Voxelwise optimization of hemodynamic lags to improve regional CVR estimates in breath-hold fMRI

Cerebrovascular Reactivity (CVR), the responsiveness of blood vessels to a vasodilatory stimulus, is an important indicator of cerebrovascular health. Assessing CVR with fMRI, we can measure the change in the Blood Oxygen Level Dependent (BOLD) response induced by a change in CO2 pressure (%BOLD/mmHg). However, there exists a temporal offset between the recorded CO2 pressure and the local BOLD response, due to both measurement and physiological delays. If this offset is not corrected for, voxel-wise CVR values will not be accurate. In this paper, we propose a framework for mapping hemodynamic lag in breath-hold fMRI data. As breath-hold tasks drive task-correlated head motion artifacts in BOLD fMRI data, our framework for lag estimation fits a model that includes polynomial terms and head motion parameters, as well as a shifted variant of the CO2 regressor (±9 s in 0.3 s increments), and the hemodynamic lag at each voxel is the shift producing the maximum total model R2 within physiological constraints. This approach is evaluated in 8 subjects with multi-echo fMRI data, resulting in robust maps of hemodynamic delay that show consistent regional variation across subjects, and improved contrast-to-noise compared to methods where motion regression is ignored or performed earlier in preprocessing.