Diffusion-time (td)-dependent diffusion kurtosis imaging (tDKI) [[1], [2], [3]] is a promising diffusion MRI (dMRI) method for measuring td-dependent changes in the non-Gaussian diffusion components of dMRI signals [[4], [5], [6]], enabling the simultaneous assessment of microstructural size and membrane integrity. We previously demonstrated that these novel markers are sensitive to cellular-level pathological changes following ischemic brain injury and can help identify subacutely salvageable tissue [7]. The transmembrane water exchange time (τex) quantified by tDKI has also been proposed as a marker of cellular metabolism [8], associated with increased metabolic activity in cancer [1,9,10], and is thought to be mediated primarily by the aquaporin channels on the cell membrane. The estimation of τex relies on the Kärger model, which requires a sufficiently long td to ensure that each compartment is fully coarse-grained [5,11]. For long-td scans, diffusion-weighted stimulated echo acquisition mode (DW-STEAM) [12] offers advantages over the more commonly used pulsed-gradient spin-echo (PGSE) dMRI, because the STEAM signal decays with T1 relaxation rather than T2 relaxation during the mixing time that determines the td [13].
Although DW-STEAM provides desirable signal-to-noise ratio (SNR) for measurements at long td, the butterfly gradients, which consist of slice-selective and crusher gradients, introduce additional diffusion weighting during the signal restoration phase. Moreover, this unwanted diffusion weighting varies at different tds, leading to bias in tDKI measurements. Previous studies using DW-STEAM [14,15] have typically followed standard PGSE practice and have neglected this effect. The impact of unwanted diffusion weighting on dMRI studies manifests in two major aspects. First, the additional diffusion weighting changes the effective diffusion direction [16], thereby compromising the experimental design. Some studies [[17], [18], [19]] calculated the actual b-values retrospectively, which resulted in different effective b-values at different tds, introducing biases for microstructural model fitting. Other studies [3,13,20] used compensatory acquisitions, loading corrected gradient vectors into the scanner to eliminate bias. Unfortunately, these studies did not account for the cross-term diffusion weighting during gradient correction; instead, they corrected the cross-terms in post-processing using the b-matrix [21]. The second issue arises from unwanted different weighting in the reference measurements or b0 images, which is often omitted or inaccurately reported in many studies [[22], [23], [24], [25], [26]] (i.e., as b = 0 ms/μm2 or ‘non-diffusion weighted’). This can lead to underestimation of the apparent diffusion coefficient (ADC) and result in artificial td dependence, as confirmed in both in vivo human brain [27] and postmortem tissue [28]. Additionally, unlike in conventional diffusion MRI with PGSE or OGSE, where a non-diffusion-weighted (b = 0) acquisition is used as the reference scan by default, no such standard is established for STEAM-based acquisitions. This lack of a unified protocol has led to different implementations of the reference scan across studies [1,16,29], making it unclear how these variations affect tDKI estimation. Due to the limitations of DW-STEAM, its application must be carefully considered [6].
In this study, we first proposed an iterative method to optimize the diffusion gradients (not a post processing method), to correct the diffusion-direction shift caused by the butterfly gradient. Secondly, we demonstrated in both a water phantom and in vivo that ignoring the diffusion weighting in reference measurements leads to underestimation of diffusivity and kurtosis, especially at long td, and therefore introduced bias in tDKI data. Therefore, we proposed using low-b-value data as the reference and remove the crushers in both reference measurements and DWIs, which we found to be an optimal approach to prevent underestimation compared with previous correction strategies, in both in vivo and ex vivo experiments.
Comments (0)