Optimizing b‐values schemes for diffusion MRI of the brain with segmented Intravoxel Incoherent Motion (IVIM) model

Abstract

To define an optimal set of b-values for accurate derivation of diffusion MRI parameters in the brain with segmented Intravoxel Incoherent Motion (IVIM) model. Methods: Simulations of diffusion signals were performed to define an optimal set of b-values targeting different perfusion regimes, by relying on an optimization procedure which minimizes the total relative error on estimated IVIM parameters computed with a segmented fitting procedure. Then, the optimal bvalues set was acquired in vivo on healthy subjects and skull base chordoma patients to compare the optimized protocol with a clinical one. Results: The total relative error on simulations decreased of about 40% when adopting the optimal set of 13 b-values (0 10 20 40 50 60 200 300 400 1200 1300 1400 1500 s/mm 2 ), showing significant differences and increased precision on D and f estimates with respect to simulations with a non-optimized b-values set. Similarly, in vivo acquisitions demonstrated a dependency of IVIM parameters on the b-values array, with differences between the optimal set of b-values and a clinical non-optimized acquisition. IVIM parameters were compatible to literature values, with D (0.679/0.701 [0.022/0.008] ⋅10 −3 mm 2 /s), f (5.49/5.80 [0.70/1.14] %), and D* (8.25/7.67 [0.92/0.83] ⋅10 −3 mm 2 /s) median [interquartile range] estimates for white matter/gray matter in volunteers and D (0.

Evaluation of Whole Brain Intravoxel Incoherent Motion (IVIM) Imaging

“…In many IVIM-related papers, this technique is used [ 10 , 12 , 29 , 31 ]. Another option for improving SNR is using multiple excitations for a single voxel, as mentioned in [ 16 ]. In all these works, the sizes of the voxels are much larger than in this paper (i.e., 1.8 × 1.8 × 5 mm [ 29 ], 1.2 × 1.2 × 4 mm [ 10 ], 2.75 × 2.75 × 5 mm [ 12 ], 1.14 × 1.14 × 4 mm [ 16 ], and 1.5 × 1.5 × 6 mm [ 32 ]) and have a nonzero slice gap or are limited to a lower number of slices, both resulting in worse segmentation capabilities.…”

“…In all these works, the sizes of the voxels are much larger than in this paper (i.e., 1.8 × 1.8 × 5 mm [ 29 ], 1.2 × 1.2 × 4 mm [ 10 ], 2.75 × 2.75 × 5 mm [ 12 ], 1.14 × 1.14 × 4 mm [ 16 ], and 1.5 × 1.5 × 6 mm [ 32 ]) and have a nonzero slice gap or are limited to a lower number of slices, both resulting in worse segmentation capabilities. Although in most publications level is presented to be in the range of 10–30 for , some publications state is in the range of 35–53 on -value 1000 [ 14 ] or even [ 16 ], which corresponds with , which is very unlikely to achieve in any scanning. To achieve precise results at a high level, one of the works’ total acquisition time was 52 min [ 12 ].…”

“…A MUSE sequence was used to ensure the maximum available SNR level [ 15 ]. While planning an experiment, we tried selecting b -values that were accessible on the scanner and were close to an optimized set [ 16 ]. Under these conditions, the GE MR 3T Pioneer scanner with 21 channel head/neck coil was able to acquire images for 10 -values (1200, 1000, 700, 500, 200, 120, 80, 50, 20, and 10) measured in three orthogonal directions and five (no diffusion weighing).…”

Evaluation of Whole Brain Intra Voxel Incoherent Motion (IVIM) Imaging

“…In many IVIM related papers such technique is used [10,12,28,31]. Another option for improving SNR is using multiple excitations for single Voxel as mentioned in [15]. In all these works sizes of the voxels are much higher than in this paper (i. e. 1,8x1,8x5 mm [28], 1,2x1,2x4 mm [10], 2,75x2,75x5mm [12], 1.14x1.14x4mm [15], 1.5x1.5x6 mm [32]) and have nonzero slice gap or are limited to lesser number of slices, both resulting in worse segmentation capabilities.…”

“…To ensure maximum available SNR level a MUSE sequence was used [14]. While planning an experiment we tried selecting b-values accessible on the scanner, close to an optimized set [15]. Under these conditions GE MR 3T Pioneer scanner with 21 channel head/neck coil was able to acquire images for 10 𝑏-values (1200 1000 700 500 200 120 80 50 20 10) measured in three orthogonal directions and five 𝑏 (no diffusion weighing).…”