Automatic Quality Control for Resting-State BOLD-Based Cerebrovascular Reactivity Mapping

Cerebrovascular reactivity (CVR) mapping based on resting-state BOLD fMRI can be widely available for research of vascular health not only in clinical studies but also in open databases. However, as it utilizes spontaneous CO₂ fluctuations of blood as endogenous stimuli, resting-state CVR may be prone to low SNR and reproducibility if the CO₂ fluctuation of an individual is small. The automatic identification of such poor-quality CVR datasets is crucial for large-scale research. Thus, in this work, we developed an automatic quality control algorithm for resting-state CVR mapping. Utilizing a total of 51 resting-state CVR maps acquired with three scanning protocols in each healthy participant, quality control parameters reflecting common characteristics of poor-quality CVR, including pooled variance of different tissue types, proportion of negative voxels in gray matter, and the sensitivity of the BOLD signal to CVR, were extracted and then combined into one comprehensive quality evaluation index (QEI). We further evaluated its performance by leave-one-out cross-validation and correlation analyses with test–retest reproducibility. Leave-one-out cross-validation showed that QEI was significantly correlated with the reference standard of quality evaluation in all left-out cases (r = 0.766). Correlation analyses with test–retest reproducibility revealed significant positive correlations between the worse QEI and similarity index of CVR maps from two tests (r = 0.809, 0.890, 0.396, and 0.654 for data from four open databases). The proposed QEI performed not only in good agreement with visual inspection but can also adapt in resting-state CVR from multiple age groups and scanning protocols, paving the way for the clinical applications of resting-state CVR mapping technology.