanding of the relationship between exercise‐induced health variable change and change in parameter‐specific patterns of HF‐motion. Future studies will need to consider these factors to further refine and improve upon these methods. Funding This work was supported by the National Institute on Aging (R01AG055500), National Institute of General Medical Sciences (T32GM108540), and the Office of Research Infrastructure Program Division of Construction and Instruments (S10OD025025). Conflicts of Interest The authors declare no conflicts of interest. Supporting information Supplemental Figure S1: Each panel depicts differences in baseline (pre) HF‐motion (mean of three runs per session) between the Discovery (left panel boxplot) and Premier (right panel boxplot) GE scanner models. Panels are arranged such that the top row refers to HF‐motion data from the x‐, y‐, and z‐translations, whereas the bottom row refers to HF‐motion data in the x‐ (pitch), y‐ (roll), and z‐rotation (yaw). Each panel's x‐axis label describes the scanner model. The y‐axis denotes the proportion of average relative power > 0.1 Hz for each parameter. Within each panel, linear model standardized effect sizes, degree of freedom, and uncorrected p values are reported in the top right corner in blue text. Supplemental Figure S2: From left to right, each panel depicts the relationship between respiratory variability and each measure of HF‐motion for the x‐, y‐, z‐translation (top row), and x‐, y‐, z‐rotations (bottom row) in baseline (pre) data subjects with ≤ 15 bpm (n = 58). Data points reflect the relationship between each predictor and outcome measure after adjustment of model covariates estimated by the effect_plot package from the R software linear model object. Blue text reports the standardized effect size (β), standard error (SE), and p value from the main effect term of each model. RV median denotes that the x‐axis measure represents the average median respiratory variability estimate based on the average of respiratory runs acquired at baseline (pre). Y‐axis reflects the range of HF‐motion for each parameter estimate. Shading represents a 95% confidence interval. Supplemental Table S1: Standardized coefficients from separately fit linear models for Respiratory rate and Respiratory variability were adjusted for sex, age, and scanner model. Bonferroni‐corrected significance threshold: p < 0.0083. Bolded p‐values survive this correction. +Denotes marginal effects that are defined as 0.0083 <= p < 0.05. Respiratory rate (RR) and respiratory variability (RV) were modelled separately. Supplemental Figure S3: Distribution of respiratory rate values across all 116 baseline (pre) subjects. X‐axis depicts the respiratory rate in breaths per minute (bpm), and the y‐axis reflects the count of subjects. Eligible subjects are defined at less than or equal to 15 bpm (n = 58) and are shown in green. Subjects with a respiratory rate above 15 bpm are colored in orange, indicating those subjects were excluded from the primary tests of respiratory rate to HF‐motion relations. Supplemental Figure S4: X‐axis depicts each health variable in their native range (left CRF as relative VO2max, center BMI, right WC). Y‐axis reflects the average session level framewise displacement (FD). Data depicts the relationship between variables in 122 baseline participants. Blue text reports the standardized model effect size (β), standard error (SE), and p value from the main effect term of each model. Data from the R software linear model object was plotted based on an adjustment for covariates that included age, sex, and scanner model. Shaded area reflects 95% confidence intervals. Results do not change when outliers > 2.5sds are removed. Supplemental Figure S5: X‐axis depicts the ∆health variable in native range (left ∆CRF as relative VO2max, and right ∆BMI). Y‐axis reflects the average session level ∆framewise displacement (∆FD). Data depicts the relationship between each predictor and outcome variable across change analysis (n = 84) participants. Blue text reports the standardized model effect size (β), standard error (SE), and p value from the main effect term of each model. Data from the R software linear model object was plotted based on an adjustment for covariates that included age, sex, intervention group, and scanner model. Shaded area reflects 95% confidence intervals. Results do not change when outliers above 2.5sds are removed.