2267
Accelerated 3D isotropic multi-echo UTE using CG-SENSE and Deep Learning Reconstruction for CT-like Imaging & Ultrashort T2* Mapping
Hung Phi Do1, Bekku Mitsuhiro2, Dawn Berkeley1, Brian Tymkiw1, Wissam AlGhuraibawi1, and Mo Kadbi1
1Canon Medical Systems USA, Inc., Tustin, CA, United States, 2Canon Medical Systems Corporation, Otawara, Japan
1Canon Medical Systems USA, Inc., Tustin, CA, United States, 2Canon Medical Systems Corporation, Otawara, Japan
Synopsis
Keywords: Whole Joint, Bone
Motivation: MRI provides superior soft-tissue contrast, however, cortical bone and ultrashort-T2* tissues are invisible in routine clinical MRI. Multi-echo Ultrashort TE (mecho-UTE) may allow visualization and quantification of bone and ultrashort-T2* tissues, however, its acquisition time is often long.
Goal(s): We used CG-SENSE and Deep Learning Reconstruction (CG-SENSE+DLR) to accelerate mecho-UTE, making it clinically feasible.
Approach: 5-min, 3-min, and 2-min mecho-UTE scans were prospectively acquired and reconstructed with Gridding and CG-SENSE+DLR. Resolution, sharpness, and T2* were measured and compared.
Results: CG-SENSE+DLR allows 2- and 3-min mecho-UTE with improved resolution and sharpness compared to 5-min mecho-UTE with Gridding reconstruction.
Impact: Routine MRI and mecho-UTE enable comprehensive MSK imaging, providing soft-tissue contrast and visualization-and-quantification of bone and ultrashort-T2* tissues. CG-SENSE+DLR allows accelerated 2- and 3-minunte mecho-UTE, which would enable widespread clinical utilization to demonstrate the value of mecho-UTE compared to CT.
Introduction
MRI provides superior soft-tissue contrast compared to CT, however, in musculoskeletal (MSK) conditions, where bone assessment is needed, a CT scan is often required. Single-echo Zero TE (ZTE) and Ultrashort TE (UTE) have the potential for radiation-free imaging of cortical bone, however, poor resolution, long scan time, and fundamental tradeoffs between scan time, SNR, and resolution[1] need to be addressed before wider adoption for clinical evaluation.In addition to CT-like contrast, dual-echo and multi-echo UTE can provide visualization of ultrashort-T2* tissues (tendon, ligament, meniscus, etc.) and quantification of bone porosity index[2] and quantitative T2* of ultrashort T2* tissues[3]–[7], which correlated well with CT bone density measurements and ultrashort-T2* tissue health, respectively. While providing more information compared to single-echo counterpart, multi-echo UTE is even more demanding in terms of scan time requirements.
As seen in Figure 1, routine FSE2D and mecho-UTE would allow comprehensive MSK imaging by providing excellent soft-tissue contrast and visualization/quantification ultrashort-T2* tissues (cortical bone, tendon, meniscus, ligament, etc.), respectively. This work aimed to use CG-SENSE[8] and Deep Learning Denoising Reconstruction[9], [10] (CG-SENSE+DLR) to accelerate and improve the resolution and sharpness of multi-echo UTE (mecho-UTE), making it viable for clinical adoption. Our goal is not to replace CT, but rather to enable a radiation-free bone imaging option for patients who are CT-contraindicated and need radiation-exposure minimization (for example, repeated CT exams, pediatric patients, etc.).
Methods
Data Collection:Experiments were performed on four subjects (four knees and one shoulder) using a 3T scanner with approved IRB. 5-min, 3-min, and 2-min 3D-isotropic 0.8mm3 [reconstructed as 0.4mm3] resolution 4-echo mecho-UTE scans were prospectively acquired using the protocols in Figure 2. Three scans were reconstructed with both gridding and CG-SENSE+DLR.
Data Analysis:
Logarithmic inversion[11] was applied to the first echo (TE1 = 0.096ms) image to generate CT-like images. The second echo was subtracted from the first echo (TE1-TE2) for better delineation/visualization of ultrashort-T2* tissues (tendons, ligaments, etc.). Mono-exponential fitting was used to generate T2* map. Region of interests on tendons were drawn for T2* measurements. Full-Width at Half-Maximum (FWHM) and Relative Edge SHarpness (RESH)[9] were measured to assess resolution and sharpness, as seen in Figure 3.
Results and Discussion
Figure 4 shows CT-like, subtraction, and T2* map images. Figure 5 shows quantitative comparisons of T2*, FWHM, and RESH.Qualitatively, as shown in Figure 4, CG-SENSE+DLR images show improved image quality and sharpness compared to Gridding counterparts. Furthermore, 3-min and 2-min with CG-SENSE+DLR show improved image quality and sharpness compared to the 5-min with Gridding.
Quantitatively, as shown in Figure 5, tendon T2* values were similar between scans and reconstructions and in the range from literature[12], [13]. As seen in the bar plot, mean T2* values were increased in Gridding reconstruction with shorter acquisition times (likely due to streaking and blurring artifacts) while those in CG-SENSE+DLR stayed relatively consistent around 4.76-4.79ms.
Compared to Gridding counterparts, CG-SENSE+DLR images have an average of 22.76% smaller FWHM and 3.24 times higher RESH. Compared to 5-min Gridding, 3-min and 2-min CG-SENSE+DLR have an average of 30.59% and 8.71% smaller FWHM, respectively and 3.23 and 2.17 times higher RESH, respectively.
Conclusion
This work demonstrated that CG-SENSE+DLR allows accelerated 2- and 3-minute 3D-isotropic 0.8mm3 [reconstructed as 0.4mm3] 4-echo mecho-UTE acquisition with improved resolution and sharpness compared to 5-min mecho-UTE with Gridding reconstruction. The accelerated mecho-UTE may enable widespread clinical adoption to evaluate its diagnostic performance and economic value of comprehensive MSK MRI (routine FSE2D and mecho-UTE) compared to CT or to CT & MRI.Acknowledgements
No acknowledgement found.References
[1] M. C. Florkow, K. Willemsen, V. V. Mascarenhas, E. H. G. Oei, M. van Stralen, and P. R. Seevinck, “Magnetic Resonance Imaging Versus Computed Tomography for Three-Dimensional Bone Imaging of Musculoskeletal Pathologies: A Review,” J. Magn. Reson. Imaging, vol. 56, no. 1, pp. 11–34, 2022, doi: 10.1002/jmri.28067.
[2] C. S. Rajapakse, M. Bashoor-Zadeh, C. Li, W. Sun, A. C. Wright, and F. W. Wehrli, “Volumetric Cortical Bone Porosity Assessment with MR Imaging: Validation and Clinical Feasibility,” Radiology, vol. 276, no. 2, pp. 526–535, Aug. 2015, doi: 10.1148/radiol.15141850.
[3] E. Einarsson et al., “Relating MR relaxation times of ex vivo meniscus to tissue degeneration through comparison with histopathology,” Osteoarthr. Cartil. Open, vol. 2, no. 2, p. 100061, Apr. 2020, doi: 10.1016/j.ocarto.2020.100061.
[4] E. B. Rubin et al., “Advanced MRI Approaches for Evaluating Common Lower Extremity Injuries in Basketball Players: Current and Emerging Techniques,” J. Magn. Reson. Imaging, vol. n/a, no. n/a, doi: 10.1002/jmri.29019.
[5] A. Williams, Y. Qian, S. Golla, and C. R. Chu, “UTE-T2∗ mapping detects sub-clinical meniscus injury after anterior cruciate ligament tear,” Osteoarthritis Cartilage, vol. 20, no. 6, pp. 486–494, Jun. 2012, doi: 10.1016/j.joca.2012.01.009.
[6] A.-S. Agergaard et al., “UTE T2* mapping of tendinopathic patellar tendons: an MRI reproducibility study,” Acta Radiol. Stockh. Swed. 1987, vol. 62, no. 2, pp. 215–224, Feb. 2021, doi: 10.1177/0284185120918807.
[7] L. M. Wilms et al., “UTE-T2* versus conventional T2* mapping to assess posterior cruciate ligament ultrastructure and integrity—an in-situ study,” Quant. Imaging Med. Surg., vol. 12, no. 8, pp. 4190–4201, Aug. 2022, doi: 10.21037/qims-22-251.
[8] K. P. Pruessmann, M. Weiger, P. Börnert, and P. Boesiger, “Advances in sensitivity encoding with arbitrary k-space trajectories,” Magn. Reson. Med., vol. 46, no. 4, pp. 638–651, 2001, doi: 10.1002/mrm.1241.
[9] R. M. Lebel, “Performance characterization of a novel deep learning-based MR image reconstruction pipeline,” ArXiv200806559 Cs Eess, Aug. 2020, Accessed: Sep. 28, 2021. [Online]. Available: http://arxiv.org/abs/2008.06559
[10] M. Kidoh et al., “Deep Learning Based Noise Reduction for Brain MR Imaging: Tests on Phantoms and Healthy Volunteers,” Magn. Reson. Med. Sci., vol. 19, no. 3, pp. 195–206, 2020, doi: 10.2463/mrms.mp.2019-0018.
[11] F. Wiesinger et al., “Zero TE MR bone imaging in the head,” Magn. Reson. Med., vol. 75, no. 1, pp. 107–114, 2016, doi: 10.1002/mrm.25545.
[12] S. J. Breda et al., “Tissue-Specific T2* Biomarkers in Patellar Tendinopathy by Subregional Quantification Using 3D Ultrashort Echo Time MRI,” J. Magn. Reson. Imaging, vol. 52, no. 2, pp. 420–430, 2020, doi: 10.1002/jmri.27108.
[13] Y. Qiao, H.-Y. Tao, K. Ma, Z.-Y. Wu, J.-X. Qu, and S. Chen, “UTE-T2⁎ Analysis of Diseased and Healthy Achilles Tendons and Correlation with Clinical Score: An In Vivo Preliminary Study,” BioMed Res. Int., vol. 2017, p. 2729807, 2017, doi: 10.1155/2017/2729807.
Figures
Figure 1: Study Overview. In combination with routine
FSE2D, mecho-UTE may enable comprehensive one-stop-shop MSK imaging. The goal
of this work is to accelerate the 3D isotropic mecho-UTE using CG-SENSE and
Deep Learning-based Denoising Reconstruction, making it clinically feasible.
Figure 2: Sequence parameters.
Figure 3: An example of a drawn feature profile (A), measured Full-Width at Half-Maximum (FWHM) (B), and calculated Relative Edge SHarpness (RESH) (C), which was a ratio of maximum CG-SENSE+DLR’s gradient magnitude over that of Gridding reconstruction.
Figure 4: Representative CT-like (A), subtraction (B), and T2* map (C) images from three scans with Gridding and CG-SENSE+DLR reconstructions. Red arrowhead is pointing to cortical bone, while yellow arrows are pointing to shoulder tendon, which were invisible on routine clinical MSK MRI.
Figure 5: The table (top) shows quantitative comparisons of measured FWHM, RESH, and tendon’s T2* between Gridding vs. CG-SENSE+DLR and between 5-min Gridding vs. 3-min CG-SENSE+DLR and vs. 2-min CG-SENSE+DLR. The bar plot (bottom) visualizes the tendon’s average T2* values, as shown in the first row of the top table. T2* measurements are more consistent across acquisitions compared to those in Gridding.
DOI: https://doi.org/10.58530/2024/2267