The Effect of Stimulus Timing and Hemodynamic Delay on Measured BOLD fMRI Cerebrovascular Reactivity

2611 

Abstract Submission 

Rebecca Clements¹, Molly Bright¹

¹Northwestern University, Chicago, IL

Rebecca Clements  
Northwestern University
Chicago, IL

Molly Bright  
Northwestern University
Chicago, IL

Cerebrovascular reactivity (CVR), the response of blood vessels to a vasoactive stimulus, is an important indicator of cerebrovascular health. CVR is often measured using BOLD fMRI combined with breath-holds (BHs) or CO2 inhalation. During the scan, end-tidal CO2 (PETCO2) is measured and used as a regressor in a general linear model to calculate CVR in units of %BOLD/mmHg. However, due to measurement and physiological delays, there is a spatially varying offset in timing between the recorded PETCO2 and the BOLD response [6]. Despite several methods to account for this offset and improve CVR estimations [1,7,9], many existing CVR studies do not correct for delay [8,10]. The validity of these studies is not well understood, particularly in patient groups with prolonged delays. This work aims to use simulated and real fMRI data to understand how stimulus timing and hemodynamic delay affect CVR accuracy when delay is not corrected.

BOLD fMRI data were simulated for 10, 15, and 20s repeated BHs alternating with 30s rest periods using a time-scaled ramp design convolved with the canonical HRF (Fig1A). BOLD fMRI data were also simulated for 20, 40, 60, and 80s of hypercapnia induced using gas inhalation, alternating with similar length rest periods, by convolving a block design with the HRF (Fig1B). Simulated timeseries, which were approximately 8.5 minutes total (TR=1s), were shifted from 0-28s in 2s increments. Missing values were replaced by the mean of the simulated dataset. CVR was calculated in each simulated dataset using a general linear model, with unshifted data as the independent variable (representing the PETCO2 regressor) and shifted data as the dependent variable (representing the fMRI time series), both of which were downsampled to a 2s TR to better represent a typical fMRI TR. In these "ideal" simulations, CVR should be 1 (arbitrary units), as variance in the simulated data is fully attributed to simulated CO2 changes without scaling. Percent error of the CVR estimate was calculated for each simulated CO2 protocol and shift.

Real 18s repeated BH fMRI data for 15 healthy adults (25±5y, 5M) were used to better understand how delay affects CVR accuracy in noisier data (see [4] for methods). Both uncorrected and delay corrected CVR maps were calculated for each scan using phys2cvr [6,7]. Using FSL's Harvard-Oxford cortical atlas in native space, average uncorrected CVR, delay-corrected CVR, delay, and percent error of uncorrected CVR were calculated within gray matter in each cortical parcel(Fig2A).

15s repeated BH fMRI data were also acquired in a 31yo male with unilateral Moyamoya disease (occluded right MCA; see [2]). Uncorrected and delay-corrected CVR maps, and the percent error of the CVR calculation for each cortical parcel (separated by L and R hemispheres), were calculated.

For all simulations, as the shift (delay) increased, the CVR percent error also increased (Fig1C). As the length of the hypercapnia period decreased, CVR percent error increased at a faster rate. This suggests that longer periods of hypercapnia mitigate, but do not fully account for, the effects of delay on CVR accuracy when delay is not corrected. Features of real data, such as task-correlated motion, could further increase CVR error.

In real fMRI data, as the absolute value of delay increased, the number of parcels with CVR percent error>20% increased substantially (Fig2B). Similarly, in the participant with Moyamoya disease, the parcels in the affected right hemisphere had greater delay values and CVR errors (Fig2C). This supports the simulation results, suggesting that as hemodynamic delay increases, the importance of CVR delay correction also increases.

Delay correction is essential for accurate CVR results, even when long periods of hypercapnia are induced. Measuring hemodynamic delay not only leads to more accurate CVR measurements, but could also provide a complementary biomarker for cerebrovascular diseases [3,5].

BOLD fMRI ²

Cerebral Metabolism and Hemodynamics ¹

Cerebral Blood Flow

Cerebrovascular Disease

Data analysis

Design and Analysis

Experimental Design

FUNCTIONAL MRI

Modeling