Magnetic Resonance Spectroscopy Quantification Aided by Deep Estimations of Imperfection Factors and Macromolecular Signal(中文，English)

Dicheng Chen^1,#, Meijin Lin^2,#, Huiting Liu¹,Jiayu Li¹,Yirong Zhou¹, Taishan Kang³,Liangjie Lin⁴,Zhigang Wu⁴,Jiazheng Wang⁴,Jing Li⁵,Jianzhong Lin³, Xi Chen⁶,Di Guo⁷,Xiaobo Qu^1,*

¹ Department of Electronic Science, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen 361104, China.
² Department of Applied Marine Physics and Engineering, and Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen 361104, China.
³ Department of Radiology, Zhongshan Hospital affiliated to Xiamen University, Xiamen 361004, China.
⁴ Philips, Beijing 100016, China.
⁵ Xingaoyi Medical Equipment Company, Yuyao 315400, China.
⁶ McLean Hospital, Harvard Medical School, Belmont, MA 02478, USA.
⁷ School of Computer and Information Engineering, Xiamen University of Technology, Xiamen 361024, China.
^# equally contributed to this work.
^* Email: quxiaobo <at> xmu.edu.cn

Citation

D. Chen, M. Lin, H. Liu, J. Li, Y. Zhou, T. Kang, L. Lin, Z. Wu, J. Wang, J. Li, J. Lin, X. Chen, D. Guo and X. Qu, Magnetic resonance spectroscopy quantification aided by deep estimations of imperfection factors and overall macromolecular signal , IEEE Transactions on Biomedical Engineering, 10.1109/TBME.2024.3354123, 2024.

Synopsis

Magnetic Resonance Spectroscopy (MRS) is an important technique for biomedical detection. However, it is challenging to accurately quantify metabolites with proton MRS[1] due to serious overlaps of metabolite signals, imperfections because of non-ideal acquisition conditions, and interference with strong background signals mainly from macromolecules. The most popular method, LCModel, adopts complicated non-linear least square to quantify metabolites and addresses these problems by designing empirical priors such as basis-sets, imperfection factors. However, when the signal-to-noise ratio of MRS signal is low, the solution may have large deviation[2]. Recently, deep learning has made significant progress in MRS processing[3]-[5]. Its database learning with labels has directly and successfully quantified metabolite concentrations from input spectrum, especially from low SNR spectrum, in an end-to-end manner. However, the end-to-end strategy has limited generalization, and the network should be retrained if any metabolite concentration for the test is out of the range from the trained data.

Main Context

In this work, we designed an MRS Quantification neural Network (QNet) with two modules to extract the IFs (Imperfection Factors) and overall MM(MacroMolecule) signal. Rather than using end-to-end deep learning, QNet employs LLS(Linear Least Squares) aided by the deep neural network of the two modules to quantify metabolite concentrations (see Fig. 1). The LLS in the proposed QNet takes part in the backpropagation of network, feedbacking the quantification error to into metabolite spectrum estimation, which greatly improves the generalization. We also applied the quantum mechanical and exponential model to synthesize the basis-set and incorporate in vivo parameter information to form many spectra to train QNet. Both simulation and in vivo experiments were conducted to compare QNet with LCModel[6]. The former experiments show that QNet can achieve more accurate quantification and be more robust to noise. The latter experiments demonstrate that the quantification results of QNet and LCModel are strong consistent in metabolites tNAA and tCho when SNR is high, whereas QNet can be more stable when SNR decreases.

Meanwhile, QNet has been deployed to CloudBrain platform[7] (see Fig. 2), which can be used for free. For details, please scan the QR code on the upper right of Fig. 2 or the resource link below.

Fig. 1. The procedure of QNet: (a) Procedure of spectrum synthesis, (b) deep learning part with two modules to extract IFs and the overall MM signal, respectively, (c) LLS part using the estimate results from (b) to predict metabolite concentration, and (d) the spectral fits estimated by the real part of the modulated basis set and the predicted . is one of the QNet final outputs and modulated by the predicted IFs from the network in (b) and the concentrations estimated by LLS from (c).

Fig. 2. Example of QNet for MRS quantification of human brain metabolites. (a) Quantification results and fitting analysis of 17 major human brain metabolites, and (b) and (c) corresponding quantification fitting for each metabolite. For more statistical analysis functions and free use, please enter the CloudBrain platform by scanning the QR code on the top right or the resource link below.

Source

CloudBrian-MRS link:https://csrc.xmu.edu.cn/CloudBrain.html

Acknowledgements

This work was supported by the National Natural Science Foundation of China (62122064, 61971361, 62331021, 62371410), the Natural Science Foundation of Fujian Province of China (2023J02005 and 2021J011184), the National Key R&D Program of China (2023YFF0714200), the President Fund of Xiamen University (20720220063), and Nanqiang Outstanding Talent Program of Xiamen University.

References

[1] R. Graaf, "In vivo NMR spectroscopy: Principles and techniques" Wiley-Blackwell, book, 2019.
[2] M. Marjańska et al., "Results and interpretation of a fitting challenge for MR spectroscopy set up by the MRS study group of ISMRM," Magn. Reson. Med., vol. 87, no. 1, pp. 11-32, 2022.
[3] Q. Yang et al., "Physics-driven synthetic data learning for biomedical magnetic resonance: The imaging physics-based data synthesis paradigm for artificial intelligence," IEEE Signal Process Mag., vol. 40, no. 2, pp. 129-140, 2023.
[4] F. Lam et al., "High-dimensional MR spatiospectral imaging by integrating physics-based modeling and data-driven machine learning: Current progress and future directions," IEEE Signal Process Mag., vol. 40, no. 2, pp. 101-115, 2023.
[5] D. Chen et al., "Magnetic resonance spectroscopy deep learning denoising using few in vivo data," IEEE Trans. Comput. Imaging, vol. 9, pp. 448-458, 2023.
[6] S. W. Provencher, "Estimation of metabolite concentrations from localized in vivo proton NMR spectra," Magn. Reson. Med., vol. 30, no. 6, pp. 672-679, 1993.
[7] X. Chen et al., “Cloud Brain-MRS: An intelligent cloud computing platform for in vivo magnetic resonance spectroscopy preprocessing, quantification, and analysis,” J. Magn. Reson., vol. 254, p. 107601, 2024.