Quantitative mapping of cerebrovascular reactivity amplitude and delay with breath-hold BOLD fMRI when end-tidal CO 2 quality is low

Rebecca G. Clements, Kristina M. Zvolanek, Neha Reddy, Kimberly J. Hemmerling, Roza G. Bayrak, Catie Chang, Molly G. Bright

Abstract

Cerebrovascular reactivity (CVR) is a critical indicator of brain health, measuring the ability of cerebral blood vessels to dilate in response to vasoactive stimuli like carbon dioxide ($CO_2$). While breath-holding tasks are commonly used in fMRI to assess CVR, obtaining accurate end-tidal $CO_2$ ($P_{ET}CO_2$) measurements via nasal cannula requires high patient compliance, which is often lacking in clinical populations. This study explores a deep learning-based "rescue" pipeline that predicts $P_{ET}CO_2$ from robust respiration belt measurements (RVT) and rescales the output to standard units (%BOLD/mmHg). Our results demonstrate that this method produces CVR amplitude and delay maps with high absolute agreement to ground truth, even in cases of cerebrovascular pathology like Moyamoya disease.

Introduction

CVR mapping serves as a "stress test" for the brain's vasculature. By measuring the BOLD fMRI response to changes in arterial $CO_2$, clinicians can identify regions of impaired blood flow regulation. Standardizing these measurements into units of %BOLD/mmHg is essential for cross-subject comparisons and longitudinal tracking. However, the gold-standard measurement of $P_{ET}CO_2$ is notoriously difficult to obtain; it requires participants to breathe exclusively through their nose and perform specific exhalations. When these requirements are not met, the resulting $CO_2$ data is often too noisy for quantitative analysis.

Previous efforts have used Respiration Volume per Time (RVT) as a surrogate, but RVT lacks standard units and often fails to capture the precise temporal dynamics necessary for mapping CVR delay. This work addresses these gaps by combining deep learning with a novel rescaling strategy.

Methods

Deep Learning Pipeline

The researchers developed a 1D Fully Convolutional Network (FCN) trained to transform RVT signals into predicted $P_{ET}CO_2$ waveforms. Unlike models trained on resting-state data, this FCN was specifically optimized for breath-holding tasks. A specialized "peak-focused" loss function was employed to ensure the model accurately captured the hypercapnic peaks, which are the most critical segments for CVR calculation.

The "One-Trial" Calibration

To convert the arbitrary units of RVT or predicted $CO_2$ into standard mmHg, the authors introduced a rescaling method. They hypothesized that if a participant can provide just one high-quality breath-hold trial (a single clean exhalation) during a scan, that segment can serve as a "gold standard" to calibrate the entire session's predicted waveform. This significantly lowers the compliance threshold for patients.

Experimental Design

The model was trained and validated using a diverse set of breath-holding protocols to ensure generalizability. The datasets included both fixed-duration and randomized breath-hold tasks.

Summary of experimental task parameters for the four datasets used in model training and validation.

Results

Accuracy and Reliability

The predicted $P_{ET}CO_2$ regressor showed exceptional performance, achieving a mean Fisher’s z-transformed correlation of 1.14 with ground-truth measurements. In terms of CVR amplitude, the median error in gray matter was only 0.037 %BOLD/mmHg—a level of precision sufficient to distinguish between different age groups or clinical states.

Sensitivity to Pathology

The method's clinical utility was tested in a patient with Moyamoya disease, characterized by narrowed intracranial arteries and significant CVR delays. The predicted $CO_2$ regressor successfully identified regions of extreme vascular delay with a Dice similarity coefficient (DSC) of 0.66 compared to ground truth. In contrast, the raw RVT-based method failed to accurately map these delays, yielding a DSC of only 0.16.

Discussion

This study demonstrates that deep learning can effectively bridge the gap between robust but non-quantitative respiration metrics (RVT) and the precise $P_{ET}CO_2$ data required for clinical CVR mapping. By capturing the temporal "shape" of the $CO_2$ response more accurately than raw RVT, the FCN provides superior maps of CVR delay, which is often a more sensitive marker of disease than amplitude alone.

While the current model was trained on healthy individuals, future work will focus on expanding the training set to include patients with chronic respiratory or cardiovascular conditions. Ultimately, this "rescue" pipeline could be integrated into real-time fMRI monitoring, allowing technicians to save datasets that would otherwise be discarded due to poor sensor quality.

Summary updated

Abstract

Cerebrovascular reactivity (CVR), the ability of cerebral blood vessels to dilate or constrict in response to a vasoactive stimulus, is a clinically useful measure of cerebrovascular health. CVR is often measured using a breath-hold task to modulate blood CO<sub>2</sub> levels during an fMRI scan. Measuring end-tidal CO<sub>2</sub> (P<sub>ET</sub>CO<sub>2</sub>) with a nasal cannula during the task allows CVR amplitude to be calculated in standard units (vascular response per unit change in CO<sub>2</sub>, or %BOLD/mmHg) and CVR delay to be calculated in seconds. The use of standard units allows for normative CVR ranges to be established and for CVR comparisons to be made across subjects and scan sessions. Although breath holding can be successfully performed by diverse patient populations, obtaining accurate P<sub>ET</sub>CO<sub>2</sub> measurements requires additional task compliance; specifically, participants must breathe exclusively through their nose and exhale immediately before and after each breath hold. Meeting these requirements is challenging, even in healthy participants, and this has limited the translational potential of breath-hold fMRI for CVR mapping. Previous work has focused on using alternative regressors such as respiration volume per time (RVT), derived from respiration-belt measurements, to map CVR. Because measuring RVT does not require additional task compliance from participants, it is a more feasible measure than P<sub>ET</sub>CO<sub>2</sub>. However, using RVT does not produce CVR amplitude in standard units. In this work, we explored how to achieve CVR amplitude maps, in standard units, and CVR delay maps, when breath-hold task P<sub>ET</sub>CO<sub>2</sub> data quality is low. First, we evaluated whether RVT could be scaled to units of mmHg using a subset of P<sub>ET</sub>CO<sub>2</sub> data of sufficiently high quality. Second, we explored whether a P<sub>ET</sub>CO<sub>2</sub> timeseries predicted from RVT using deep learning allows for more accurate CVR measurements. Using a dense-mapping breath-hold fMRI dataset, we showed that both rescaled RVT and rescaled, predicted P<sub>ET</sub>CO<sub>2</sub> can be used to produce maps of CVR amplitude in standard units and CVR delay with strong absolute agreement to ground-truth maps. The rescaled, predicted P<sub>ET</sub>CO<sub>2</sub> regressor resulted in superior accuracy for both CVR amplitude and delay. In an individual with regions of increased CVR delay due to Moyamoya disease, the predicted P<sub>ET</sub>CO<sub>2</sub> regressor also provided greater sensitivity to pathology than RVT. Ultimately, this work will increase the clinical applicability of CVR in populations exhibiting decreased task compliance.
BibTeX
@inproceedings{clements2025quantitative,
  author = {Rebecca G. Clements and Kristina M. Zvolanek and Neha Reddy and Kimberly J. Hemmerling and Roza G. Bayrak and Catie Chang and Molly G. Bright},
  title = {{Quantitative mapping of cerebrovascular reactivity amplitude and delay with breath-hold BOLD fMRI when end-tidal CO <sub>2</sub> quality is low}},
  booktitle = {Imaging},
  year = {2025},
  doi = {10.1162/imag\_a\_00536},
}
Figure