Bayesian deep learning-based 1 H-MRS of the brain: Metabolite quantification with uncertainty estimation using Monte Carlo dropout.

PURPOSE: To develop a Bayesian convolutional neural network (BCNN) with Monte Carlo dropout sampling for metabolite quantification with simultaneous uncertainty estimation in deep learning-based proton MRS of the brain. METHODS: Human brain spectra were simulated using basis spectra for 17 metabolites and macromolecules (N = 100 000) at 3.0 Tesla. In addition, actual in vivo spectra (N = 5) were modified by adjusting SNR and linewidth with increasing severity of spectral degradation (N = 50). A BCNN was trained on the simulated spectra to generate a noise-free, line-narrowed, macromolecule signal-removed, metabolite-only spectrum from a typical human brain spectrum. At inference, each input spectrum was Monte Carlo dropout sampled (50 times), and the resulting mean spectrum and variance spectrum were used for metabolite quantification and uncertainty estimation, respectively. RESULTS: Using the simulated spectra, the mean absolute percent errors of the BCNN-predicted metabolite content were < 10% for Cr, Glu, Gln, mI, NAA, and Tau (< 5% for Glu, NAA, and mI). For all metabolites, the correlations (r's) between the ground-truth error and BCNN-predicted uncertainty ranged 0.72-0.94 (0.83 ± 0.06; p < 0.001). Using the modified in vivo spectra, the extent of variation in the estimated metabolite content against the increasing severity of spectral degradation tended to be smaller with BCNN than with linear combination of model spectra (LCModel). Overall, the variation in metabolite content tended to be more highly correlated with the uncertainty from BCNN than with the Cramér-Rao lower-bounds from LCModel (0.938 ± 0.019 vs. 0.881 ± 0.057 [p = 0.115]). CONCLUSION: The BCNN with Monte Carlo dropout sampling may be used in deep learning-based MRS for the estimation of uncertainty in the machine-predicted metabolite content, which is important in the clinical application of deep learning-based MRS.

Deep learning-based target metabolite isolation and big data-driven measurement uncertainty estimation in proton magnetic resonance spectroscopy of the brain.

PURPOSE: The aim of this study was to develop a method for metabolite quantification with simultaneous measurement uncertainty estimation in deep learning-based proton magnetic resonance spectroscopy (1 H-MRS). METHODS: The reliability of metabolite quantification depends on signal-to-noise ratio (SNR), linewidth, and degree of spectral overlap (DSO), and therefore knowledge about these factors may be utilized in measurement uncertainty estimation in deep learning-based 1 H-MRS. While SNR and linewidth are typically estimated from a representative singlet, DSO needs to be estimated metabolite-specifically. We developed convolutional neural networks (CNNs) capable of isolating target metabolite signal on simulated rat brain spectra at 9.4T, such that, in addition to metabolite content, the signal-to-background ratio (SBR) as a quantitative metric of DSO can be estimated directly from CNN-output for each metabolite. The CNN-predicted SBR was adjusted according to its pre-defined relationship to the ground-truth SBR by exploiting the big spectral data (N = 80 000), and used for measurement uncertainty estimation together with the SNR and linewidth from the CNN-input spectrum. The proposed method was tested first on the simulated spectra in comparison with LCModel and jMRUI and further on in vivo spectra. RESULTS: The proposed method outperformed LCModel and jMRUI in both quantitative accuracy and measurement uncertainty estimation. Using in vivo data, the metabolite concentrations from the proposed method were close to the reported ranges with the measurement uncertainty of glutamine, glutamate, myo-inositol, N-acetylaspartate, and Tau less than 10%. CONCLUSION: The proposed method may be used for metabolite quantification with measurement uncertainty estimation in rat brain at 9.4T by exploiting the spectral isolation capability of the CNNs and the availability of big spectral data.

Reconstruction of spectra from truncated free induction decays by deep learning in proton magnetic resonance spectroscopy.

PURPOSE: To explore the applicability of convolutional neural networks (CNNs) in the reconstruction of spectra from truncated FIDs (tFIDs) in 1 H-MRS, which can be valuable in situations in which data sampling is highly limited, such as spectroscopic magnetic resonance fingerprinting. METHODS: Rat brain FIDs were simulated at 9.4 T based on in vivo data (N = 11) and randomly truncated by retaining 8, 16, 32, 64, 128, 256, 512, and 1024 (null truncation) points (denoted as tFID8 , tFID16 , … tFID1024 ). Using a U-net, 3 CNNs were individually trained (N = 40 000) in time domain only (FID to FID [FID CNNFID ]), in frequency domain only (spectrum to spectrum [spec CNNspec ]), and across the domains (FID to spectrum [FID CNNspec ]) to map the truncated data to their fully sampled versions. The CNNs were tested on the simulated data (N = 5000), and the CNN with the best performance was further tested on the in vivo data, for which the CNN-predicted fully sampled data were analyzed using the LCModel and the results were compared with those from the original, fully sampled data. RESULTS: The best result on the simulated data was obtained with spec CNNspec , which effectively recovered the spectral details even for those input spectra that appear as a hump due to substantial FID truncation (spectra from tFID16 and tFID32 ). Overall, its performance was significantly degraded on the in vivo data. Nonetheless, using spec CNNspec , several coupled spins in addition to the major singlets can be quantified from tFID128 with the error no larger than 10%. CONCLUSION: Upon the availability of more realistically simulated training data, CNNs can also be used in the reconstruction of spectra from truncated FIDs.

Intact metabolite spectrum mining by deep learning in proton magnetic resonance spectroscopy of the brain.

PURPOSE: To develop a robust method for brain metabolite quantification in proton magnetic resonance spectroscopy (1 H-MRS) using a convolutional neural network (CNN) that maps in vivo brain spectra that are typically degraded by low SNR, line broadening, and spectral baseline into noise-free, line-narrowed, baseline-removed intact metabolite spectra. METHODS: A CNN was trained (n = 40 000) and tested (n = 5000) on simulated brain spectra with wide ranges of SNR (6.90-20.74) and linewidth (10-20 Hz). The CNN was further tested on in vivo spectra (n = 40) from five healthy volunteers with substantially different SNR, and the results were compared with those from the LCModel analysis. A Student t test was performed for the comparison. RESULTS: Using the proposed method the mean-absolute-percent-errors (MAPEs) in the estimated metabolite concentrations were 12.49% ± 4.35% for aspartate, creatine (Cr), γ-aminobutyric acid (GABA), glucose, glutamine, glutamate, glutathione (GSH), myo-Inositol (mI), N-acetylaspartate, phosphocreatine (PCr), phosphorylethanolamine, and taurine over the whole simulated spectra in the test set. The metabolite concentrations estimated from in vivo spectra were close to the reported ranges for the proposed method and the LCModel analysis except mI, GSH, and especially Cr/PCr for the LCModel analysis, and phosphorylcholine to glycerophosphorylcholine ratio (PC/GPC) for both methods. The metabolite concentrations estimated across the in vivo spectra with different SNR were less variable with the proposed method (~10% or less) than with the LCModel analysis. CONCLUSION: The robust performance of the proposed method against low SNR may allow a subminute 1 H-MRS of human brain, which is an important technical development for clinical studies.

Parameterization of spectral baseline directly from short echo time full spectra in 1 H-MRS.

PURPOSE: To investigate the feasibility of parameterizing macromolecule (MM) resonances directly from short echo time (TE) spectra rather than pre-acquired, T1 -weighted, metabolite-nulled spectra in 1 H-MRS. METHODS: Initial line parameters for metabolites and MMs were set for rat brain spectra acquired at 9.4 Tesla upon a priori knowledge. Then, MM line parameters were optimized over several steps with fixed metabolite line parameters. The proposed method was tested by estimating metabolite T1 . The results were compared with those obtained with two existing methods. Furthermore, subject-specific, spin density-weighted, MM model spectra were generated according to the MM line parameters from the proposed method for metabolite quantification. The results were compared with those obtained with subject-specific, T1 -weighted, metabolite-nulled spectra. RESULTS: The metabolite T1 were largely in close agreement among the three methods. The spin density-weighted MM resonances from the proposed method were in good agreement with the T1 -weighted, metabolite-nulled spectra except for the MM resonance at ∼3.2 ppm. The metabolite concentrations estimated by incorporating these two different spectral baselines were also in good agreement except for several metabolites with resonances at ∼3.2 ppm. CONCLUSION: The MM parameterization directly from short-TE spectra is feasible. Further development of the method may allow for better representation of spectral baseline with negligible T1 -weighting. Magn Reson Med 78:836-847, 2017. © 2016 International Society for Magnetic Resonance in Medicine.

Neurometabolic profiles of the substantia nigra and striatum of MPTP-intoxicated common marmosets: An in vivo proton MRS study at 9.4 T.

Given the strong coupling between the substantia nigra (SN) and striatum (STR) in the early stage of Parkinson's disease (PD), yet only a few studies reported to date that have simultaneously investigated the neurochemistry of these two brain regions in vivo, we performed longitudinal metabolic profiling in the SN and STR of 1-methyl-1,2,3,6-tetrahydropyridine (MPTP)-intoxicated common marmoset monkey models of PD (n = 10) by using proton MRS (1 H-MRS) at 9.4 T. T2 relaxometry was also performed in the SN by using MRI. Data were classified into control, MPTP_2weeks, and MPTP_6-10 weeks groups according to the treatment duration. In the SN, T2 of the MPTP_6-10 weeks group was lower than that of the control group (44.33 ± 1.75 versus 47.21 ± 2.47 ms, p < 0.05). The N-acetylaspartate to total creatine ratio (NAA/tCr) and γ-aminobutyric acid to tCr ratio (GABA/tCr) of the MPTP_6-10 weeks group were lower than those of the control group (0.41 ± 0.04 versus 0.54 ± 0.08 (p < 0.01) and 0.19 ± 0.03 versus 0.30 ± 0.09 (p < 0.05), respectively). The glutathione to tCr ratio (GSH/tCr) was correlated with T2 for the MPTP_6-10 weeks group (r = 0.83, p = 0.04). In the STR, however, GABA/tCr of the MPTP_6-10 weeks group was higher than that of the control group (0.25 ± 0.10 versus 0.16 ± 0.05, p < 0.05). These findings may be an in vivo depiction of the altered basal ganglion circuit in PD brain resulting from the degeneration of nigral dopaminergic neurons and disruption of nigrostriatal dopaminergic projections. Given the important role of non-human primates in translational studies, our findings provide better understanding of the complicated evolution of PD.

Unsupervised anomaly detection using generative adversarial networks in 1H-MRS of the brain.

The applicability of generative adversarial networks (GANs) capable of unsupervised anomaly detection (AnoGAN) was investigated in the management of quality of 1H-MRS human brain spectra at 3.0 T. The AnoGAN was trained in an unsupervised manner solely on simulated normal brain spectra and used for filtering out abnormal spectra with a broad range of abnormalities, which were simulated by including abnormal ranges of SNR, linewidth and metabolite concentrations and spectral artifacts such as ghost, residual water, and lipid. The AnoGAN was able to filter out those spectra with SNR less than ~11-12 dB with an accuracy of ~80% or higher (assuming a normal SNR range to be 15-18 dB). It also detected with an accuracy of ~80% or higher those spectra, in which NAA levels were reduced by ~25-30% or more from the lower bound and elevated by ~20-30% or more from the upper bound of the normal concentration range (7.5-17 mmol/L), while the concentrations of the rest of the metabolites were all within the normal ranges. Despite the fact that those spectra contaminated with ghost, residual water or lipid have never been involved in the training or optimization of the AnoGAN, they were correctly classified as abnormal regardless of the types of the artifacts, depending solely on their intensity. Although the current version of our AnoGAN requires further technical improvement particularly for the detection of linewidth-associated abnormality and validation on in vivo data, our unsupervised deep learning-based approach could be an option in addition to those previously reported supervised deep learning-based approaches in the binary classification of spectral quality with an extended abnormal spectra regime.

In-Vivo Proton Magnetic Resonance Spectroscopy of 2-Hydroxyglutarate in Isocitrate Dehydrogenase-Mutated Gliomas: A Technical Review for Neuroradiologists.

The diagnostic and prognostic potential of an onco-metabolite, 2-hydroxyglutarate (2HG) as a proton magnetic resonance spectroscopy (1H-MRS) detectable biomarker of the isocitrate dehydrogenase (IDH)-mutated (IDH-MT) gliomas has drawn attention of neuroradiologists recently. However, due to severe spectral overlap with background signals, quantification of 2HG can be very challenging. In this technical review for neuroradiologists, first, the biochemistry of 2HG and its significance in the diagnosis of IDH-MT gliomas are summarized. Secondly, various 1H-MRS methods used in the previous studies are outlined. Finally, wereview previous in vivo studies, and discuss the current status of 1H-MRS in the diagnosis of IDH-MT gliomas.

On the Utility of Short Echo Time (TE) Single Voxel 1H-MRS in Non-Invasive Detection of 2-Hydroxyglutarate (2HG); Challenges and Potential Improvement Illustrated with Animal Models Using MRUI and LCModel.

Mutations in isocitrate dehydrogenase 1 and 2 (IDH1/2) are frequently found in brain tumors, and the resulting onco-metabolite, 2-hydroxyglutarate (2HG), has been suggested to be a potential diagnostic and prognostic biomarker of the diseases. Indeed, recent studies have demonstrated the feasibility of non-invasively detecting 2HG by using proton magnetic resonance spectroscopy (1H-MRS). Due to severe spectral overlaps of 2HG with its background metabolites and spectral baselines, however, the majority of those previous studies employed spectral editing methods with long echo times (TEs) instead of the most commonly used short TE approach with spectral fitting. Consequently, the results obtained with spectral editing methods may potentially be prone to errors resulting from substantial signal loss due to relaxation. Given that the spectral region where the main signal of 2HG resides is particularly sensitive to spectral baseline in metabolite quantification, we have investigated the impact of incorporating voxel-specifically measured baselines into the spectral basis set on the performance of the conventional short TE approach in 2HG detection in rodent models (Fisher 344 rats; n = 19) of IDH1/2 mutant-overexpressing F98 glioma at 9.4T. Metabolite spectra were acquired (SPECIAL sequence) for a tumor region and the contralateral normal region of the brain for each animal. For the estimation of spectral baselines metabolite-nulled spectra were obtained (double-inversion-recovery SPECIAL sequence) for each individual voxels. Data were post-processed with and without the measured baselines using MRUI and LCModel-the two most widely used data post-processing packages. Our results demonstrate that in-vivo detection of 2HG using the conventional short TE approach is challenging even at 9.4T. However, incorporation of voxel-specifically measured spectral baselines may potentially improve its performance. Upon more thorough validation in a larger number of animals and more importantly in human patients, the potential utility of the proposed short TE acquisition with voxel-specific baseline measurement approach in 2HG detection may need to be considered in the study design.