arXiv subjects

S. M. Shermer

Publications and source records attributed to S. M. Shermer.

4 recordsLinked to original sources

arXiv2026

The Sim-to-Real Gap in MRS Quantification: A Systematic Deep Learning Validation for GABA

Magnetic resonance spectroscopy (MRS) is used to quantify metabolites in vivo and estimate biomarkers for conditions ranging from neurological disorders to cancers. Quantifying low-concentration metabolites such as GABA ($γ$-aminobutyric acid) is challenging due to low signal-to-noise ratio (SNR) and spectral overlap. We investigate and validate deep learning for quantifying complex, low-SNR, overlapping signals from MEGA-PRESS spectra, devise a convolutional neural network (CNN) and a Y-shaped autoencoder (YAE), and select the best models via Bayesian optimisation on 10,000 simulated spectra from slice-profile-aware MEGA-PRESS simulations. The selected models are trained on 100,000 simulated spectra. We validate their performance on 144 spectra from 112 experimental phantoms containing five metabolites of interest (GABA, Glu, Gln, NAA, Cr) with known ground truth concentrations across solution and gel series acquired at 3 T under varied bandwidths and implementations. These models are further assessed against the widely used LCModel quantification tool. On simulations, both models achieve near-perfect agreement (small MAEs; regression slopes $\approx 1.00$, $R^2 \approx 1.00$). On experimental phantom data, errors initially increased substantially. However, modelling variable linewidths in the training data significantly reduced this gap. The best augmented deep learning models achieved a mean MAE for GABA over all phantom spectra of 0.151 (YAE) and 0.160 (FCNN) in max-normalised relative concentrations, outperforming the conventional baseline LCModel (0.220). A sim-to-real gap remains, but physics-informed data augmentation substantially reduced it. Phantom ground truth is needed to judge whether a method will perform reliably on real data.

eess.SP↗

arXiv2021

Benchmarking GABA Quantification: A Ground Truth Data Set and Comparative Analysis of TARQUIN, LCModel, jMRUI and Gannet

Many tools exist for the quantification of GABA-edited magnetic resonance spectroscopy (MRS) data. Despite a recent consensus effort by the MRS community, literature comparing them is sparse but indicates a methodological bias. While invivo data sets can ascertain the level of agreement between tools, ground-truth is required to establish accuracy, and investigate the sources of discrepancy. We present a novel approach to benchmarking GABA quantification tools, using several series of phantom experiments with iterated GABA concentration. Each series presents a different set of background metabolites and environmental conditions allowing comparison of not only individual estimates, but the ability of tools to characterise changes in GABA across a range of potential confounds. The methodology of the phantom experiments is presented, as well as characterisation of the data. We also perform an initial comparative analysis of several common MRS quantification tools and in-house code, to illustrate utility of the dataset and the potential bias introduced by different quantification methods. The GABA-to-NAA ratios reported by each tool are compared to the ground-truth, and estimation accuracy is assessed by linear regression of this relationship. While the linearity of GABA-to-NAA gradients is generally captured by all of the tools, a large variation in the slope, offset and environmental stability of the gradient is observed. The primary driver of differences in linear combination modelling is the choice of basis function. However, tools employing a common basis and pre-processing still produce differences on the order of 4%. Less-strictly parametrised fitting approaches appear to improve the robustness of quantification, but accurate modelling of edit efficiency calculations is still necessary to avoid systematic offsets. In general, the level of variation suggests that comparisons across...

physics.med-ph↗

arXiv2019

Comparison of R1 Mapping Protocols: What are we measuring?

Purpose: Recent work highlights the breadth of reported spin-lattice relaxation rates ($R_1$) for individual tissues. One potential source of variation is the protocol used to determine $R_1$. The methodological dependence of R1 and relaxivity $r_1$ are investigated. Methods: $R_1$ is quantified in gel phantoms with varying concentration of MnCl2, and a small cohort of three healthy volunteers using different acquisition methods. Siemens inversion recovery (IR) and saturation recovery (SR) protocols are applied to phantoms and volunteers. Variable flip angle (VFA) protocols are additionally applied to phantoms. $R_1$ is quantified using single voxel fits, and distributions examined for regions in the thalamus, and cerebellum as well as grey and white matter. Phantoms exclude boundary fits and relaxivity is quantified across the full concentration range. Normality of $R_1$ distributions is assessed by Kolmogorov-Smirnov score, and inter-sequence agreement by two-sample t-test. Results: Phantom relaxivity is found to be 7.16 Hz/mM, 9.22 Hz/mM and 10.65 Hz/mM to 11.91 Hz/mM for IR, SR and VFA methods, respectively. In vivo $R_1$ exhibit low intra-participant variation for IR. SR $R_1$ are lower than IR values with inter- and intra-participant variation on the same order. Brain regions and phantoms mapped with different protocols varied significantly with t-test p-values between 0 and 5E-10. Conclusion: Results suggest a significant protocol dependence of $R_1$, and corresponding relaxivity, suggesting inter-method comparisons should be attempted tentatively, if at all.

physics.med-ph↗

arXiv2019

MRSNet: Metabolite Quantification from Edited Magnetic Resonance Spectra With Convolutional Neural Networks

Quantification of metabolites from magnetic resonance spectra (MRS) has many applications in medicine and psychology, but remains a challenging task despite considerable research efforts. For example, the neurotransmitter $γ$-aminobutyric acid (GABA), present in very low concentration in vivo, regulates inhibitory neurotransmission in the brain and is involved in several processes outside the brain. Reliable quantification is required to determine its role in various physiological and pathological conditions. We present a novel approach to quantification of metabolites from MRS with convolutional neural networks --- MRSNet. MRSNet is trained to perform the multi-class regression problem of identifying relative metabolite concentrations from given input spectra, focusing specifically on the quantification of GABA, which is particularly difficult to resolve. Typically it can only be detected at all using special editing acquisition sequences such as MEGA-PRESS. A large range of network structures, data representations and automatic processing methods are investigated. Results are benchmarked using experimental datasets from test objects of known composition and compared to state-of-the-art quantification methods: LCModel, jMRUI (AQUES, QUEST), TARQUIN, VeSPA and Gannet. The results show that the overall accuracy and precision of metabolite quantification is improved using convolutional neural networks.

eess.IV↗