Multi-Parametric Molecular Imaging of the Brain Using Optimized Multi-TE Subspace MRSI

Objective: To develop a novel multi-TE MR spectroscopic imaging (MRSI) approach to enable label-free, simultaneous, high-resolution mapping of several molecules and their biophysical parameters in the brain. Methods: The proposed method uniquely integrated an augmented molecular-component-specific subspace model for multi-TE 1H-MRSI signals, an estimation-theoretic experiment optimization (nonuniform TE selection) for molecule separation and parameter estimation, a physics-driven subspace learning strategy for spatiospectral reconstruction and molecular quantification, and a new accelerated multi-TE MRSI acquisition for generating high-resolution data in clinically relevant times. Numerical studies, phantom and in vivo experiments were conducted to validate the optimized experiment design and demonstrate the imaging capability offered by the proposed method. Results: The proposed TE optimization improved estimation of metabolites, neurotransmitters and their T2’s over conventional TE choices, e.g., reducing variances of neurotransmitter concentration by ~ 40% and metabolite T2 by ~ 60%. Simultaneous metabolite and neurotransmitter mapping of the brain can be achieved at a nominal resolution of 3.4 × 3.4 × 6.4 mm3. High-resolution, 3D metabolite T2 mapping was made possible for the first time. The translational potential of the proposed method was demonstrated by mapping biochemical abnormality in a post-traumatic epilepsy (PTE) patient. Conclusion: The feasibility for high-resolution mapping of metabolites/neurotransmitters and metabolite T2’s within clinically relevant time was demonstrated. We expect our method to offer richer information for revealing and understanding metabolic alterations in neurological diseases. Significance: A novel multi-TE MRSI approach was presented that enhanced the technological capability of multi-parametric molecular imaging of the brain. The proposed method presents new technology development and application opportunities for providing richer molecular level information to uncover and comprehend metabolic changes relevant in various neurological applications.

where s(n, TE), ŝ(n, TE; θ) and ξ(n, TE) denote the measured FID signals, model prediction by Eq. ( 2) in the main text and white Gaussian noise with a standard deviation δ, respectively.The variables n and TE index FID sampling point and TE.We defined the set of unknown spectral parameters θ as: θ = [c; T 2 ; T 2 ; ∆f ; ϕ; g] , with each vectorized component being (3) The log-likelihood function of s(n, TE) can be derived as where L(s(n, TE); θ) denotes the likelihood function (Gaussian).Then each block entry (denoted as F θ 1 ,θ 2 ) of the overall FIM, F(θ), can be calculated as: The final FIM can be formed as The lower triangle of F(θ) is the conjugate transpose of the upper triangle part, thus the whole matrix can be generated after obtaining all the entries above the diagonal using Eq. ( 5).Detailed calculations for each entry can be found similarly in Refs.[1,2] Supplementary Figures The 2-TE acquisition, in an equivalent time comparison, yielded the lowest CRLB for GABA while the difference between 1 and 2 TEs for Glx is small.Noting that the optimal 2-TE combinations for Glx and GABA (marked by the blue arrows) are different, which can be due to the inherent resonance structure and J-coupling differences between these two components.Considering that GABA estimation is more challenging, we chose the 2 TEs of 65 and 80 ms which minimize the GABA CRLB for in vivo acquisition.Table S1: Metabolite T 2 statistics (mean ± standard deviation) in grey and white matters computed from the same dataset (Fig. 9 in the main text) across different voxels.NAA T 2 in WM is significantly higher than GM, while Cr and Cho show nonsignificant difference between the two tissue types.The overall T 2 ranges for different metabolites are consistent with previously published results.
Table S2: Regional metabolite T 2 's (mean ± standard deviation) summarized from all the volunteers' data.

Figure S1 :
Figure S1: CRLB analysis and TE optimization for the estimation of Glx (Row 1) and GABA (Row 2).Column 1 shows the minimum CRLB w.r.t. the number of TEs identified using a greedy search (sequentially adding new TEs into the optimization) without considering macromolecule signals (without MM).Columns 2 and 3 show the CRLBs for all possible 2-TE combinations (in the range of 35 to 140 ms with 15 ms increment) without and with MM signals, respectively.Note that the range of CRLBs without MM is [0, 20] in the figure, and the range of CRLBs with MM is [0, 50].The 2-TE acquisition, in an equivalent time comparison, yielded the lowest CRLB for GABA while the difference between 1 and 2 TEs for Glx is small.Noting that the optimal 2-TE combinations for Glx and GABA (marked by the blue arrows) are different, which can be due to the inherent resonance structure and J-coupling differences between these two components.Considering that GABA estimation is more challenging, we chose the 2 TEs of 65 and 80 ms which minimize the GABA CRLB for in vivo acquisition.

R1. 1 Figure
Figure S2: CRLB analysis and TE optimization for metabolite T 2 estimation.Columns 1 and 2 show the CRLB of NAA, Cr, Cho T 2 's and their combination (in ms) w.r.t.2-TE choices.The TE combinations with the lowest CRLB were marked by the blue arrows.The plot on the right shows the minimum CRLB at different numbers of TEs acquired (jointly considering NAA, Cr and Cho), under an equivalent time constraint (repeated TE selections allowed while adding new TEs).The globally minimal CRLB can be achieved with 4 TEs (i.e., 35, 200, 245, and 275 ms), which were used for in vivo data collection.

Figure S3 :
Figure S3: Subspace learning and adaptation: After obtaining the learned multi-TE subspace (shown in the form of spectral basis functions here) from the training data (first column), we projected high-SNR experimental data onto the learned subspace to get reference signals (second column).Then a voxel-wised FIR filter was estimated to adapt the reference signals to the actual experimental data (with additional lineshape variations).Smaller residuals can be observed after this adaptation (third column), indicating better representation capability.Finally, an adapted subspace (last column) can be generated from the refitted data.

Figure S4 :
Figure S4: Experimental setup from a healthy volunteer scan.The shaded areas are the OVS bands for subcutaneous fat suppression (only 6 showing up in this in-plane view, the other 2 set at the top and bottom of the imaging volume).The green box is the shimming volume.

Figure S5 :
Figure S5: 3D metabolite and neurotransmitter mapping results from volunteer #2.The figure arrangement is the same as Fig. 7 in the main text.

Figure S6 :
Figure S6: 3D metabolite and neurotransmitter mapping results from volunteer #3.The figure arrangement is the same as above.

Figure S7 :
Figure S7: High-resolution 3D metabolite T 2 mapping results from volunteer 2. The figure arrangement is the same as Fig. 9 in the main text.

Figure S8 :
Figure S8: High-resolution 3D metabolite T 2 mapping results from volunteer 3. The figure arrangement is the same as above.

Figure S9 :
Figure S9: Comparison of optimal 4-TE (top row) and literature 4-TE (second row) acquisitions for metabolite T 2 mapping.Regional values are shown in the bottom panel.For the optimal 4-TE acquisition: NAA T 2 is 234±15 ms in WM and 222±24 ms in GM, Cr T 2 is 158±8 ms in WM and 154±11 ms in GM, Cho T 2 is 269±12 ms in WM and 271±21 ms in GM.For the literature 4 TEs: NAA T 2 is 232±25 ms in WM and 224±31 ms in GM, Cr T 2 is 131±10 ms in WM and 136±13 ms in GM, and Cho T 2 is 300±28 ms in WM and 298±40 ms in GM.All regional statistics were obtained within the same subject across the imaging slices shown in the figure.The optimal 4-TE acquisition reduced the variance and mitigated the underestimation and overestimation for Cr and Cho T 2 's.

Figure S10 :
Figure S10: Separated spectral components for a representative 2-TE in vivo spectra: both the overall spectrum and the separated components are shown in the same plot for each TE.

Figure S11 :
Figure S11: Separated spectral components for a representative 4-TE in vivo spectra.