Background: Functional MRI is ubiquitously used to study post-stroke recovery. However, the fMRI-derived haemodynamic responses are vulnerable to vascular insult which can result in reduced magnitude and temporal delays (lag) in the haemodynamic response function (HRF). The aetiology of HRF lag remains controversial, and a better understanding of it is required to ensure accurate interpretation of post-stroke fMRI studies. In this longitudinal study, we investigate the relationship between haemodynamic lag and cerebrovascular reactivity (CVR) following stroke.
Methods: Voxelwise lag maps were calculated relative to a mean grey matter reference signal for 27 healthy controls and 59 patients with stroke across two timepoints (~2 weeks and ~4 months post-stroke), and two conditions: resting-state and breath-holding. The breath-holding condition was additionally used to calculate CVR in response to hypercapnia. HRF lag was computed for both conditions across tissue compartments: lesion, perilesional tissue, unaffected tissue of the lesioned hemisphere, and their homologue regions in the unaffected hemisphere. CVR and lag maps were correlated. Group, condition, and time effects were assessed using ANOVA analyses.
Results: Compared with the average grey matter signal, a relative haemodynamic lead was observed in the primary sensorimotor cortices in resting-state and bilateral inferior parietal cortices in breath-holding condition. Whole-brain haemodynamic lag was significantly correlated across conditions irrespective of group, with regional differences across conditions suggestive of a neural network pattern. Patients showed relative lag in the lesioned hemisphere which significantly reduced over time. Breath-hold derived lag and CVR had no significant voxel-wise correlation in controls, or patients within the lesioned hemisphere or the homologous regions of the lesion and perilesional tissue in the right hemisphere (mean r<0.1).
Conclusion: The contribution of altered CVR to HRF lag was negligible. We suggest that HRF lag is largely independent of CVR, and could partly reflect intrinsic neural network dynamics amongst other factors.