Linear and Sigmoidal Cerebrovascular Reactivity in fMRI and Transcranial Doppler Ultrasound
Poster No:
1501
Submission Type:
Abstract Submission
Authors:
Genevieve Hayes1, Sierra Sparks1, Joana Pinto1, Daniel Bulte1
Institutions:
1University of Oxford, Oxford, Oxfordshire
First Author:
Co-Author(s):
Introduction:
Cerebrovascular reactivity (CVR) describes the brain's ability to regulate blood flow in response to vasoactive stimuli such as CO2, reflecting cerebrovascular health and adaptability. CVR is most commonly evaluated using a 2-point measure: at baseline and during a stimulus (e.g., breath-hold, gas-inhalation). However, this approach does not allow for non-linear evaluation of the response (Bhogal et al., 2014), complicating comparisons between common CVR imaging modalities. In this context, we present an analysis of dynamic CVR to a ramp stimulus using transcranial Doppler ultrasound (TCD) and blood oxygen level dependent (BOLD) MRI.
Methods:
Dynamics of the cerebral blood flow response to a ramped breathing protocol were assessed in 25 healthy participants (13F, aged 33±11 years) using TCD and BOLD MRI. An overview of the acquisition and analysis pipeline is presented in Fig. 1. Both modalities assessed the cerebrovascular response to a ramp gas protocol (3 cycles of 5 deep breaths, 30s of air, 40s of air + 5% CO2, and 40s of air + 10% CO2. Blood flow velocity in the right middle cerebral artery (MCAv) was measured continuously using a clinical TCD (Doppler-BoxX, DWL) in a first session. Participants completed the same protocol imaged with 3T BOLD MRI (Siemens Prisma scanner, TR/TE=0.8s/0.03s, 2.4mm isotropic FOV) in a second session.
Data processing and analysis was performed using FSL and custom Python scripts. The end-tidal peaks in the CO2 time-course were selected automatically. The BOLD MRI images were preprocessed with motion correction, spatial smoothing (FWHM=4mm), B0 unwarping, fieldmap correction, and high pass filtering (275s). The mean BOLD signal was calculated from active voxels of the right parietal lobe using the MNI152 brain atlas. For the TCD signal, a rolling mean of the MCAv was used to smooth the pulsatile signal. To account for delays, a bulk shift was applied to each PETCO2 trace to maximise its cross-correlation with the mean MCAv or BOLD signal. MCAvmean was normalised relative to the baseline period to correct for probe angle variations.
To characterise CVR, both linear and sigmoidal models (equations in Fig. 1) were fit to the PETCO2 vs. normalised MCAvmean, and PETCO2 vs. %BOLD signal. In the sigmoid model, the minimum blood flow, a, and span, d, parameters were fixed (normalised MCAvmean: a=0.2, d=2.5; %BOLD: a=-8, d=18). Parameters 1/b, the slope of the linear regime, and c, the PETCO2 value for the inflection point were fit to the data for each subject and modality independently. Correlation was tested using the Pearson correlation coefficient (p<0.05 significance) and r-value correlation coefficient (r<0.4 weak, 0.4≤r<0.7 moderate, r≥0.7 strong correlation).
Data processing and analysis was performed using FSL and custom Python scripts. The end-tidal peaks in the CO2 time-course were selected automatically. The BOLD MRI images were preprocessed with motion correction, spatial smoothing (FWHM=4mm), B0 unwarping, fieldmap correction, and high pass filtering (275s). The mean BOLD signal was calculated from active voxels of the right parietal lobe using the MNI152 brain atlas. For the TCD signal, a rolling mean of the MCAv was used to smooth the pulsatile signal. To account for delays, a bulk shift was applied to each PETCO2 trace to maximise its cross-correlation with the mean MCAv or BOLD signal. MCAvmean was normalised relative to the baseline period to correct for probe angle variations.
To characterise CVR, both linear and sigmoidal models (equations in Fig. 1) were fit to the PETCO2 vs. normalised MCAvmean, and PETCO2 vs. %BOLD signal. In the sigmoid model, the minimum blood flow, a, and span, d, parameters were fixed (normalised MCAvmean: a=0.2, d=2.5; %BOLD: a=-8, d=18). Parameters 1/b, the slope of the linear regime, and c, the PETCO2 value for the inflection point were fit to the data for each subject and modality independently. Correlation was tested using the Pearson correlation coefficient (p<0.05 significance) and r-value correlation coefficient (r<0.4 weak, 0.4≤r<0.7 moderate, r≥0.7 strong correlation).
·Fig.1: Data acquisition and analysis pipeline
Results:
24/25 subjects were included in the analysis, with 1 excluded due to noisy TCD data. Comparison plots between the TCD and BOLD MRI CVR fitting parameters are presented in Fig. 2.
Across all subjects, the mean change in PETCO2 from the bottom of the ramp to peak hypercapnia was 27.5±3.9mmHg in TCD and 28.1±4.8mmHg in MRI. The mean baseline PETCO2 during the TCD and MRI was 39.5±3.9mmHg and 43.2±3.6mmHg respectively. The average MCAvmean across subjects ranged from 79±11% to 170±20% of baseline and the average BOLD signal span was 7.1±1.3%.
Comparing the CVR between the two sessions, the PETCO2 vs. MCAvmean and PETCO2 vs. BOLD showed significant correlation (p=0.022, r=0.467). The slope parameter of the sigmoidal fit had the strongest correlation between methods (p=0.014, r=0.493). The inflection point of the sigmoid fit also showed statistical significance in correlation between modalities (p=0.038, r=0.426).
Across all subjects, the mean change in PETCO2 from the bottom of the ramp to peak hypercapnia was 27.5±3.9mmHg in TCD and 28.1±4.8mmHg in MRI. The mean baseline PETCO2 during the TCD and MRI was 39.5±3.9mmHg and 43.2±3.6mmHg respectively. The average MCAvmean across subjects ranged from 79±11% to 170±20% of baseline and the average BOLD signal span was 7.1±1.3%.
Comparing the CVR between the two sessions, the PETCO2 vs. MCAvmean and PETCO2 vs. BOLD showed significant correlation (p=0.022, r=0.467). The slope parameter of the sigmoidal fit had the strongest correlation between methods (p=0.014, r=0.493). The inflection point of the sigmoid fit also showed statistical significance in correlation between modalities (p=0.038, r=0.426).
·Fig. 2: Parameter comparison plots between TCD and MRI fitting
Conclusions:
The CVR response to a simple ramp PETCO2 protocol showed a significant within-subject correlation when comparing linear and sigmoidal CVR using TCD and MRI. Features extracted from the sigmoid model, including the slope and inflection point warrant further investigation as they may inform on cerebrovascular health beyond a linear fit alone.
Modeling and Analysis Methods:
Methods Development 1
Novel Imaging Acquisition Methods:
BOLD fMRI
Physiology, Metabolism and Neurotransmission:
Cerebral Metabolism and Hemodynamics 2
Keywords:
Cerebral Blood Flow
Data analysis
Experimental Design
FUNCTIONAL MRI
Modeling
MRI
ULTRASOUND
1|2Indicates the priority used for review