A novel nonlinear analysis of blood flow dynamics applied to the human lung.

The spatial/temporal dynamics of blood flow in the human lung can be measured noninvasively with magnetic resonance imaging (MRI) using arterial spin labeling (ASL). We report a novel data analysis method using nonlinear prediction to identify dynamic interactions between blood flow units (image voxels), potentially providing a probe of underlying vascular control mechanisms. The approach first estimates the linear relationship (predictability) of one voxel time series with another using correlation analysis, and after removing the linear component, it estimates the nonlinear relationship with a numerical mutual information approach. Dimensionless global metrics for linear prediction (FL) and nonlinear prediction (FNL) represent the average amplitude of fluctuations in one voxel estimated by another voxel, as a percentage of the global average voxel flow. A proof-of-principle test of this approach analyzed experimental data from a study of high-altitude pulmonary edema (HAPE), providing two groups exhibiting known differences in vascular reactivity. Subjects were mountaineers divided into HAPE-susceptible (S, n = 4) and HAPE-resistant (R, n = 5) groups based on prior history at high altitudes. Dynamic ASL measurements in the lung in normoxia (N, [Formula: see text] = 0.21) and hypoxia (H, [Formula: see text] = 0.13 ± 0.01) were compared. The nonlinear prediction metric FNL decreased with hypoxia [7.4 ± 1.3 (N) vs. 6.3 ± 0.7 (H), P = 0.03] and was significantly different between groups [7.4 ± 1.2 (R) vs. 6.2 ± 14.1 (S), P = 0.03]. This proof-of-principle test demonstrates that this nonlinear analysis approach applied to ASL data is sensitive to physiological effects even in small subject cohorts, and it potentially can be used in a wide range of studies in health and disease in the lung and other organs.NEW & NOTEWORTHY Pulmonary blood flow varies both spatially and temporally. We describe a novel technique to uncover nonlinear interactions in dynamic images of pulmonary blood flow measured using MRI, based on mutual information between the flow fluctuations in different imaging voxels. In a proof-of-principle study, we show that the proposed metric of nonlinear interactions was sufficiently sensitive to detect a difference between small cohorts of subjects susceptible to high-altitude pulmonary edema (HAPE) and subjects resistant to HAPE.