Introduction

Interleaved block-segmented DTI (iblocks-DTI)¹ has been recently proposed to achieve high-resolution DTI, with 1) further reduced distortion artifacts and 2) more robust phase variation correction at high magnetic fields or with high b-values than the conventional interleaved DTI. However, iblocks-DTI needs to acquire five patch patterns for covering a complete k-space with increased scan time, making it less practical for clinical use. Hence, view-sharing iblocks-DTI (VSiblocks-DTI)² was proposed to achieve reasonable scan time by sharing peripheral blocks among neighboring diffusion directions and decreasing the number of required patch patterns. However, VSiblocks-DTI requires a smooth ordering of diffusion directions to maintain the similarity of diffusion contrast among the shared k-space blocks. In previous study, VSiblocks-DTI has only been assessed with a smooth ordering of diffusion directions and a matrix size equal to 128×128. The effects of matrix sizes and orderings of diffusion directions on VSiblocks-DTI performance have not been investigated. In this study, we assess the performance of previously proposed VSiblocks-DTI with different orderings of diffusion-weighted directions and three matrix sizes. Moreover, we propose a novel least difference block sharing (LDBS) method for VSiblocks-DTI, to enable robust VSiblocks-DTI under different conditions.

Materials and methods

The least difference block sharing (LDBS) scheme for VSiblocks-DTI
Figure 1a demonstrates the data acquisition and sharing scheme in the previously proposed VSiblocks-DTI. For the data acquisition, only one target pattern including the central block (pattern #1, #2, or #3) was acquired for each diffusion direction and pattern #4 or #5 was acquired once every three diffusion directions to provide necessary peripheral high-frequency data. Afterward, for each diffusion direction, the unacquired peripheral blocks were filled with patterns from neighboring diffusion directions. This previous sharing method is referred to as neighbor sharing in the following content. In the proposed least difference block sharing (LDBS) scheme (Figure 1b), DWI images were first reconstructed with neighbor sharing for calculating the image difference values in the next step. Then, for each diffusion direction, unacquired peripheral blocks were filled with patterns shared from a diffusion direction providing the least image difference value from the current direction, thereby minimizing the differences between the shared blocks and the desired correct data.
Experiments
Hybrid simulations were performed to investigate the performance of VS iblocks-DTI with either neighbor sharing or LDBS under different conditions. Firstly, two sets of four-shot interleaved DTI data with 64 diffusion directions were acquired on a 1.5 T MRI scanner using a 19-channel coil with the following scan parameters: TE/TR = 90.5(full-echo)/4000 ms, FOV = 24x24 cm², matrix size = 192×192, and b-value = 750 s/mm². One set was acquired with a smooth ordering of diffusion directions³ (Figure 2a) while the other one with a randomized ordering (Figure 2b). Second, central k-space data with 3 different sizes: 1) 128×128, 2) 160×160, and 3) 192×192 were obtained from these data to simulate the data acquired with different matrix sizes. MUSE⁴ reconstruction results of respective four-shot interleaved data served as the gold standard for calculating DTI tensor. Afterward, 6 sets of VS iblocks-DTI data (three different matrix sizes and two different orderings of diffusion directions) were generated and reconstructed by POCSMUSE⁵. DTI tensor calculations from neighbor sharing and LDBS were quantitatively evaluated with three error indexes: 1) angular deviation of the principal eigenvector V1_error, 2) percentage fractional anisotropy deviation FA_error, and 3) percentage mean diffusivity deviation MD_error.

Results and Discussion

First, colored FA maps, quantitative tensor error maps and values in Figures 2, 3, 4, and 5 showed that LDBS provided better DTI fiber visualization and smaller quantitative tensor errors than neighbor sharing under all six different conditions. Second, results in Figures 2, 3, and 5 demonstrated that the DTI tensors produced by neighbor sharing possessed larger errors with smaller matrix sizes, suggesting that the reduced size of central blocks could affect reconstruction performances. Furthermore, for neighbor sharing results, tensor calculations with a randomized ordering of diffusion directions showed larger errors than those with a smooth ordering. Therefore, both matrix sizes and the ordering of diffusion directions could reduce the performance of VSiblocks-DWI with neighbor sharing. In comparison, with LDBS, tensor calculation errors under all six different conditions were very subtle (V1_error < 3.27 degrees; FA_error < 4.37%; MD_error < 1.24%), and no obvious differences in tensor errors among the six different conditions were observed (maximum difference in V1_error: 0.46 degrees; FA_error: 0.58%; MD_error: 0.24%). Thus, we conclude that the proposed LDBS is a robust sharing method for VSiblocks-DTI to provide accurate DTI tensor calculation, therefore benefiting further applications of VSiblocks-DTI and the development of high-resolution DTI techniques.

Acknowledgements

The work was in part supported by grants from Hong Kong Research Grant Council (GRF17106820, GRF17125321, GRF14206723, and ECS24213522).