Inter-subject variability in hypercapnic normalization of the BOLD fMRI response.

In the application of hypercapnic normalization to functional magnetic resonance imaging (fMRI) studies, the blood oxygenation level dependent (BOLD) response to a functional stimulus is typically divided by the BOLD response to a hypercapnic challenge. While some prior studies have shown that hypercapnic normalization can reduce inter-subject BOLD variability, other studies have found an increase in inter-subject variability. In this study we used measures of baseline cerebral blood flow (CBF) and the functional BOLD and CBF responses to both visual stimuli and hypercapnia to assess the effect of hypercapnic normalization on inter-subject variability. We found that the functional and hypercapnic BOLD and CBF responses all exhibited a significant inverse dependence on baseline CBF. In contrast, the maximum BOLD response was independent of baseline CBF and was not a major source of inter-subject BOLD variability. Division of the functional BOLD response by the hypercapnic BOLD response increased inter-subject variability in the normalized responses as compared to the original responses, reflecting the presence of a systematic bias term that was inversely dependent on the hypercapnic BOLD response. This systematic bias resulted from a positive intercept term in the linear relationship between the functional and hypercapnic BOLD responses. This positive intercept term reflected a steeper inverse dependence of the hypercapnic CBF response on baseline CBF, as compared to the functional CBF response. In contrast to the results obtained with normalization based on division, normalized responses obtained by using the hypercapnic BOLD response as a covariate were unaffected by the systematic bias and exhibited reduced inter-subject variability. The findings of this study indicate that the positive intercept in the linear relationship between functional and hypercapnic BOLD responses should be carefully considered in the hypercapnic normalization of BOLD fMRI data.