Robust estimation of dynamic cerebrovascular reactivity using breath-holding fMRI: application in diabetes and hypertension

Abstract

Breath-holding (BH) tasks during functional magnetic resonance imaging (fMRI) acquisitions are gaining popularity for non-invasive mapping of carbon-dioxide (CO2) driven cerebrovascular reactivity (CVR), which is a valuable clinical marker of vascular function. However, compliance to BH tasks is often unclear, and the ability to record end-tidal CO2often limited, rendering the optimal analysis of BH fMRI data a challenge. In this work, we demonstrate an adaptive data-driven approach for estimating CVR from BH fMRI data that minimizes errors due to subject non-compliance and regional CVR time delay variability. Building on previous work, we propose a frequency-domain-based approach for CVR estimation without the need for end-tidal CO2(PETCO2) recordings. CVR amplitude is estimated in units of %ΔBOLD directly from the data-driven BH frequency. Serious deviations from the designed task paradigm were suppressed and thus did not bias the estimated CVR values. We demonstrate our method in detecting regional CVR amplitude and time-lag differences in a group of 56 individuals, consisting of healthy (CTL), hypertensive (HT) and diabetic-hypertensive (DM+HT) groups of similar ages and sex ratios. The CVR amplitude was lowest in HT+DM, and HT had a lower CVR amplitude than CTL regionally but the voxelwise comparison did not yield statistical significance. Notably, we demonstrate that the voxelwise CVR time delay estimated in Fourier domain is a more sensitive marker of vascular dysfunction than CVR amplitude. While HT+DM seems to confer longer CVR delays, HT seems to confer shorter delays than CTL. These are the first MRI-based observations of CVR time delay differences between diabetic-hypertensive patients and healthy controls. These results demonstrate the feasibility of extracting CVR amplitude and CVR time delay using BH challenges without PETCO2recordings, and the unique clinical value of CVR time-delay information.

In search of a unifying theory of white matter aging: associations with age in perfusion detectable in advance of microstructure

“…CBF, in contrast, appears to decline contemporaneously with microstructure. The superior sensitivity of vascular timing information to pathology is consistent with recent reports on the unique sensitivity of vascular-response latency (D. Kim et al, 2021;McKetton et al, 2019;Nanayakkara et al, 2023). A notable limitation to these findings is that, while CBF declined quadratically in all tracts by the UNC, the predominantly linear declines in ATT restricted the extent of possible CCF comparisons.…”

“…Those axons with reduced or malformed myelin sheaths are then more vulnerable to age-related culling of vulnerable WM fibers, and associated WM regions at higher risk of pathological declines (Kohama et al, 2012;Tang et al, 1997). Related to perfusion quantification, vascular response latency has been demonstrated as a more sensitive marker of pathology than response amplitude (D. Kim et al, 2021;McKetton et al, 2019;Nanayakkara et al, 2023), so it remains unclear whether perfusion amplitude or timing is an earlier marker of possible WM degeneration.…”

Individual variability in the relationship between physiological and resting-state fMRI metrics

“…Indeed, previous comparisons of CVR estimated in RS or during a gas challenge showed a few differences between the two, notably the presence of an undershoot in spontaneous CVR (and the lack of one in induced CVR), and a small difference in CVR magnitude (Prokopiou, Pattinson, Wise & Mitsis, 2019). While previous studies reporting a modest relationship between CVR and RSF suggest to interpret CVR and RSF as two different aspects of vascular changes, only partially sharing similar characteristics (Bailes et al, 2023;Chen et al, 2023;Lipp et al, 2015;Liu et al, 2013), further studies addressing specific aspects of vascular responses to different CO 2 stimuli (or the lack of) are required. Yet, using CVR estimates obtained from CO 2 recordings during the RS run could be an alternative approach to using BH-induced CVR estimates in specific pathological cases where compliance to a task is impossible to obtain, e.g.…”

“…Golestani et al, 2016; Lipp et al, 2015; Murphy, Harris & Wise, 2011; Pinto, Jorge, Sousa, Vilela & Figueiredo, 2016; Zvolanek et al, 2023) Moreover, improved modelling and denoising can improve CVR estimation as well, for instance by adopting advanced acquisition techniques and lag-based CVR estimations (cfr. Donahue et al, 2016; Moia et al, 2021), deep learning methods to extract CVR (Hou et al, 2023), or, as more recently proposed, algorithms leveraging analyses in the frequency domain, rather than in the time domain (Nanayakkara, Meusel, Anderson & Chen, 2023; Xu et al, 2023).…”

“…ALFF and RSFA have both been proposed and adopted as low-cost approximations of CVR, due to their significant cross-sectional relationship with induced CVR (Chen et al, 2023;Golestani et al, 2016;Kannurpatti et al, 2014) or the within-subject(s), within-session spatial agreement of the two measures (De Vis, Bhogal, Hendrikse, Petersen & Siero, 2018;Kannurpatti & Biswal, 2008;Kannurpatti et al, 2011;Liu et al, 2013). However, this relationship is a debated topic in the field, due to contrasting evidence from different studies or settings.…”