Intravoxel incoherent motion (IVIM) perfusion imaging has been shown to be applicable in clinical brain examinations, but those images are known to be noisy. To better quantify the necessary conditions to produce homogenous IVIM perfusion images in the brain, we studied the properties of signal noise as function of b-value and of localization in the brain. We compared the image quality of the perfusion maps as function of number of average to the maximal quality IVIM perfusion maps obtained during a 1 hour acquisition time.
The behavior of the distribution of the signal intensities as function of b-value, showed an overall decrease in standard deviation with increased b-value (Figure 1), Interestingly, the behavior of the shape of the distribution was very different in GM compared to WM, with a much larger dependence of kurtosis and average skewness on b-value in GM compared to WM. In GM, the kurtosis first increased with higher b-values between 0 and 600 s/mm2, to drop again above 600 s/mm2 while also the average skewness dropped sharply between 0 and 600 s/mm2. As expected, the standard deviation dropped as function of number of averages, with the biggest drop observed between 1 average and 5 averages. In proportion to the signal mean value, the Coefficient of Variation (CV) for all b values was below 20% in GM and below 10% in WM, with 1, 5, 10, and 20 repetitions.
In both white and gray matter, the signal intensity distributions of each voxels was deviating from the Gaussian distribution, with the largest observed at b=900 s/mm2 (D’agostio test for normality, Figure 2 a). The signal-to-noise ratio (SNR) for each b-value increased as function of number of averages, and was the increase was the most significant in the first 5 averages (Figure 2 b). Interestingly, the slope of increase was steeper in GM compared to WM.
The diffusion coefficient parameteric maps were homogenous with a single acquisition, and did not improve significantly with increased number of averages. The images homogeneity of the parameteric maps of f and D* improved gradually with the number of signal averages, and this improvement was the most significant between average 1 to 5 (Figure 3).
The Figure 4 shows difference to our standard 20 average maps. The most prominent change in maximum difference was found to be achieved with 2-3 averages.
1 Federau C, O'Brien K, Meuli R, Hagmann P, Maeder P (2014) Measuring brain perfusion with intravoxel incoherent motion (IVIM): initial clinical experience. J Magn Reson Imaging 39:624-632
2 Federau C, Sumer S, Becce F et al (2014) Intravoxel incoherent motion perfusion imaging in acute stroke: initial clinical experience. Neuroradiology 56:629-635
3 Suo S, Cao M, Zhu W et al (2016) Stroke assessment with intravoxel incoherent motion diffusion-weighted MRI. NMR Biomed 29:320-328
4 Yao Y, Zhang S, Tang X et al (2016) Intravoxel incoherent motion diffusion-weighted imaging in stroke patients: initial clinical experience. Clin Radiol 71:938 e911-936
5 Smith SM, Jenkinson M, Woolrich MW et al (2004) Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage 23:S208-219
6 King DE (2009), Dlib-ml: A machine learning toolkit, J Mach Learn Res 10: 1755-1758