Session Item

The apparent diffusion coefficient obtained from diffusion-weighted imaging (DWI) is a promising biomarker for early response assessment of head and neck (H&N) cancer. Non-Gaussian Intravoxel Incoherent Motion (NG-IVIM) DWI, provides additional information on perfusion and restricted (intra-cellular) diffusion, potentially interesting for response assessment. However, precise estimation of the NG-IVIM DWI parameters may come with considerable scan times. Therefore, the goal of this study was to find the optimal tradeoff between scan time and parameter precision for NG-IVIM DWI in the H&N region and provide the corresponding optimal set of b-values.

The signal S_b of a NG-IVIM DWI scan at b-value b in a voxel is given by S_b = S₀(1-f) exp(-bD+(1/6)(bD)²K)+S₀f exp(-bD^*), where S₀ is the signal at b=0 s/mm². The NG-IVIM DWI parameters of interest (f the perfusion fraction, D* the pseudo-diffusion coefficient, D the diffusion coefficient and K the kurtosis, related to restricted diffusion) can be obtained by fitting a multi b-value acquisition. The number and choice of b-values strongly influences the parameter precision. Therefore, as a first step, the b-values for a set size varying between 5 and 30 b-values were optimized using a cost function with 3 components: (1) Cramér-Rao lower bound as a measure of precision, (2) echo time (TE) correction for signal decrease at higher TE, and (3) a constraint to avoid b-values below 0 and above 1500 s/mm².

Next, b-values from all optimized sets were combined into one acquisition (b=1x0, 2x10, 1x20, 6x80, 1x130, 1x280, 1x570, 1x640, 2x750, 1x760, 2x770, 2x780, 5x790, 8x1500 s/mm²), which was imaged twice in 16 healthy volunteers in one session on a GE MR450w 1.5T scanner. A 20 b-value reference set from literature (b=(0,10a¹,…,10a^(n-1), a=1.3, n=20)) was imaged as well. Distortion correction, intra- and inter-volume registration were applied to all acquisitions. DWI parameters were fitted for each voxel in tonsil ROIs using an in-house fitting script. The importance of parameter optimization was quantified using the SD of the DWI parameters. The effect of number of b-values (i.e. scan time) was expressed by the SD of the differences of the mean in the tonsils (repeatability).

Parameter precision of the optimized set of 20 b-values was significantly better compared to the 20 b-value reference set in D (factor 1.5),  f (factor 1.7), and K (factor 2.3) (Figure 1). Yet, the precision of D* was decreased (factor 0.7). A gradual increase in parameter precision was observed with increasing set size from 5 to 15 b-values for all parameters. In general, parameter precision did not improve by using more than 15 b-values (Table 1).

 

The b-value set of 15 images (b=1x0,1x10,2x80,1x130,1x570,2x770,2x780,1x790,4x1500 s/mm²) yielded the optimal tradeoff between scan time and parameter precision. Its precision was comparable to the larger b-value sets, while scan time was reduced by 50% from 12 to 6 minutes in our experiments.