Controlling for the effect of arterial-CO2 fluctuations in resting-state fMRI: Comparing end-tidal CO2 clamping and retroactive CO2 correction

Machine learning-based estimation of respiratory fluctuations in a healthy adult population using resting state BOLD fMRI and head motion parameters

PURPOSE

External physiological monitoring is the primary approach to measure and remove effects of low-frequency respiratory variation from BOLD-fMRI signals. However, the acquisition of clean external respiratory data during fMRI is not always possible, so recent research has proposed using machine learning to directly estimate respiratory variation (RV), potentially obviating the need for external monitoring. In this study, we propose an extended method for reconstructing RV waveforms directly from resting state BOLD-fMRI data in healthy adult participants with the inclusion of both BOLD signals and derived head motion parameters.

METHODS

In the proposed method, 1D convolutional neural networks (1D-CNNs) used BOLD signals and head motion parameters to reconstruct the RV waveform for the whole fMRI scan time. Resting-state fMRI data and associated respiratory records from the Human Connectome Project in Young Adults (HCP-YA) dataset are used to train and test the proposed method.

RESULTS

Compared to using only BOLD-fMRI data for a CNN input, this approach yielded improvements of 14% in mean absolute error, 24% in mean square error, 14% in correlation, and 12% in dynamic time warping. When tested on independent datasets, the method demonstrated generalizability, even in data with different TRs and physiological conditions.

CONCLUSION

This study shows that the respiratory variations could be reconstructed from BOLD-fMRI data in the young adult population, and its accuracy could be improved using supportive data such as head motion parameters. The method also performed well on independent datasets with different experimental conditions.

Widespread Autonomic Physiological Coupling Across the Brain-Body Axis

The brain is closely attuned to visceral signals from the body's internal environment, as evidenced by the numerous associations between neural, hemodynamic, and peripheral physiological signals. We show that these brain-body co-fluctuations can be captured by a single spatiotemporal pattern. Across several independent samples, as well as single-echo and multi-echo fMRI data acquisition sequences, we identify widespread co-fluctuations in the low-frequency range (0.01 - 0.1 Hz) between resting-state global fMRI signals, neural activity, and a host of autonomic signals spanning cardiovascular, pulmonary, exocrine and smooth muscle systems. The same brain-body co-fluctuations observed at rest are elicited by arousal induced by cued deep breathing and intermittent sensory stimuli, as well as spontaneous phasic EEG events during sleep. Further, we show that the spatial structure of global fMRI signals is maintained under experimental suppression of end-tidal carbon dioxide (PETCO2) variations, suggesting that respiratory-driven fluctuations in arterial CO2 accompanying arousal cannot explain the origin of these signals in the brain. These findings establish the global fMRI signal as a significant component of the arousal response governed by the autonomic nervous system.

Effects of clamping end-tidal CO2 on neurofluidic low-frequency oscillations

In recent years, low-frequency oscillations (LFOs) (0.01-0.1 Hz) have been a subject of interest in resting-state functional magnetic resonance imaging research. They are believed to have many possible driving mechanisms, from both regional and global sources. Internal fluctuations in the partial pressure of CO2 (PCO2) has long been thought of as one of these major driving forces, but its exact contributions compared with other mechanisms have yet to be fully understood. This study examined the effects of end-tidal PCO2 (PetCO2) oscillations on LF cerebral hemodynamics and cerebrospinal fluid (CSF) dynamics under "clamped PetCO2" and "free-breathing" conditions. Under clamped PetCO2, a participant's PetCO2 levels were fixed to their baseline average, whereas PetCO2 was not controlled in free breathing. Under clamped PetCO2, the fractional amplitude of hemodynamic LFOs in the occipital and sensorimotor cortex and temporal lobes were found to be significantly reduced. Additionally, the fractional amplitude of CSF LFOs, measured at the fourth ventricle, was found to be reduced by almost one-half. However, the spatiotemporal distributions of blood and CSF delay times, as measured by cross-correlation in the LF domain, were not significantly altered between conditions. This study demonstrates that, while PCO2 oscillations significantly mediate LFOs, especially those observed in the CSF, other mechanisms are able to maintain LFOs, with high correlation, even in their absence.

Whole-brain analysis of CO2 chemosensitive regions and identification of the retrotrapezoid and medullary raphé nuclei in the common marmoset (Callithrix jacchus)

Respiratory chemosensitivity is an important mechanism by which the brain senses changes in blood partial pressure of CO2 (PCO2). It is proposed that special neurons (and astrocytes) in various brainstem regions play key roles as CO2 central respiratory chemosensors in rodents. Although common marmosets (Callithrix jacchus), New-World non-human primates, show similar respiratory responses to elevated inspired CO2 as rodents, the chemosensitive regions in marmoset brain have not been defined yet. Here, we used c-fos immunostainings to identify brain-wide CO2-activated brain regions in common marmosets. In addition, we mapped the location of the retrotrapezoid nucleus (RTN) and raphé nuclei in the marmoset brainstem based on colocalization of CO2-induced c-fos immunoreactivity with Phox2b, and TPH immunostaining, respectively. Our data also indicated that, similar to rodents, marmoset RTN astrocytes express Phox2b and have complex processes that create a meshwork structure at the ventral surface of medulla. Our data highlight some cellular and structural regional similarities in brainstem of the common marmosets and rodents.

Interoceptive Awareness of the Breath Preserves Attention and Language Networks amidst Widespread Cortical Deactivation: A Within-Participant Neuroimaging Study

Interoception, the representation of the body's internal state, serves as a foundation for emotion, motivation, and wellbeing. Yet despite its centrality in human experience, the neural mechanisms of interoceptive attention are poorly understood. The Interoceptive/Exteroceptive Attention Task (IEAT) is a novel neuroimaging paradigm that compares behavioral tracking of the respiratory cycle (Active Interoception) to tracking of a visual stimulus (Active Exteroception). Twenty-two healthy participants completed the IEAT during two separate scanning sessions (N = 44) as part of a randomized control trial of mindful awareness in body-oriented therapy (MABT). Compared with Active Exteroception, Active Interoception deactivated somatomotor and prefrontal regions. Greater self-reported interoceptive sensibility (MAIA scale) predicted sparing from deactivation within the anterior cingulate cortex (ACC) and left-lateralized language regions. The right insula, typically described as a primary interoceptive cortex, was only specifically implicated by its deactivation during an exogenously paced respiration condition (Active Matching) relative to self-paced Active Interoception. Psychophysiological interaction (PPI) analysis characterized Active Interoception as promoting greater ACC connectivity with lateral prefrontal and parietal regions commonly referred to as the dorsal attention network (DAN). In contrast to evidence relating accurate detection of liminal interoceptive signals such as the heartbeat to anterior insula activity, interoceptive attention toward salient signals such as the respiratory cycle may involve reduced cortical activity but greater ACC-DAN connectivity, with greater sensibility linked to reduced deactivation within the ACC and language-processing regions.

Direct machine learning reconstruction of respiratory variation waveforms from resting state fMRI data in a pediatric population

In many functional magnetic resonance imaging (fMRI) studies, respiratory signals are unavailable or do not have acceptable quality due to issues with subject compliance, equipment failure or signal error. In large databases, such as the Human Connectome Projects, over half of the respiratory recordings may be unusable. As a result, the direct removal of low frequency respiratory variations from the blood oxygen level-dependent (BOLD) signal time series is not possible. This study proposes a deep learning-based method for reconstruction of respiratory variation (RV) waveforms directly from BOLD fMRI data in pediatric participants (aged 5 to 21 years old), and does not require any respiratory measurement device. To do this, the Lifespan Human Connectome Project in Development (HCP-D) dataset, which includes respiratory measurements, was used to both train a convolutional neural network (CNN) and evaluate its performance. Results show that a CNN can capture informative features from the BOLD signal time course and reconstruct accurate RV timeseries, especially when the subject has a prominent respiratory event. This work advances the use of direct estimation of physiological parameters from fMRI, which will eventually lead to reduced complexity and decrease the burden on participants because they may not be required to wear a respiratory bellows.

Generating dynamic carbon-dioxide traces from respiration-belt recordings: Feasibility using neural networks and application in functional magnetic resonance imaging

INTRODUCTION

In the context of functional magnetic resonance imaging (fMRI), carbon dioxide (CO2) is a well-known vasodilator that has been widely used to monitor and interrogate vascular physiology. Moreover, spontaneous fluctuations in end-tidal carbon dioxide (PETCO2) reflects changes in arterial CO2 and has been demonstrated as the largest physiological noise source for denoising the low-frequency range of the resting-state fMRI (rs-fMRI) signal. However, the majority of rs-fMRI studies do not involve CO2 recordings, and most often only heart rate and respiration are recorded. While the intrinsic link between these latter metrics and CO2 led to suggested possible analytical models, they have not been widely applied.

METHODS

In this proof-of-concept study, we propose a deep-learning (DL) approach to reconstruct CO2 and PETCO2 data from respiration waveforms in the resting state.

RESULTS

We demonstrate that the one-to-one mapping between respiration and CO2 recordings can be well predicted using fully convolutional networks (FCNs), achieving a Pearson correlation coefficient (r) of 0.946 ± 0.056 with the ground truth CO2. Moreover, dynamic PETCO2 can be successfully derived from the predicted CO2, achieving r of 0.512 ± 0.269 with the ground truth. Importantly, the FCN-based methods outperform previously proposed analytical methods. In addition, we provide guidelines for quality assurance of respiration recordings for the purposes of CO2 prediction.

DISCUSSION

Our results demonstrate that dynamic CO2 can be obtained from respiration-volume using neural networks, complementing the still few reports in DL of physiological fMRI signals, and paving the way for further research in DL based bio-signal processing.

fMRI studies evaluating central respiratory control in humans

A plethora of neural centers in the central nervous system control the fundamental respiratory pattern. This control is ensured by neurons that act as pacemakers, modulating activity through chemical control driven by changes in the O2/CO2 balance. Most of the respiratory neural centers are located in the brainstem, but difficult to localize on magnetic resonance imaging (MRI) due to their small size, lack of visually-detectable borders with neighboring areas, and significant physiological noise hampering detection of its activity with functional MRI (fMRI). Yet, several approaches make it possible to study the normal response to different abnormal stimuli or conditions such as CO2 inhalation, induced hypercapnia, volitional apnea, induced hypoxia etc. This review provides a comprehensive overview of the majority of available studies on central respiratory control in humans.

Physiological noise modeling in fMRI based on the pulsatile component of photoplethysmograph

The blood oxygenation level-dependent (BOLD) contrast mechanism allows the noninvasive monitoring of changes in deoxyhemoglobin content. As such, it is commonly used in functional magnetic resonance imaging (fMRI) to study brain activity since levels of deoxyhemoglobin are indirectly related to local neuronal activity through neurovascular coupling mechanisms. However, the BOLD signal is severely affected by physiological processes as well as motion. Due to this, several noise correction techniques have been developed to correct for the associated confounds. The present study focuses on cardiac pulsatility fMRI confounds, aiming to refine model-based techniques that utilize the photoplethysmograph (PPG) signal. Specifically, we propose a new technique based on convolution filtering, termed cardiac pulsatility model (CPM) and compare its performance with the cardiac-related RETROICOR (Card-RETROICOR), which is a technique commonly used to model fMRI fluctuations due to cardiac pulsatility. Further, we investigate whether variations in the amplitude of the PPG pulses (PPG-Amp) covary with variations in amplitude of pulse-related fMRI fluctuations, as well as with the systemic low frequency oscillations (SLFOs) component of the fMRI global signal (GS - defined as the mean signal across all gray matter voxels). Capitalizing on 3T fMRI data from the Human Connectome Project, CPM was found to explain a significantly larger fraction of the fMRI signal variance compared to Card-RETROICOR, particularly for subjects with larger heart rate variability during the scan. The amplitude of the fMRI pulse-related fluctuations did not covary with PPG-Amp; however, PPG-Amp explained significant variance in the GS that was not attributed to variations in heart rate or breathing patterns. Our results suggest that the proposed approach can model high-frequency fluctuations due to pulsation as well as low-frequency physiological fluctuations more accurately compared to model-based techniques commonly employed in fMRI studies.

A practical modification to a resting state fMRI protocol for improved characterization of cerebrovascular function

Cerebrovascular reactivity (CVR), defined here as the Blood Oxygenation Level Dependent (BOLD) response to a CO₂ pressure change, is a useful metric of cerebrovascular function. Both the amplitude and the timing (hemodynamic lag) of the CVR response can bring insight into the nature of a cerebrovascular pathology and aid in understanding noise confounds when using functional Magnetic Resonance Imaging (fMRI) to study neural activity. This research assessed a practical modification to a typical resting-state fMRI protocol, to improve the characterization of cerebrovascular function. In 9 healthy subjects, we modelled CVR and lag in three resting-state data segments, and in data segments which added a 2–3 minute breathing task to the start of a resting-state segment. Two different breathing tasks were used to induce fluctuations in arterial CO₂ pressure: a breath-hold task to induce hypercapnia (CO₂ increase) and a cued deep breathing task to induce hypocapnia (CO₂ decrease). Our analysis produced voxel-wise estimates of the amplitude (CVR) and timing (lag) of the BOLD-fMRI response to CO₂ by systematically shifting the CO₂ regressor in time to optimize the model fit. This optimization inherently increases gray matter CVR values and fit statistics. The inclusion of a simple breathing task, compared to a resting-state scan only, increases the number of voxels in the brain that have a significant relationship between CO₂ and BOLD-fMRI signals, and improves our confidence in the plausibility of voxel-wise CVR and hemodynamic lag estimates. We demonstrate the clinical utility and feasibility of this protocol in an incidental finding of Moyamoya disease, and explore the possibilities and challenges of using this protocol in younger populations. This hybrid protocol has direct applications for CVR mapping in both research and clinical settings and wider applications for fMRI denoising and interpretation.

The Effect of CO2 on Resting-State Functional Connectivity: Isocapnia vs. Poikilocapnia

The normal variability in breath size and frequency results in breath-to-breath variability of end-tidal PCO₂ (P_ETCO₂), the measured variable, and arterial partial pressure of carbon dioxide (PaCO₂), the independent variable affecting cerebral blood flow (CBF). This study examines the effect of variability in PaCO₂ on the pattern of resting-state functional MRI (rs-fMRI) connectivity. A region of interest (ROI)-to-ROI and Seed-to-Voxel first-level bivariate correlation, hemodynamic response function (hrf)-weighted analysis for measuring rs-fMRI connectivity was performed during two resting-state conditions: (a) normal breathing associated with breath-to-breath variation in PaCO₂ (poikilocapnia), and (b) normal breathing with breath-to-breath variability of P_ETCO₂ dampened using sequential rebreathing (isocapnia). End-tidal PCO₂ (P_ETCO₂) was used as a measurable surrogate for fluctuations of PaCO₂. During poikilocapnia, enhanced functional connections were found between the cerebellum and inferior frontal and supramarginal gyrus (SG), visual cortex and occipital fusiform gyrus; and between the primary visual network (PVN) and the hippocampal formation. During isocapnia, these associations were not seen, rather enhanced functional connections were identified in the corticostriatal pathway between the putamen and intracalacarine cortex, supracalcarine cortex (SCC), and precuneus cortex. We conclude that vascular responses to variations in P_ETCO₂, account for at least some of the observed resting state synchronization of blood oxygenation level-dependent (BOLD) signals.

Using carpet plots to analyze transit times of low frequency oscillations in resting state fMRI

A "carpet plot" is a 2-dimensional plot (time vs. voxel) of scaled fMRI voxel intensity values. Low frequency oscillations (LFOs) can be successfully identified from BOLD fMRI and used to study characteristics of neuronal and physiological activity. Here, we evaluate the use of carpet plots paired with a developed slope-detection algorithm as a means to study LFOs in resting state fMRI (rs-fMRI) data with the help of dynamic susceptibility contrast (DSC) MRI data. Carpet plots were constructed by ordering voxels according to signal delay time for each voxel. The slope-detection algorithm was used to identify and calculate propagation times, or "transit times", of tilted vertical edges across which a sudden signal change was observed. We aim to show that this metric has applications in understanding LFOs in fMRI data, possibly reflecting changes in blood flow speed during the scan, and for evaluating alternative blood-tracking contrast agents such as inhaled CO₂. We demonstrate that the propagations of LFOs can be visualized and automatically identified in a carpet plot as tilted lines of sudden intensity change. Resting state carpet plots produce edges with transit times similar to those of DSC carpet plots. Additionally, resting state carpet plots indicate that edge transit times vary at different time points during the scan.