2731
Super Resolution MRI using Multiple Signal Classification (MUSIC) Reconstruction1Institute of Biophysics, Chinese Academy of Sciences, Beijing, China, 2University of Chinese Academy of Sciences, Beijing, China, 3Institute of Artificial Intelligence, Hefei Comprehensive National Science Center, Hefei, China
Synopsis
Keywords: Vascular, Blood vessels, Super-Resolution
Motivation: The Fourier transform (FT) reconstruction has convenient implementation and stable performance; however, it has the problem of poor resolving power.
Goal(s): Our goal is to bypass the Fourier transform to obtain MR images, thereby solving the problem of poor resolution and achieving super-resolution imaging.
Approach: We were inspired by array signal processing theory and proposed an approach based on the Multiple Signal Classification (MUSIC) algorithm called MUSIC-MRI.
Results: Our phantom experiments suggest that the resolution ability of MUSIC-MRI is approximately 2x2 better than that of the 2D Fourier transform. Our in-vivo vascular imaging experiments show that the MUSIC-MRI significantly promotes the actual resolution.
Impact: MUSIC-MRI can break through the Rayleigh Limit of Fourier transform and significantly increase the actual resolution ability of the reconstructed images. Scientists or clinicians may use MUSIC-MRI to image very small structures and lesions without modifying MR sequences.
Introduction
The severe sidelobe issue associated with Fourier transformation and the non-convergent characteristics of partial Fourier coefficient summation result in diminished resolving power, which the Rayleigh Limit indicates. We are inspired by radar theory and propose MUSIC-MRI (MUltiple SIgnal Classification) reconstructions to overcome the resolution dilemma of FT. Our experiments show that MUSIC-MRI has super-resolution ability in several in-vivo vascular imaging tasks, such as carotid artery, perforating artery, and deep medullary vein in CADASIL patients.Methods
The modern MRI performs FT for orthogonal basis decomposition on k-space data to get images, while the MUSIC-MRI algorithm mainly includes three steps: performing auto-correlation on the magnetic resonance signal, performing feature decomposition to obtain the target subspace, and performing negative correlation inner product with array manifold to obtain the image. The phantom experiments were performed on a human 3T MR research system (Siemens Healthcare, Erlangen, Germany) with a NOVA coil (Time Medical Holding, CA, USA). We used normalized signal intensity for contrast assessment. The high-contrast spatial resolution slice in the ACR phantom was used for resolution evaluation. The slice is composed of specifically designed hole grids to overcome the partial volume effect. The gradient echo (GRE) sequence was scanned with the following parameters: field of view (FOV) = 200×200 mm2, matrix = 192×192, nominal resolution = 1.04×1.04 mm2, thickness = 2.0mm, TR = 111.0ms, TE = 21.0ms, FA = 20°. The carotid bifurcation imaging experiments were performed on a human 3T MR research system (Siemens Healthcare, Erlangen, Germany) with a NOVA coil. The 2D time of flight (TOF) sequence was scanned with the following parameters: FOV = 130×130mm2, matrix = 64×64( and 80x80, 128x128), nominal resolution = 2.03×2.03mm2 (and 1.63x1.63 mm2, 0.81x0.81 mm2), thickness = 2.0mm, TR = 18.0ms, TE = 5.43ms, FA = 20°. The perforating artery imaging experiments were performed on a human 7T MR research system with a NOVA coil. The 2D time of fight (2D-TOF) sequence was scanned with the following parameters: FOV = 170×170mm2, matrix = 320×320, nominal resolution = 0.59×0.59mm2, thickness = 1.5mm, TR = 30.0ms, TE = 5.92ms, FA = 20°. The deep medullary vein imaging experiments were performed on a human 7T MR research system with a NOVA coil. The gradient echo (GRE) sequence was scanned with the following parameters: FOV = 170×170mm2, matrix = 192×192( and 224x224, 320x320), nominal resolution = 0.89×0.89mm2 (and 0.76x0.76 mm2, 0.53x0.53 mm2), thickness = 1.0mm, TR = 872.0ms, TE = 25.0ms, FA = 20°.Results
The reconstructed images of the high-contrast object are shown in Figure 1. The spatial resolution of MUSIC-MRI with an encoding matrix of 192x192 is equivalent to that of FT-MRI with an encoding matrix of 384x384. The reconstructed images of the low-contrast object detectability are shown in Figure 2. The MUSIC-MRI delineates the lower contrast hole that cannot be resolved by FT-MRI with the same encoding matrix. The reconstructed images of the carotid bifurcation layer are shown in Figure 3. The images obtained by MUSIC-MRI with an encoding matrix of 64x64 can distinguish carotid bifurcation, while FT-MRI needs an encoding matrix of 128x128. Figure 4 shows that MUSIC-MRI has significant resolution ability and image contrast improvement in perforating artery imaging. Figure 5 shows that in deep vein imaging of CADASIL patients, MUSIC-MRI is significantly superior to FT-MRI, especially in the lesion area, even if the encoding matrix is only about half large.Discussion
Our study shows that MR images can be obtained without using FT. The phantom and carotid artery imaging experiments suggest the resolution ability of MUSIC-MRI is approximately 4x better than that of the FT-MRI when reconstructing the same data. Thus, breaking through the Rayleigh Limit and achieving super-resolution. Figure 2 confirms that MUSIC-MRI is capable of resolving the low-contrast object from the k-space data with a small encoding matrix, while FT-MRI can hardly detect the weak signals in this case. Nonetheless, MUSIC-MRI has a consistently better performance in perforating artery imaging whose signals are very weak. However, there are still obvious limitations in the current version of MUSIC-MRI. First, the computation is highly dependent on large-scale SVD, which is very time-consuming. This problem may be considerably alleviated by using high-performance GPUs. Second, the multi-channel data fusion of MUSIC-MRI is relatively difficult and needs further development of the algorithm.Conclusion
A novel reconstruction approach MUSIC-MRI is proposed to obtain super-resolution MR images without modifying the scan sequence. The resolution of MUSIC-MRI images is nearly 4x better than FT-MRI in several in-vivo 2D vascular imaging tasks, such as carotid artery, perforating artery, and deep medullary vein.Acknowledgements
We acknowledge Dr. Jing An from Siemens Shenzhen Magnetic Resonance Ltd. for her support in the experiments. This work was supported in part by National Natural Science Foundation of China (82271985, 82001804, 81961128030), Youth Innovation Promotion Association CAS (2022093), National Science and Technology Innovation 2030 Major Program (2022ZD0211900, 2022ZD0211901), Ministry of Science and Technology of China grant (2019YFA0707103), and National Nature Science Foundation of China grant (31730039).References
1. AUE W, BARTHOLDI E, ERNST R R, 1976. Two-dimensional spectroscopy. application to nuclear magnetic resonance[J]. The Journal of Chemical Physics, 64(5): 2229-2246.
2. ERNST R R, 1992. Nuclear magnetic resonance fourier transform spectroscopy (nobel lecture)[J]. Angewandte Chemie International Edition in English, 31(7): 805-823
3. Schmidt, Ralph. "Multiple emitter location and signal parameter estimation." IEEE transactions on antennas and propagation 34.3 (1986): 276-2804.
4. R. O. Schmidt. A Signal Subspace Approach to Multiple Emitter Location and Spectral Estimation. Ph.D. dissertation, Stanford University, Stanford, CA, 1981.
5. CANDÈS E J, FERNANDEZ-GRANDA C, 2014. Towards a mathematical theory of superresolution[J]. Communications on pure and applied Mathematics, 67(6): 906-956]
6. CANDèS E, ROMBERG J, TAO T, 2006. Robust uncertainty principles: Exact signal reconstruction from highly incomplete frequency information[J/OL]. IEEE Transactions on Information Theory, 52(2): 489-509.
Figures
