Cardiac q-space trajectory imaging by motion-compensated tensor-valued diffusion encoding in human heart in vivo

Abstract

Purpose: Tensor-valued diffusion encoding can probe more specific features of tissue microstructure than what is available by conventional diffusion weighting. In this work, we investigate the technical feasibility of tensor-valued diffusion encoding at high b-values with q-space trajectory imaging (QTI) analysis, in the human heart in vivo.

Methods: Ten healthy volunteers were scanned on a 3T scanner. We designed time-optimal gradient waveforms for tensor-valued diffusion encoding (linear and planar) with second-order motion compensation. Data were analyzed with QTI. Normal values and repeatability were investigated for the mean diffusivity (MD), fractional anisotropy (FA), microscopic FA (μFA), isotropic, anisotropic and total mean kurtosis (MKi, MKa, and MKt), and orientation coherence (C_c ). A phantom, consisting of two fiber blocks at adjustable angles, was used to evaluate sensitivity of parameters to orientation dispersion and diffusion time.

Results: QTI data in the left ventricular myocardium were MD = 1.62 ± 0.07 μm² /ms, FA = 0.31 ± 0.03, μFA = 0.43 ± 0.07, MKa = 0.20 ± 0.07, MKi = 0.13 ± 0.03, MKt = 0.33 ± 0.09, and C_c = 0.56 ± 0.22 (mean ± SD across subjects). Phantom experiments showed that FA depends on orientation dispersion, whereas μFA was insensitive to this effect.

Conclusion: We demonstrated the first tensor-valued diffusion encoding and QTI analysis in the heart in vivo, along with first measurements of myocardial μFA, MKi, MKa, and C_c . The methodology is technically feasible and provides promising novel biomarkers for myocardial tissue characterization.

Keywords: cardiac microstructure; diffusion tensor imaging; motion-compensated diffusion encoding; q-space trajectory imaging; tensor-valued diffusion encoding; tissue characterization.

FIGURE 1

Schematic microscopic (subvoxel) and macroscopic (voxel) tensors corresponding to underlying histology illustrated by layers of (left to right) coherently organized cardiomyocytes (e.g., in healthy myocardium), dispersed cardiomyocytes (e.g., in the presence of intravoxel rotation); multiple populations of cardiomyocytes (e.g., at the right ventricle insertion points); and randomly organized cardiomyocytes (e.g., in cardiomyocyte disarray). Histology slide (random configuration) adapted from Nichols et al. FA, fractional anisotropy.

FIGURE 2

(Left) Vertical cross section and 3D view of phantom consisting of two blocks of hollow fibers that are rotated from interblock angle = 0° (parallel; as shown) to 90° (perpendicular) with respect to one another (black arrows). Slice locations 1–3 are shown in yellow. (Top right) Scanning electron microscopy image of fiber cross section at ×1000 magnification. (Bottom right) Histogram of pore diameters with 3‐μm bins up to 30 μm. All pores with diameters greater than 30 μm were grouped in a single bin, which accounts for the high area fraction.

FIGURE 3

(Top to bottom) Effective diffusion‐encoding gradient waveforms for b‐tensor encoding, zeroth to third‐order gradient moments; and encoding power spectrum for (left to right) planar b‐tensor encoding (PTE), linear b‐tensor encoding (LTE) (long t_d), and LTE (short t_d) corresponding to intermediate, low, and high mean frequencies, respectively. Waveforms were compensated for up to second‐order motion, and residual M₃ can be seen at the end of the waveforms (see inset). Encoding power spectrum (in hertz) is defined as the Fourier transform of q(t). There is a clear distinction in the encoding power spectrum between LTE (long t_d) and LTE (short t_d), while PTE contains contributions from both LTE waveforms. The three colors (red, gray, and blue) correspond to orthogonal directions. Other sequence features were omitted for clarity. Waveforms were generated using a modified NOW toolbox.

FIGURE 4

Dependence of mean diffusivity (MD; black lines), FA (black lines), and microscopic FA (μFA; red lines) on interblock angle and (left to right) slice positions 1 to 3 (mean ± SD across voxels). Diffusion times (ms) given for pulsed gradient spin‐echo (PGSE) experiments, and mean frequencies of the power spectrum of q(t) given for PGSE and q‐space magic angle spinning (qMAS) experiments. MD increases and FA decreases as the diffusion time is decreased. FA decreases with increasing interblock angle in Slice 2, simulating increased fiber orientation dispersion. In contrast, μFA remains relatively insensitive to interblock angle, highlighting the strength of μFA as a metric of anisotropy that is not biased by orientation dispersion.

FIGURE 5

(Top) Normalized signal‐attenuation curves showing divergence between linear (red; b _Δ = 1) and planar (black; b _Δ = −0.5) tensor encoding, particularly at higher b‐values. Data are reported in an example region of interest (ROI) in the myocardium of 1 healthy volunteer. Signals were averaged across the ROI and used to fit a covariance model. Data were acquired in five shells and plotted with offsets in b for better visualization of data and fitting. (Middle) Powder average of PTE and LTE log signal images, sorted by b‐value. (Bottom) Relative difference images between PTE and LTE show that contrast generally increases with b‐value. MKa, anisotropic mean kurtosis; MKi, isotropic mean kurtosis.

FIGURE 6

(Top) Example cardiac q‐space trajectory imaging (cQTI) maps and (bottom) values in a middle short‐axis slice in the left‐ventricular (LV) wall (N = 10). MD (in μm²/ms), FA, μFA, MKa, MKi, total mean kurtosis (MKt), and microscopic orientation coherence (C _c) shown. Mean ± SD are reported across subjects. The measurements were more consistent across the septal, anterior, and lateral walls. In the inferior wall, higher MD and lower FA, μFA, and MKa were observed, which may be associated with susceptibility effects near the posterior vein.

FIGURE 7

Bland–Altman plots showing agreement in cQTI between Scans 1 and 2 within ROIs. Mean difference ± 1.96 SD are given by black and red lines, respectively (N = 9). These data can serve as a benchmark for future improvement, and can be valuable when planning future studies.

FIGURE 8

MD and FA histograms for LTE (Long t_d) and LTE (Short t_d). Mean MD = [1.61, 1.50] μm²/ms and mean FA = [0.37, 0.32], respectively (vertical lines). Data were pooled across subjects (N = 9) and ROIs.