Combined arterial spin labeling and diffusion-weighted imaging for noninvasive estimation of capillary volume fraction and permeability-surface product in the human brain

Abstract

A number of two-compartment models have been developed for the analysis of arterial spin labeling (ASL) data, from which both cerebral blood flow (CBF) and capillary permeability-surface product (PS) can be estimated. To derive values of PS, the volume fraction of the ASL signal arising from the intravascular space (v(bw)) must be known a priori. We examined the use of diffusion-weighted imaging (DWI) and subsequent analysis using the intravoxel incoherent motion model to determine v(bw) in the human brain. These data were then used in a two-compartment ASL model to estimate PS. Imaging was performed in 10 healthy adult subjects, and repeated in five subjects to test reproducibility. In gray matter (excluding large arteries), mean voxel-wise v(bw) was 2.3±0.2 mL blood/100 g tissue (all subjects mean±s.d.), and CBF and PS were 44±5 and 108±2 mL per 100 g per minute, respectively. After spatial smoothing using a 6-mm full width at half maximum Gaussian kernel, the coefficient of repeatability of CBF, v(bw) and PS were 8 mL per 100 g per minute, 0.4 mL blood/100 g tissue, and 13 mL per 100 g per minute, respectively. Our results show that the combined use of ASL and DWI can provide a new, noninvasive methodology for estimating v(bw) and PS directly, with reproducibility that is sufficient for clinical use.

Figure 1

Example of model fitting in a single voxel. (A) dM map in one subject, shown here at TI=1.8 seconds, after registration to MNI (Montreal Neurological Institute) standard space. The white circle shows the approximate location of the voxel (enlarged for clarity). Plots of the intravoxel incoherent motion (IVIM) (B) and arterial spin labeling (ASL) (C) data from this voxel. Filled circles=measured data, solid black lines=fitted data, gray dashed lines=95% confidence interval from wild bootstrap analysis (see Results section).

Figure 2

Overview of the segmentation process. (A) T₁ map, produced using data from the multi-TI non-selective, non-background suppressed arterial spin labeling (ASL) data. (B) Relative arterial volume fraction (rV_a) map at the same location. (C) Data from the T₁ and rV_a map were used to segment the gray matter (gray voxels), while excluding regions that are likely to contain large arteries (white voxels). All data are shown after registration to MNI (Montreal Neurological Institute) standard space.

Figure 3

(A) Mean fitted value of v_bw (vascular volume fraction) as a function of input v_bw value from the diffusion-weighted imaging (DWI) Monte Carlo simulations, at signal-to-noise ratio (SNR)=120. (B) Variation of mean fitted permeability-surface product (PS) as a function of input PS value from the arterial spin labeling (ASL) Monte Carlo simulations, at SNR=10, with v_bw=2 mL blood/100 g tissue (mean value from in vivo data). In both plots, filled circles represent the mean fitted value after 1,000 iterations, and error bars represent s.d. The dashed lines represent input value=fitted value. (C) 2D precision plot of fitted PS values. Color indicates SD of the fitted PS value over 1,000 Monte Carlo simulations, as a function of both input v_bw value (horizontal axis) and input PS value (vertical axis). Typical parameters found in human gray matter at 1.5 T were used to generate all synthetic data (Table 2). The color reproduction of this figure is available on the Journal of Cerebral Blood Flow and Metabolism journal online.

Figure 4

(Top row) Maps of fitted cerebral blood flow (CBF) (left), blood volume fraction (v_bw) (middle), and permeability-surface product (PS) (right) in a representative subject, before segmentation and spatial smoothing. (Bottom row) Voxel-wise repeatability index (RI) across five subjects in the same slice, after segmentation of white matter and large arteries.