Enhanced efficiency and sensitivity in detection of acute ischemic brain injury using fast diffusion kurtosis imaging
Yin Wu1,2, Jinsuh Kim3, Suk-Tak Chan1, Iris Yuwen Zhou1, Yingkun Guo1, Takahiro Igarashi1, Hairong Zheng2, Gang Guo4, and Phillip Zhe Sun1

1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Charlestown, MA, United States, 2Paul C. Lauterbur Research Centre for Biomedical Imaging, Shenzhen Key Laboratory for MRI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, People's Republic of, 3Department of Radiology, University of Illusions at Chicago, Chicago, IL, United States, 4Department of Radiology, Xiamen 2nd Hospital, Xiamen, China, People's Republic of


We systematically compare the diagnostic value of conventional tensor-based DKI and recently proposed fast DKI protocols using an acute stroke rodent model. The measures and volumes of diffusion and kurtosis lesions were in good agreement between the two DKI methods. Importantly, contrast-to-noise ratio (CNR) of mean kurtosis using the fast DKI protocol was significantly higher than that of the routine method with its CNR efficiency approximately doubles. Therefore, our results demonstrated excellent performance of the fast DKI protocol in characterizing acute ischemic tissue injury, which may facilitate translation of the fast DKI approach in the acute stroke setting.


Diffusion kurtosis imaging (DKI) can offer a useful complementary tool to routine diffusion MRI for improved stratification of tissue damage in acute ischemic stroke 1. However its relatively long imaging time has hampered its clinical applications. Recently proposed fast DKI approach substantially shortens the imaging time 2, but its sensitivity for imaging acute stroke has yet been fully described. In this study, we compared the conventional tensor-model DKI and fast DKI methods by means of the contrast-to-noise ratio (CNR) and CNR efficiency to elucidate its diagnostic value in the acute stroke setting.


MRI: Eleven rats were induced unilateral stroke with a standard intraluminal MCAO procedure and imaged with a 4.7 T MR scanner 60 minutes after the procedure. Five slices (slice thickness/gap = 1.8/0.2 mm) were acquired with a single-shot EPI sequence. Imaging parameters include: FOV = 20x20 mm2, matrix size = 48x48, diffusion duration/diffusion time = 6/20 ms, TR/TE = 2500/36.6 ms, one reference image of b = 0 s/mm2, NSA = 4. For the conventional DKI protocol, two b-values of 1000 and 2500 s/mm2 were applied in fifteen diffusion directions. The scan time was 5 min and 10 s. For the fast DKI protocol, three images of b = 1000 s/mm2 were applied along gradient directions of (1,0,0), (0,1,0) and (0,0,1), and nine images of b = 2500 s/mm2 along diffusion directions of $$$\widehat{n}^{(1)}=(1,0,0)^T$$$, $$$\widehat{n}^{(1+)}=(0,1,1)^T$$$ and $$$\widehat{n}^{(1-)}=(0,1,-1)^T$$$, and similarly for i =2 and 3. Note that the superscript i in $$$\widehat{n}^{(i)}$$$ labels the position of the “1”, while in $$$\widehat{n}^{(i+)}$$$ and $$$\widehat{n}^{(i-)}$$$ it labels the position of the “0”. The fast DKI scan time was 2 min and 10 s.

Image analysis: For conventional DKI, fractional anisotropy (FA), axial (D) and radial (D) diffusivity, axial (K) and radial (K) kurtosis, and tensor-based MDtensor and MKtensor were obtained using DKE. For the fast DKI, MDfast was calculated as the mean of MDx,y,z as 3: $$$MD\scriptsize x,y,z \normalsize = \frac{(b_{1}+b_{3})D_{x,y,z}^{(12)}-(b_{1}+b_{2})D_{x,y,z}^{(13)}}{b_{3}-b_{2}}$$$, where $$$D_{x,y,z}^{(ij)}=\frac{lnS(b_i)/S(0)-lnS(b_j)/S(0)}{b_j-b_i}$$$, i = 1, j = 2, 3, and b1 = 0, b2 = 1000, and b3 = 2500 s/mm2. Furthermore, MKfast was obtained from 2: $$ MK_{fast}=\frac{\frac{6}{15}[\sum_{i=1}^3ln\frac{S(b_3,\widehat{n}^{(i)})}{S(0)}+2\sum_{i=1}^3ln\frac{S(b_3,\widehat{n}^{{(i+)}})}{S(0)}+2\sum_{i=1}^3ln\frac{S(b_3,\widehat{n}^{(i-)})}{S(0)}]+6\cdot b_3\cdot MD_{fast}}{b_3^2\cdot MD_{fast}^2}.$$

Image segmentation: The ischemic lesion was defined by thresholding at two standard deviations (SD) below the baseline MD of the contralateral normal brain. The reference ROI was designated by mirroring the segmented lesion into the contralateral brain. The MKtensor, MDfast and MKfast lesions were similarly defined.

Data Analysis: The CNR and CNR efficiency were calculated as 4: $$$CNR=(S_{ischemia}-S_{contralateral})/{\sqrt{{(\sigma_{ischemia}^2+\sigma_{contralateral}^2)}/2}}$$$ and $$$CNR/\sqrt{scan time}$$$. Two-tailed Student’s t-test was performed between paired measurements in ipsilateral ischemic and contralateral normal regions. Measurement differences across the ROIs were tested using one-way ANOVA with Bonferroni correction.


Ischemic regions showed significant decrease in diffusion indices (MDtensor: 32.1±2.1%, D: 28.4±2.1%, and D: 34.9±2.3%), and increase of FA (22.5±4.5%) and kurtosis indices (MKtensor: 47.1±7.3%, K: 50.6±6.2%, and K: 37.4±7.6%) compared to those in the contralateral regions. The magnitude of percentage changes were significantly higher in MKtensor and K than other indices. In addition, the difference between the percentage changes of MKtensor and K was insignificant. Fig. 1 compares MD and MK maps between the two DKI methods. MD and MK lesions were overlaid on a diffusion-weighted image and were respectively colored in red and green, exhibiting noticeable mismatch. MKfast map shows significantly higher CNR between the reference and ischemic regions (1.6±0.2), compared to that of MKtensor map (1.3±0.2). The acquisition time of the fast DKI method was about 50% shorter than that of the conventional DKI protocol, leading to 1.9 times higher gain of CNR efficiency. Lesion sizes of MD and MK are highly correlated between the two DKI methods (Fig. 2). Multi-slice and 3D volume rendering of typical ischemic diffusion and kurtosis lesions derived from the fast DKI method shows MKfast lesion significantly smaller than MDfast lesion, with their normalized lesion area ratio being 0.75±0.11. Although MDfast values were significantly reduced in all ischemic lesions from the contralateral normal area, differences of MDfast values across the three lesions were statistically insignificant (Table 1). In comparison, MKfast across the three lesion areas was significantly elevated from the normal area, with significant difference of MKfast across the three lesion areas.


Our study recapitulated that mean kurtosis is one of the most sensitive parameters to detect acute stroke injury, and demonstrated the enhanced sensitivity and efficiency of the fast DKI method for stratification of ischemic tissue injury during acute stroke. The fast DKI method provides important information for resolving heterogeneous DWI lesion, particularly translatable in the acute stroke setting.


National Basic Research Program of China (2015CB755500), NSFC (81571668 and 81471721), Shenzhen Science and Technology Program (JCYJ20140610151856743), NIH/NINDS (1R21NS085574 and 1R01NS083654).


[1] Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K. Diffusional kurtosis imaging: The quantification of non-gaussian water diffusion by means of magnetic resonance imaging. Magn Reson Med 2005;53(6):1432-1440.

[2] Hansen B, Lund TE, Sangill R, Jespersen SN. Experimentally and computationally fast method for estimation of a mean kurtosis. Magn Reson Med 2013;69(6):1754-1760.

[3] Jensen JH, Helpern JA. MRI quantification of non-Gaussian water diffusion by kurtosis analysis. NMR Biomed 2010;23(7):698-710.

[4] Sun PZ, Lu J, Wu Y, Xiao G, Wu R. Evaluation of the dependence of CEST-EPI measurement on repetition time, RF irradiation duty cycle and imaging flip angle for enhanced pH sensitivity. Phys Med Biol 2013;58:N229-N240.


Fig. 1 Comparison of MD and MK maps between the two DKI methods. Lesions of MD and MK were overlaid on a diffusion-weighted image and colored in red and green, respectively. Regions with concurrent diffusivity and kurtosis abnormalities were shown in deep green, and regions with only kurtosis abnormality were in light green. The unit of diffusivity is μm2/ms.

Fig. 2 Correlation of lesion volume size of (a) mean diffusion and (b) mean kurtosis determined from the conventional and fast DKI methods. The linear regression is shown as a solid line, and the identity line shown as a dash-dotted line.

Fig. 3 Demonstration of fast DKI in an acute stroke rat. (a) Multi-slice diffusion and kurtosis lesion mismatch from fast DKI. (b) 3D illustration of mean diffusion/mean kurtosis lesions and their mismatch in a representative acute stroke rat using the fast DKI method.

Table 1. Summary of mean diffusion and kurtosis values in the contralateral normal area and three distinctive lesion areas obtained using the fast DKI method.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)