Poster R9 - Evaluation of a Novel 3D Diffusion-Weighted MRI Sequence for Improved Breast Characterization

Introduction: Diffusion-weighted imaging (DWI) has become a highly valuable MR technique in the assessment of breast cancer with many institutions incorporating DWI sequences into their clinical breast imaging protocols. Single-shot EPI (ss-EPI) sequences are the reference standard for DW breast images mainly due to their short acquisition times and robustness to subject motion. However, a significant drawback of DW-EPI is its susceptibility to magnetic field inhomogeneities, which can cause geometric distortions and incomplete fat suppression in the resulting images. In this work we present a novel 3D sequence that reduces artifacts in breast DW images by using a fast spin-echo readout, variable-density sampling and a phase correction algorithm. We show preliminary results of our Variable Density Cartesian Acquisition with Spiral Profile Reordering (VD-CASPR) sequence in the breast of three healthy volunteers.

Materials and

Methods: In VD-CASPR, each Ky-Kz profile (or readout in the Kx direction) is acquired following a pseudo spiral trajectory on a cartesian grid. At the beginning of each echo train, a user-defined number of profiles in the center of k-space is consistently sampled. The remaining ETL – N profiles are then sampled following a pseudo-spiral trajectory that is calculated based on a golden-step angle algorithm. At the end of the acquisition, we will effectively have a fully sampled outer shell of k-space and an oversampled inner shell of k-space. This oversampled region increases the overall SNR of the sequence and also serves as navigators to correct for phase errors that result from the multi-shot acquisition.

K-space data were exported to MATLAB (Mathworks Inc., Natick, Massachusetts) for phase correction, which was previously validated on phantoms. For our algorithm, we created two datasets: the first containing the undersampled k-space acquired in each shot, and the second containing the center-most profiles that are sampled across all echo trains. Both datasets were Fourier transformed and then the phase of the center-only dataset was subtracted from the undersampled dataset in image space. Finally, the phase-corrected data were averaged across shots to generate the final phase-corrected images.

Results, Conclusions, and Discussions: VD-CASPR images suffered from severe void artifacts and blurring, which were significantly improved once the phase correction algorithm was applied [Figure 2]. Additionally, VD-CASPR with phase correction was able to achieve improved image quality with high SNR at different b-values in both axial and sagittal orientations [Figures 3, 4]. Furthermore, VD-CASPR with phase correction showed excellent fat suppression at b = 0 s/mm2 and adequate fat suppression at b = 200, 500 s/mm2. In comparison, single shot DW-EPI had significant geometric distortions in axial and sagittal images. DW-TSE images did not suffer from geometric distortions but exhibited incomplete fat suppression near tissue interfaces and longer scan times compared to VD-CASPR, even with parallel imaging reconstruction. Lastly, VD-CASPR was able to achieve significantly reduced slice thickness in sagittal acquisitions, which could be highly beneficial for tumor detection and segmentation.

This work presents the application of our novel 3D sequence for volumetric diffusion weighted imaging in the breast. Phase corrected VD-CASPR achieved high quality images at low and intermediate b-values (b = 200, 500 s/mm2) that were robust to B0 inhomogeneities and phase induced errors. VD-CASPR was the diffusion sequence that most closely aligned to the T2-weighted acquisition. This highly valuable feature would allow for perfect co-registration with T2 and T1-weighted images for tumor characterization. Future work will explore additional reconstruction steps to improve image quality at higher b-values (b = 800 s/mm2), where noise is substantially increased.

Acknowledgements (Optional): This work was supported by NIH/NCI grants U01CA207091, R01CA283663, and CPRIT RP190049. The authors would like to thank all volunteers for their participation in this study.