Impact of acquisition and modeling parameters on the test-retest reproducibility of edited GABA.

Literature values vary widely for within-subject test-retest reproducibility of gamma-aminobutyric acid (GABA) measured with edited magnetic resonance spectroscopy (MRS). Reasons for this variation remain unclear. Here, we tested whether three acquisition parameters-(1) sequence complexity (two-experiment MEscher-GArwood Point RESolved Spectroscopy [MEGA-PRESS] vs. four-experiment Hadamard Encoding and Reconstruction of MEGA-Edited Spectroscopy [HERMES]); (2) editing pulse duration (14 vs. 20 ms); and (3) scanner frequency drift (interleaved water referencing [IWR] turned ON vs. OFF)-and two linear combination modeling variations-(1) three different coedited macromolecule models (called "1to1GABA", "1to1GABAsoft", and "3to2MM" in the Osprey software package); and (2) 0.55- versus 0.4-ppm spline baseline knot spacing-affected the within-subject coefficient of variation of GABA + macromolecules (GABA+). We collected edited MRS data from the dorsal anterior cingulate cortex from 20 participants (mean age: 30.8 ± 9.5 years; 10 males). Test and retest scans were separated by removing the participant from the scanner for 5-10 min. Each acquisition consisted of two MEGA-PRESS and two HERMES sequences with editing pulse durations of 14 and 20 ms (referred to here as MEGA-14, MEGA-20, HERMES-14, and HERMES-20; all TE = 80 ms, 224 averages). We identified the best test-retest reproducibility following postprocessing with a composite model of the 0.9- and 3-ppm macromolecules ("3to2MM"); this model performed particularly well for the HERMES data. Furthermore, sparser (0.55- compared with 0.4-ppm) spline baseline knot spacing yielded generally better test-retest reproducibility for GABA+. Replicating our prior results, linear combination modeling in Osprey compared with simple peak fitting in Gannet resulted in substantially better test-retest reproducibility. However, reproducibility did not consistently differ for MEGA-PRESS compared with HERMES, for 14- compared with 20-ms editing pulses, or for IWR-ON versus IWR-OFF. These results highlight the importance of model selection for edited MRS studies of GABA+, particularly for clinical studies that focus on individual patient differences in GABA+ or changes following an intervention.

Improvements in precision and accuracy of complex- relative to real-domain linear combination model spectral fitting not necessarily recovered by zero filling.

Although the information obtained from in vivo proton magnetic resonance spectroscopy (1H MRS) presents a complex-valued spectrum, spectral quantification generally employs linear combination model (LCM) fitting using the real spectrum alone. There is currently no known investigation comparing fit results obtained from LCM fitting over the full complex data versus the real data and how these results might be affected by common spectral preprocessing procedure zero filling. Here, we employ linear combination modeling of simulated and measured spectral data to examine two major ideas: first, whether use of the full complex rather than real-only data can provide improvements in quantification by linear combination modeling and, second, to what extent zero filling might influence these improvements. We examine these questions by evaluating the errors of linear combination model fits in the complex versus real domains against three classes of synthetic data: simulated Lorentzian singlets, simulated metabolite spectra excluding the baseline, and simulated metabolite spectra including measured in vivo baselines. We observed that complex fitting provides consistent improvements in fit accuracy and precision across all three data types. While zero filling obviates the accuracy and precision benefit of complex fitting for Lorentzian singlets and metabolite spectra lacking baselines, it does not necessarily do so for complex spectra including measured in vivo baselines. Overall, performing linear combination modeling in the complex domain can improve metabolite quantification accuracy relative to real fits alone. While this benefit can be similarly achieved via zero filling for some spectra with flat baselines, this is not invariably the case for all baseline types exhibited by measured in vivo data.

Vespa: Integrated applications for RF pulse design, spectral simulation and MRS data analysis.

PURPOSE: The Vespa package (Versatile Simulation, Pulses, and Analysis) is described and demonstrated. It provides workflows for developing and optimizing linear combination modeling (LCM) fitting for 1 H MRS data using intuitive graphical user interface interfaces for RF pulse design, spectral simulation, and MRS data analysis. Command line interfaces for embedding workflows in MR manufacturer platforms and utilities for synthetic dataset creation are included. Complete provenance is maintained for all steps in workflows. THEORY AND METHODS: Vespa is written in Python for compatibility across operating systems. It embeds the PyGAMMA spectral simulation library for spectral simulation. Multiprocessing methods accelerate processing and visualization. Applications use the Vespa database for results storage and cross-application access. Three projects demonstrate pulse, sequence, simulation, and data analysis workflows: (1) short TE semi-LASER single-voxel spectroscopy (SVS) LCM fitting, (2) optimizing MEGA-PRESS (MEscher-GArwood Point RESolved Spectroscopy) flip angle and LCM fitting, and (3) creating a synthetic short TE dataset. RESULTS: The LCM workflows for in vivo basis set creation and spectral analysis showed reasonable results for both the short TE semi-LASER and MEGA-PRESS. Examples of pulses, simulations, and data fitting are shown in Vespa application interfaces for various steps to demonstrate the interactive workflow. CONCLUSION: Vespa provides an efficient and extensible platform for characterizing RF pulses, pulse design, spectral simulation optimization, and automated LCM fitting via an interactive platform. Modular design and command line interface make it easy to embed in other platforms. As open source, it is free to the MRS community for use and extension. Vespa source code and documentation are available through GitHub.

Feasibility and implications of using subject-specific macromolecular spectra to model short echo time magnetic resonance spectroscopy data.

Expert consensus recommends linear-combination modeling (LCM) of 1 H MR spectra with sequence-specific simulated metabolite basis function and experimentally derived macromolecular (MM) basis functions. Measured MM basis functions are usually derived from metabolite-nulled spectra averaged across a small cohort. The use of subject-specific instead of cohort-averaged measured MM basis functions has not been studied widely. Furthermore, measured MM basis functions are not widely available to non-expert users, who commonly rely on parameterized MM signals internally simulated by LCM software. To investigate the impact of the choice of MM modeling, this study, therefore, compares metabolite level estimates between different MM modeling strategies (cohort-mean measured; subject-specific measured; parameterized) in a lifespan cohort and characterizes its impact on metabolite-age associations. 100 conventional (TE = 30 ms) and metabolite-nulled (TI = 650 ms) PRESS datasets, acquired from the medial parietal lobe in a lifespan cohort (20-70 years of age), were analyzed in Osprey. Short-TE spectra were modeled in Osprey using six different strategies to consider the MM baseline. Fully tissue- and relaxation-corrected metabolite levels were compared between MM strategies. Model performance was evaluated by model residuals, the Akaike information criterion (AIC), and the impact on metabolite-age associations. The choice of MM strategy had a significant impact on the mean metabolite level estimates and no major impact on variance. Correlation analysis revealed moderate-to-strong agreement between different MM strategies (r > 0.6). The lowest relative model residuals and AIC values were found for the cohort-mean measured MM. Metabolite-age associations were consistently found for two major singlet signals (total creatine (tCr])and total choline (tCho)) for all MM strategies; however, findings for metabolites that are less distinguishable from the background signals associations depended on the MM strategy. A variance partition analysis indicated that up to 44% of the total variance was related to the choice of MM strategy. Additionally, the variance partition analysis reproduced the metabolite-age association for tCr and tCho found in the simpler correlation analysis. In summary, the inclusion of a single high signal-to-noise ratio MM basis function (cohort-mean) in the short-TE LCM leads to more lower model residuals and AIC values compared with MM strategies with more degrees of freedom (Gaussian parametrization) or subject-specific MM information. Integration of multiple LCM analyses into a single statistical model potentially allows to identify the robustness in the detection of underlying effects (e.g., metabolite vs. age), reduces algorithm-based bias, and estimates algorithm-related variance.

Continuous Automated Analysis Workflow for MRS Studies.

Magnetic resonance spectroscopy (MRS) can non-invasively measure levels of endogenous metabolites in living tissue and is of great interest to neuroscience and clinical research. To this day, MRS data analysis workflows differ substantially between groups, frequently requiring many manual steps to be performed on individual datasets, e.g., data renaming/sorting, manual execution of analysis scripts, and manual assessment of success/failure. Manual analysis practices are a substantial barrier to wider uptake of MRS. They also increase the likelihood of human error and prevent deployment of MRS at large scale. Here, we demonstrate an end-to-end workflow for fully automated data uptake, processing, and quality review.The proposed continuous automated MRS analysis workflow integrates several recent innovations in MRS data and file storage conventions. They are efficiently deployed by a directory monitoring service that automatically triggers the following steps upon arrival of a new raw MRS dataset in a project folder: (1) conversion from proprietary manufacturer file formats into the universal format NIfTI-MRS; (2) consistent file system organization according to the data accumulation logic standard BIDS-MRS; (3) executing a command-line executable of our open-source end-to-end analysis software Osprey; (4) e-mail delivery of a quality control summary report for all analysis steps.The automated architecture successfully completed for a demonstration dataset. The only manual step required was to copy a raw data folder into a monitored directory.Continuous automated analysis of MRS data can reduce the burden of manual data analysis and quality control, particularly for non-expert users and multi-center or large-scale studies and offers considerable economic advantages.

Comparison of linear combination modeling strategies for edited magnetic resonance spectroscopy at 3 T.

J-difference-edited spectroscopy is a valuable approach for the in vivo detection of γ-aminobutyric-acid (GABA) with magnetic resonance spectroscopy (MRS). A recent expert consensus article recommends linear combination modeling (LCM) of edited MRS but does not give specific details regarding implementation. This study explores different modeling strategies to adapt LCM for GABA-edited MRS. Sixty-one medial parietal lobe GABA-edited MEGA-PRESS spectra from a recent 3-T multisite study were modeled using 102 different strategies combining six different approaches to account for co-edited macromolecules (MMs), three modeling ranges, three baseline knot spacings, and the use of basis sets with or without homocarnosine. The resulting GABA and GABA+ estimates (quantified relative to total creatine), the residuals at different ranges, standard deviations and coefficients of variation (CVs), and Akaike information criteria, were used to evaluate the models' performance. Significantly different GABA+ and GABA estimates were found when a well-parameterized MM3co basis function was included in the model. The mean GABA estimates were significantly lower when modeling MM3co , while the CVs were similar. A sparser spline knot spacing led to lower variation in the GABA and GABA+ estimates, and a narrower modeling range-only including the signals of interest-did not substantially improve or degrade modeling performance. Additionally, the results suggest that LCM can separate GABA and the underlying co-edited MM3co . Incorporating homocarnosine into the modeling did not significantly improve variance in GABA+ estimates. In conclusion, GABA-edited MRS is most appropriately quantified by LCM with a well-parameterized co-edited MM3co basis function with a constraint to the nonoverlapped MM0.93 , in combination with a sparse spline knot spacing (0.55 ppm) and a modeling range of 0.5-4 ppm.

Comparison of different linear-combination modeling algorithms for short-TE proton spectra.

Short-TE proton MRS is used to study metabolism in the human brain. Common analysis methods model the data as a linear combination of metabolite basis spectra. This large-scale multi-site study compares the levels of the four major metabolite complexes in short-TE spectra estimated by three linear-combination modeling (LCM) algorithms. 277 medial parietal lobe short-TE PRESS spectra (TE = 35 ms) from a recent 3 T multi-site study were preprocessed with the Osprey software. The resulting spectra were modeled with Osprey, Tarquin and LCModel, using the same three vendor-specific basis sets (GE, Philips and Siemens) for each algorithm. Levels of total N-acetylaspartate (tNAA), total choline (tCho), myo-inositol (mI) and glutamate + glutamine (Glx) were quantified with respect to total creatine (tCr). Group means and coefficient of variations of metabolite estimates agreed well for tNAA and tCho across vendors and algorithms, but substantially less so for Glx and mI, with mI systematically estimated as lower by Tarquin. The cohort mean coefficient of determination for all pairs of LCM algorithms across all datasets and metabolites was R2¯ = 0.39, indicating generally only moderate agreement of individual metabolite estimates between algorithms. There was a significant correlation between local baseline amplitude and metabolite estimates (cohort mean R2¯ = 0.10). While mean estimates of major metabolite complexes broadly agree between linear-combination modeling algorithms at group level, correlations between algorithms are only weak-to-moderate, despite standardized preprocessing, a large sample of young, healthy and cooperative subjects, and high spectral quality. These findings raise concerns about the comparability of MRS studies, which typically use one LCM software and much smaller sample sizes.

Importance of Linear Combination Modeling for Quantification of Glutathione and γ-Aminobutyric Acid Levels Using Hadamard-Edited Magnetic Resonance Spectroscopy.

Background: J-difference-edited 1H-MR spectra require modeling to quantify signals of low-concentration metabolites. Two main approaches are used for this spectral modeling: simple peak fitting and linear combination modeling (LCM) with a simulated basis set. Recent consensus recommended LCM as the method of choice for the spectral analysis of edited data. Purpose: The aim of this study is to compare the performance of simple peak fitting and LCM in a test-retest dataset, hypothesizing that the more sophisticated LCM approach would improve quantification of Hadamard-edited data compared with simple peak fitting. Methods: A test-retest dataset was re-analyzed using Gannet (simple peak fitting) and Osprey (LCM). These data were obtained from the dorsal anterior cingulate cortex of twelve healthy volunteers, with TE = 80 ms for HERMES and TE = 120 ms for MEGA-PRESS of glutathione (GSH). Within-subject coefficients of variation (CVs) were calculated to quantify between-scan reproducibility of each metabolite estimate. Results: The reproducibility of HERMES GSH estimates was substantially improved using LCM compared to simple peak fitting, from a CV of 19.0-9.9%. For MEGA-PRESS GSH data, reproducibility was similar using LCM and simple peak fitting, with CVs of 7.3 and 8.8%. GABA + CVs from HERMES were 16.7 and 15.2%, respectively for the two models. Conclusion: LCM with simulated basis functions substantially improved the reproducibility of GSH quantification for HERMES data.