Rapid and automated lipid profiling by nuclear magnetic resonance spectroscopy using neural networks.

Nuclear magnetic resonance (NMR) spectroscopy is a powerful tool for quantitative metabolomics; however, quantification of metabolites from NMR data is often a slow and tedious process requiring user input and expertise. In this study, we propose a neural network approach for rapid, automated lipid identification and quantification from NMR data. Multilayered perceptron (MLP) networks were developed with NMR spectra as the input and lipid concentrations as output. Three large synthetic datasets were generated, each with 55,000 spectra from an original 30 scans of reference standards, by using linear combinations of standards and simulating experimental-like modifications (line broadening, noise, peak shifts, baseline shifts) and common interference signals (water, tetramethylsilane, extraction solvent), and were used to train MLPs for robust prediction of lipid concentrations. The performances of MLPS were first validated on various synthetic datasets to assess the effect of incorporating different modifications on their accuracy. The MLPs were then evaluated on experimentally acquired data from complex lipid mixtures. The MLP-derived lipid concentrations showed high correlations and slopes close to unity for most of the quantified lipid metabolites in experimental mixtures compared with ground-truth concentrations. The most accurate, robust MLP was used to profile lipids in lipophilic hepatic extracts from a rat metabolomics study. The MLP lipid results analyzed by two-way ANOVA for dietary and sex differences were similar to those obtained with a conventional NMR quantification method. In conclusion, this study demonstrates the potential and feasibility of a neural network approach for improving speed and automation in NMR lipid profiling and this approach can be easily tailored to other quantitative, targeted spectroscopic analyses in academia or industry.

Simulation of a virtual liver iron overload model and R2 * estimation using multispectral fat-water models for GRE and UTE acquisitions at 1.5 T and 3 T.

R2 *-MRI has emerged as a noninvasive alternative to liver biopsy for assessment of hepatic iron content (HIC). Multispectral fat-water R2 * modeling techniques such as the nonlinear least squares (NLSQ) fitting and autoregressive moving average (ARMA) models have been proposed for the accurate assessment of iron overload by also considering fat, which can otherwise confound R2 *-based HIC measurements in conditions of coexisting iron overload and steatosis. However, the R2 * estimation by these multispectral models has not been systematically investigated for various acquisition methods in iron overload only conditions and across the full clinically relevant range of HICs (0-40 mg Fe/g dry liver weight). The purpose of this study is to evaluate the R2 * accuracy and precision of multispectral models for various multiecho gradient echo (GRE) and ultrashort echo time (UTE) imaging acquisitions by constructing virtual iron overload models based on true histology and synthesizing MRI signals via Monte Carlo simulations at 1.5 T and 3 T, and comparing their results with monoexponential model and published in vivo R2 *-HIC calibrations. The signals were synthesized with TE1 = 1.0 ms for GRE and TE1 = 0.1 ms for UTE acquisition for varying echo spacing, ΔTE (0.1, 0.5, 1, 2 ms), and maximum echo time, TEmax (2, 4, 6, 10 ms). An iron-doped phantom study is also conducted to validate the simulation results in experimental GRE (TE1 = 1.2 ms, ΔTE = 0.72 ms, TEmax = 6.24 ms) and UTE (TE1 = 0.1 ms, ΔTE = 0.5 ms, TEmax = 6.1 ms) acquisitions. For GRE acquisitions, the multispectral ARMA and NLSQ models produced higher slopes (0.032-0.035) compared with the monoexponential model and published in vivo R2 *-HIC calibrations (0.025-0.028). However, for UTE acquisition for shorter echo spacing (≤0.5 ms) and longer maximum echo time, TEmax (≥6 ms), the multispectral and monoexponential signal models produced similar R2 *-HIC slopes (1.5 T, 0.028-0.032; 3 T, 0.014-0.016) and precision values (coefficient of variation < 25%) across the full clinical spectrum of HICs at both 1.5 T and 3 T. The phantom analysis also showed that all signal models demonstrated a significant improvement in R2 * estimation for UTE acquisition compared with GRE, confirming our simulation findings. Future work should investigate the performance of multispectral fat-water models by simulating liver models in coexisting conditions of iron overload and steatosis for accurate R2 * and fat quantification.

Quantitative MRI of diffuse liver diseases: techniques and tissue-mimicking phantoms.

Quantitative magnetic resonance imaging (MRI) techniques are emerging as non-invasive alternatives to biopsy for assessment of diffuse liver diseases of iron overload, steatosis and fibrosis. For testing and validating the accuracy of these techniques, phantoms are often used as stand-ins to human tissue to mimic diffuse liver pathologies. However, currently, there is no standardization in the preparation of MRI-based liver phantoms for mimicking iron overload, steatosis, fibrosis or a combination of these pathologies as various sizes and types of materials are used to mimic the same liver disease. Liver phantoms that mimic specific MR features of diffuse liver diseases observed in vivo are important for testing and calibrating new MRI techniques and for evaluating signal models to accurately quantify these features. In this study, we review the liver morphology associated with these diffuse diseases, discuss the quantitative MR techniques for assessing these liver pathologies, and comprehensively examine published liver phantom studies and discuss their benefits and limitations.

Hepatic Iron Quantification Using a Free-Breathing 3D Radial Gradient Echo Technique and Validation With a 2D Biopsy-Calibrated R2* Relaxometry Method.

BACKGROUND: Hepatic iron content (HIC) is an important parameter for the management of iron overload. Non-invasive HIC assessment is often performed using biopsy-calibrated two-dimensional breath-hold Cartesian gradient echo (2D BH GRE) R2* -MRI. However, breath-holding is not possible in most pediatric patients or those with respiratory problems, and three-dimensional free-breathing radial GRE (3D FB rGRE) has emerged as a viable alternative. PURPOSE: To evaluate the performance of a 3D FB rGRE and validate its R2* and fat fraction (FF) quantification with 3D breath-hold Cartesian GRE (3D BH cGRE) and biopsy-calibrated 2D BH GRE across a wide range of HICs. STUDY TYPE: Retrospective. SUBJECTS: Twenty-nine patients with hepatic iron overload (22 females, median age: 15 [5-25] years). FIELD STRENGTH/SEQUENCE: Three-dimensional radial and 2D and 3D Cartesian multi-echo GRE at 1.5 T. ASSESSMENT: R2* and FF maps were computed for 3D GREs using a multi-spectral fat model and 2D GRE R2* maps were calculated using a mono-exponential model. Mean R2* and FF values were calculated via whole-liver contouring and T2* -thresholding by three operators. STATISTICAL TESTS: Inter- and intra-observer reproducibility was assessed using Bland-Altman and intraclass correlation coefficient (ICC). Linear regression and Bland-Altman analysis were performed to compare R2* and FF values among the three acquisitions. One-way repeated-measures ANOVA and Wilcoxon signed-rank tests, respectively, were used to test for significant differences between R2* and FF values obtained with different acquisitions. Statistical significance was assumed at P < 0.05. RESULTS: The mean biases and ICC for inter- and intra-observer reproducibility were close to 0% and >0.99, respectively for both R2* and FF. The 3D FB rGRE R2* and FF values were not significantly different (P > 0.44) and highly correlated (R2 ≥ 0.98) with breath-hold Cartesian GREs, with mean biases ≤ ±2.5% and slopes 0.90-1.12. In non-breath-holding patients, Cartesian GREs showed motion artifacts, whereas 3D FB rGRE exhibited only minimal streaking artifacts. DATA CONCLUSION: Free-breathing 3D radial GRE is a viable alternative in non-breath-hold patients for accurate HIC estimation. LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY: Stage 2.

Morphological characterization of hepatic steatosis and Monte Carlo modeling of MRI signal for accurate quantification of fat fraction and relaxivity.

Chemical-shift-based fat-water MRI signal models with single- or dual-R2 * correction have been proposed for quantification of fat fraction (FF) and assessment of hepatic steatosis. However, there is a void in our understanding of which model truly mimics the underlying biophysical mechanism of steatosis on MRI signal relaxation. The purpose of this study is to morphologically characterize and build realistic steatosis models from histology and synthesize MRI signal using Monte Carlo simulations to investigate the accuracy of single- and dual-R2 * models in quantifying FF and R2 *. Fat morphology was characterized by performing automatic segmentation on 16 mouse liver histology images and extracting the radius, nearest neighbor (NN) distance, and regional anisotropy of fat droplets. A gamma distribution function (GDF) was used to generalize extracted features, and regression analysis was performed to derive relationships between FF and GDF parameters. Virtual steatosis models were created based on derived morphological and statistical descriptors, and the MRI signal was synthesized at 1.5 T and 3 T. R2 * and FF values were calculated using single- and dual-R2 * models and compared with in vivo R2 *-FF calibrations and simulated FFs. The steatosis models generated with regional anisotropy and NN distribution closely mimicked the true in vivo fat morphology. For both R2 * models, predicted R2 * values showed positive correlation with FFs, with slopes similar to those of the in vivo calibrations (P > 0.05), and predicted FFs showed excellent agreement with true FFs (R2 > 0.99), with slopes close to unity. Our study, hence, demonstrates the proof of concept for generating steatosis models from histologic data and synthesizing MRI signal to show the expected signal relaxation under conditions of steatosis. Our results suggest that a single R2 * is sufficient to accurately estimate R2 * and FF values for lower FFs, which agrees with in vivo studies. Future work involves characterizing and building steatosis models at higher FFs and testing single- and dual-R2 * models for accurate assessment of steatosis.

Quantitative Susceptibility Mapping Using a Multispectral Autoregressive Moving Average Model to Assess Hepatic Iron Overload.

BACKGROUND: R2*-MRI is clinically used to noninvasively assess hepatic iron content (HIC) to guide potential iron chelation therapy. However, coexisting pathologies, such as fibrosis and steatosis, affect R2* measurements and may thus confound HIC estimations. PURPOSE: To evaluate whether a multispectral auto regressive moving average (ARMA) model can be used in conjunction with quantitative susceptibility mapping (QSM) to measure magnetic susceptibility as a confounder-free predictor of HIC. STUDY TYPE: Phantom study and in vivo cohort. SUBJECTS: Nine iron phantoms covering clinically relevant R2* range (20-1200/second) and 48 patients (22 male, 26 female, median age 18 years). FIELD STRENGTH/SEQUENCE: Three-dimensional (3D) and two-dimensional (2D) multi-echo gradient echo (GRE) at 1.5 T. ASSESSMENT: ARMA-QSM modeling was performed on the complex 3D GRE signal to estimate R2*, fat fraction (FF), and susceptibility measurements. R2*-based dry clinical HIC values were calculated from the 2D GRE acquisition using a published R2*-HIC calibration curve as reference standard. STATISTICAL TESTS: Linear regression analysis was performed to compare ARMA R2* and susceptibility-based estimates to iron concentrations and dry clinical HIC values in phantoms and patients, respectively. RESULTS: In phantoms, the ARMA R2* and susceptibility values strongly correlated with iron concentrations (R2 ≥ 0.9). In patients, the ARMA R2* values highly correlated (R2  = 0.97) with clinical HIC values with slope = 0.026, and the susceptibility values showed good correlation (R2  = 0.82) with clinical dry HIC values with slope = 3.3 and produced a dry-to-wet HIC ratio of 4.8. DATA CONCLUSION: This study shows the feasibility that ARMA-QSM can simultaneously estimate susceptibility-based wet HIC, R2*-based dry HIC and FFs from a single multi-echo GRE acquisition. Our results demonstrate that both, R2* and susceptibility-based wet HIC values estimated with ARMA-QSM showed good association with clinical dry HIC values with slopes similar to published R2*-biopsy HIC calibration and dry-to-wet tissue weight ratio, respectively. Hence, our study shows that ARMA-QSM can provide potentially confounder-free assessment of hepatic iron overload. LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY: Stage 2.

CRAFT for NMR lipidomics: Targeting lipid metabolism in leucine-supplemented tumor-bearing mice.

Lipid profiling by 1 H-NMR has gained increasing utility in many fields because of its intrinsically quantitative, nondestructive nature and the ability to differentiate small molecules based on their spectral location. Most nuclear magnetic resonance (NMR) techniques for metabolite quantification use frequency domain analysis that involves many user-dependent steps such as phase and baseline correction and quantification by either manual integration or peak fitting. Recently, Bayesian analysis of time-domain NMR data has been shown to reduce operator bias and increase automation in NMR spectroscopy. In this study, we demonstrate the use of CRAFT (complete reduction to amplitude-frequency table), a Bayesian-based approach to automate processing in NMR-based lipidomics using lipid standards and tissue samples of healthy and tumor-bearing mice supplemented with leucine. Complex mixtures of lipid standards were prepared and examined using CRAFT to validate it against conventional Fourier transform (FT)-NMR and derive a fingerprint to be used for analyzing lipid profiles of serum and liver samples. CRAFT and FT-NMR were comparable in accuracy, with CRAFT achieving higher correlation in quantifying several lipid species. Analysis of the serum lipidome of tumor-bearing mice revealed hyperlipidemia and no signs of hepatic triglyceride accumulation compared with that of the healthy group demonstrating that the tumor-bearing mice were in a state of precachexia. Leucine-supplementation was associated with minimal changes in the lipid profile in both tissues. In conclusion, our study demonstrates that the CRAFT method can accurately identify and quantify lipids in complex lipid mixtures and murine tissue samples and, hence, will increase automation and reproducibility in NMR-based lipidomics.

Ultrashort echo time imaging for quantification of hepatic iron overload: Comparison of acquisition and fitting methods via simulations, phantoms, and in vivo data.

BACKGROUND: Current R2*-MRI techniques for measuring hepatic iron content (HIC) use various acquisition types and fitting models. PURPOSE: To evaluate the accuracy and precision of R2*-HIC acquisition and fitting methods. STUDY TYPE: Signal simulations, phantom study, and prospective in vivo cohort. POPULATION: In all, 132 patients (58/74 male/female, mean age 17.7 years). FIELD STRENGTH/SEQUENCE: 2D-multiecho gradient-echo (GRE) and ultrashort echo time (UTE) acquisitions at 1.5T. ASSESSMENT: Synthetic MR signals were created to mimic published GRE and UTE methods, using different R2* values (25-2000 s-1 ) and signal-to-noise ratios (SNR). Phantoms with varying iron concentrations were scanned at 1.5T. In vivo data were analyzed from 132 patients acquired at 1.5T. R2* was estimated by fitting using three signal models. Accuracy and precision of R2* measurements for UTE acquisition parameters (SNR, echo spacing [ΔTE], maximum echo time [TEmax ]) and fitting methods were compared for simulated, phantom, and in vivo datasets. STATISTICAL TESTS: R2* accuracy was determined from the relative error and by linear regression analysis. Precision was evaluated using coefficient of variation (CoV) analysis. RESULTS: In simulations, all models had high R2* accuracy (error <5%) and precision (CoV <10%) for all SNRs, shorter ΔTE (≤0.5 msec), and longer TEmax (≥10.1 msec); except the constant offset model overestimated R2* at the lowest SNR. In phantoms and in vivo, all models produced similar R2* values for different SNRs and shorter ΔTEs (slopes: 0.99-1.06, R2 > 0.99, P < 0.001). In all experiments, R2* results degraded for high R2* values with longer ΔTE (≥1 msec). In vivo, shorter and longer TEmax gave similar R2* results (slopes: 1.02-1.06, R2 > 0.99, P < 0.001) for the noise subtraction model for 25≤R2*≤2000 s-1 . However, both quadratic and constant offset models, using shorter TEmax (≤4.7 msec) overestimated R2* and yielded high CoVs up to ∼170% for low R2* (<250 s-1 ). DATA CONCLUSION: UTE with TEmax ≥ 10.1 msec and ΔTE ≤ 0.5 msec yields accurate R2* estimates over the entire clinical HIC range. Monoexponential fitting with noise subtraction is the most robust signal model to changes in UTE parameters and achieves the highest R2* accuracy and precision. LEVEL OF EVIDENCE: 2 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;49:1475-1488.

Autoregressive moving average modeling for hepatic iron quantification in the presence of fat.

BACKGROUND: Measuring hepatic R2* by fitting a monoexponential model to the signal decay of a multigradient-echo (mGRE) sequence noninvasively determines hepatic iron content (HIC). Concurrent hepatic steatosis introduces signal oscillations and confounds R2* quantification with standard monoexponential models. PURPOSE: To evaluate an autoregressive moving average (ARMA) model for accurate quantification of HIC in the presence of fat using biopsy as the reference. STUDY TYPE: Phantom study and in vivo cohort. POPULATION: Twenty iron-fat phantoms covering clinically relevant R2* (30-800 s-1 ) and fat fraction (FF) ranges (0-40%), and 10 patients (four male, six female, mean age 18.8 years). FIELD STRENGTH/SEQUENCE: 2D mGRE acquisitions at 1.5 T and 3 T. ASSESSMENT: Phantoms were scanned at both field strengths. In vivo data were analyzed using the ARMA model to determine R2* and FF values, and compared with biopsy results. STATISTICAL TESTS: Linear regression analysis was used to compare ARMA R2* and FF results with those obtained using a conventional monoexponential model, complex-domain nonlinear least squares (NLSQ) fat-water model, and biopsy. RESULTS: In phantoms and in vivo, all models produced R2* and FF values consistent with expected values in low iron and low/high fat conditions. For high iron and no fat phantoms, monoexponential and ARMA models performed excellently (slopes: 0.89-1.07), but NLSQ overestimated R2* (slopes: 1.14-1.36) and produced false FFs (12-17%) at 1.5 T; in high iron and fat phantoms, NLSQ (slopes: 1.02-1.16) outperformed monoexponential and ARMA models (slopes: 1.23-1.88). The results with NLSQ and ARMA improved in phantoms at 3 T (slopes: 0.96-1.04). In patients, mean R2*-HIC estimates for monoexponential and ARMA models were close to biopsy-HIC values (slopes: 0.90-0.95), whereas NLSQ substantially overestimated HIC (slope 1.4) and produced false FF values (4-28%) with very high SDs (15-222%) in patients with high iron overload and no steatosis. DATA CONCLUSION: ARMA is superior in quantifying R2* and FF under high iron and no fat conditions, whereas NLSQ is superior for high iron and concurrent fat at 1.5 T. Both models give improved R2* and FF results at 3 T. LEVEL OF EVIDENCE: 2 Technical Efficacy Stage: 2 J. Magn. Reson. Imaging 2019;50:1620-1632.

Automated vessel exclusion technique for quantitative assessment of hepatic iron overload by R2*-MRI.

BACKGROUND: Extraction of liver parenchyma is an important step in the evaluation of R2*-based hepatic iron content (HIC). Traditionally, this is performed by radiologists via whole-liver contouring and T2*-thresholding to exclude hepatic vessels. However, the vessel exclusion process is iterative, time-consuming, and susceptible to interreviewer variability. PURPOSE: To implement and evaluate an automatic hepatic vessel exclusion and parenchyma extraction technique for accurate assessment of R2*-based HIC. STUDY TYPE: Retrospective analysis of clinical data. SUBJECTS: Data from 511 MRI exams performed on 257 patients were analyzed. FIELD STRENGTH/SEQUENCE: All patients were scanned on a 1.5T scanner using a multiecho gradient echo sequence for clinical monitoring of HIC. ASSESSMENT: An automated method based on a multiscale vessel enhancement filter was investigated for three input data types-contrast-optimized composite image, T2* map, and R2* map-to segment blood vessels and extract liver tissue for R2*-based HIC assessment. Segmentation and R2* results obtained using this automated technique were compared with those from a reference T2*-thresholding technique performed by a radiologist. STATISTICAL TESTS: The Dice similarity coefficient was used to compare the segmentation results between the extracted parenchymas, and linear regression and Bland-Altman analyses were performed to compare the R2* results, obtained with the automated and reference techniques. RESULTS: Mean liver R2* values estimated from all three filter-based methods showed excellent agreement with the reference method (slopes 1.04-1.05, R2 > 0.99, P < 0.001). Parenchyma areas extracted using the reference and automated methods had an average overlap area of 87-88%. The T2*-thresholding technique included small vessels and pixels at the vessel/tissue boundaries as parenchymal area, potentially causing a small bias (<5%) in R2* values compared to the automated method. DATA CONCLUSION: The excellent agreement between reference and automated hepatic vessel segmentation methods confirms the accuracy and robustness of the proposed method. This automated approach might improve the radiologist's workflow by reducing the interpretation time and operator dependence for assessing HIC, an important clinical parameter that guides iron overload management. LEVEL OF EVIDENCE: 3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2018;47:1542-1551.

Quantitative ultrashort echo time imaging for assessment of massive iron overload at 1.5 and 3 Tesla.

PURPOSE: Hepatic iron content (HIC) quantification via transverse relaxation rate (R2*)-MRI using multi-gradient echo (mGRE) imaging is compromised toward high HIC or at higher fields due to the rapid signal decay. Our study aims at presenting an optimized 2D ultrashort echo time (UTE) sequence for R2* quantification to overcome these limitations. METHODS: Two-dimensional UTE imaging was realized via half-pulse excitation and radial center-out sampling. The sequence includes chemically selective saturation pulses to reduce streaking artifacts from subcutaneous fat, and spatial saturation (sSAT) bands to suppress out-of-slice signals. The sequence employs interleaved multi-echo readout trains to achieve dense temporal sampling of rapid signal decays. Evaluation was done at 1.5 Tesla (T) and 3T in phantoms, and clinical applicability was demonstrated in five patients with biopsy-confirmed massively high HIC levels (>25 mg Fe/g dry weight liver tissue). RESULTS: In phantoms, the sSAT pulses were found to remove out-of-slice contamination, and R2* results were in excellent agreement to reference mGRE R2* results (slope of linear regression: 1.02/1.00 for 1.5/3T). UTE-based R2* quantification in patients with massive iron overload proved successful at both field strengths and was consistent with biopsy HIC values. CONCLUSION: The UTE sequence provides a means to measure R2* in patients with massive iron overload, both at 1.5T and 3T. Magn Reson Med 78:1839-1851, 2017. © 2017 Wiley Periodicals, Inc.

Radial Ultrashort TE Imaging Removes the Need for Breath-Holding in Hepatic Iron Overload Quantification by R2* MRI.

OBJECTIVE: The objective of this study is to evaluate radial free-breathing (FB) multiecho ultrashort TE (UTE) imaging as an alternative to Cartesian FB multiecho gradient-recalled echo (GRE) imaging for quantitative assessment of hepatic iron content (HIC) in sedated patients and subjects unable to perform breath-hold (BH) maneuvers. MATERIALS AND METHODS: FB multiecho GRE imaging and FB multiecho UTE imaging were conducted for 46 test group patients with iron overload who could not complete BH maneuvers (38 patients were sedated, and eight were not sedated) and 16 control patients who could complete BH maneuvers. Control patients also underwent standard BH multiecho GRE imaging. Quantitative R2* maps were calculated, and mean liver R2* values and coefficients of variation (CVs) for different acquisitions and patient groups were compared using statistical analysis. RESULTS: FB multiecho GRE images displayed motion artifacts and significantly lower R2* values, compared with standard BH multiecho GRE images and FB multiecho UTE images in the control cohort and FB multiecho UTE images in the test cohort. In contrast, FB multiecho UTE images produced artifact-free R2* maps, and mean R2* values were not significantly different from those measured by BH multiecho GRE imaging. Motion artifacts on FB multiecho GRE images resulted in an R2* CV that was approximately twofold higher than the R2* CV from BH multiecho GRE imaging and FB multiecho UTE imaging. The R2* CV was relatively constant over the range of R2* values for FB multiecho UTE, but it increased with increases in R2* for FB multiecho GRE imaging, reflecting that motion artifacts had a stronger impact on R2* estimation with increasing iron burden. CONCLUSION: FB multiecho UTE imaging was less motion sensitive because of radial sampling, produced excellent image quality, and yielded accurate R2* estimates within the same acquisition time used for multiaveraged FB multiecho GRE imaging. Thus, FB multiecho UTE imaging is a viable alternative for accurate HIC assessment in sedated children and patients who cannot complete BH maneuvers.

Explainable AI to Facilitate Understanding of Neural Network-Based Metabolite Profiling Using NMR Spectroscopy.

Neural networks (NNs) are emerging as a rapid and scalable method for quantifying metabolites directly from nuclear magnetic resonance (NMR) spectra, but the nonlinear nature of NNs precludes understanding of how a model makes predictions. This study implements an explainable artificial intelligence algorithm called integrated gradients (IG) to elucidate which regions of input spectra are the most important for the quantification of specific analytes. The approach is first validated in simulated mixture spectra of eight aqueous metabolites and then investigated in experimentally acquired lipid spectra of a reference standard mixture and a murine hepatic extract. The IG method revealed that, like a human spectroscopist, NNs recognize and quantify analytes based on an analyte's respective resonance line-shapes, amplitudes, and frequencies. NNs can compensate for peak overlap and prioritize specific resonances most important for concentration determination. Further, we show how modifying a NN training dataset can affect how a model makes decisions, and we provide examples of how this approach can be used to de-bug issues with model performance. Overall, results show that the IG technique facilitates a visual and quantitative understanding of how model inputs relate to model outputs, potentially making NNs a more attractive option for targeted and automated NMR-based metabolomics.

Hepatic steatosis modeling and MRI signal simulations for comparison of single- and dual-R2* models and estimation of fat fraction at 1.5T and 3T.

BACKGROUND AND OBJECTIVE: Magnetic resonance imaging (MRI) has emerged as a noninvasive clinical tool for assessment of hepatic steatosis. Multi-spectral fat-water MRI models, incorporating single or dual transverse relaxation decay rate(s) (R2*) have been proposed for accurate fat fraction (FF) estimation. However, it is still unclear whether single- or dual-R2* model accurately mimics in vivo signal decay for precise FF estimation and the impact of signal-to-noise ratio (SNR) on each model performance. Hence, this study aims to construct virtual steatosis models and synthesize MRI signals with different SNRs to systematically evaluate the accuracy of single- and dual-R2* models for FF and R2* estimations at 1.5T and 3.0T. METHODS: Realistic hepatic steatosis models encompassing clinical FF range (0-60 %) were created using morphological features of fat droplets (FDs) extracted from human liver biopsy samples. MRI signals were synthesized using Monte Carlo simulations for noise-free (SNRideal) and varying SNR conditions (5-100). Fat-water phantoms were scanned with different SNRs to validate simulation results. Fat water toolbox was used to calculate R2* and FF for both single- and dual-R2* models. The model accuracies in R2* and FF estimates were analyzed using linear regression, bias plot and heatmap analysis. RESULTS: The virtual steatosis model closely mimicked in vivo fat morphology and Monte Carlo simulation produced realistic MRI signals. For SNRideal and moderate-high SNRs, water R2* (R2*W) by dual-R2* and common R2* (R2*com) by single-R2* model showed an excellent agreement with slope close to unity (0.95-1.01) and R2 > 0.98 at both 1.5T and 3.0T. In simulations, the R2*com-FF and R2*W-FF relationships exhibited slopes similar to in vivo calibrations, confirming the accuracy of our virtual models. For SNRideal, fat R2* (R2*F) was similar to R2*W and dual-R2* model showed slightly higher accuracy in FF estimation. However, in the presence of noise, dual-R2* produced higher FF bias with decreasing SNR, while leading to only marginal improvement for high SNRs and in regions dominated by fat and water. In contrast, single-R2* model was robust and produced accurate FF estimations in simulations and phantom scans with clinical SNRs. CONCLUSION: Our study demonstrates the feasibility of creating virtual steatosis models and generating MRI signals that mimic in vivo morphology and signal behavior. The single-R2* model consistently produced lower FF bias for clinical SNRs across entire FF range compared to dual-R2* model, hence signifying that single-R2* model is optimal for assessing hepatic steatosis.

Automated Segmentation and Morphological Characterization of Hepatic Steatosis and Correlation with Histopathology.

Background/objectives: Prevalence of nonalcoholic fatty liver disease (NAFLD) has increased to 25% of the world population. Hepatic steatosis is a hallmark feature of NAFLD and is assessed histologically using visual and ordinal fat grading criteria (0-3) from the Nonalcoholic Steatohepatitis (NASH) Clinical Research Network (CRN) scoring system. The purpose of this study is to automatically segment and extract morphological characteristics and distributions of fat droplets (FDs) on liver histology images and find associations with severity of steatosis. Methods: A previously published human cohort of 68 NASH candidates was graded for steatosis by an experienced pathologist using the Fat CRN grading system. The automated segmentation algorithm quantified fat fraction (FF) and fat-affected hepatocyte ratio (FHR), extracted fat morphology by calculating radius and circularity of FDs, and examined FDs distribution and heterogeneity using nearest neighbor distance and regional isotropy. Results: Regression analysis and Spearman correlation (ρ) yielded high correlations for radius (R2 = 0.86, ρ = 0.72), nearest neighbor distance (R2 = 0.82, ρ = -0.82), regional isotropy (R2 = 0.84, ρ = 0.74), and FHR (R2 = 0.90, ρ = 0.85), and low correlation for circularity (R2 = 0.48, ρ = -0.32) with FF and pathologist grades, respectively. FHR showed a better distinction between pathologist Fat CRN grades compared to conventional FF measurements, making it a potential surrogate measure for Fat CRN scores. Our results showed variation in distribution of morphological features and steatosis heterogeneity within the same patient's biopsy sample as well as between patients of similar FF. Conclusions: The fat percentage measurements, specific morphological characteristics, and patterns of distribution quantified with the automated segmentation algorithm showed associations with steatosis severity; however, future studies are warranted to evaluate the clinical significance of these steatosis features in progression of NAFLD and NASH.

Functional Plasticity Coupled With Structural Predispositions in Auditory Cortex Shape Successful Music Category Learning.

Categorizing sounds into meaningful groups helps listeners more efficiently process the auditory scene and is a foundational skill for speech perception and language development. Yet, how auditory categories develop in the brain through learning, particularly for non-speech sounds (e.g., music), is not well understood. Here, we asked musically naïve listeners to complete a brief (∼20 min) training session where they learned to identify sounds from a musical interval continuum (minor-major 3rds). We used multichannel EEG to track behaviorally relevant neuroplastic changes in the auditory event-related potentials (ERPs) pre- to post-training. To rule out mere exposure-induced changes, neural effects were evaluated against a control group of 14 non-musicians who did not undergo training. We also compared individual categorization performance with structural volumetrics of bilateral Heschl's gyrus (HG) from MRI to evaluate neuroanatomical substrates of learning. Behavioral performance revealed steeper (i.e., more categorical) identification functions in the posttest that correlated with better training accuracy. At the neural level, improvement in learners' behavioral identification was characterized by smaller P2 amplitudes at posttest, particularly over right hemisphere. Critically, learning-related changes in the ERPs were not observed in control listeners, ruling out mere exposure effects. Learners also showed smaller and thinner HG bilaterally, indicating superior categorization was associated with structural differences in primary auditory brain regions. Collectively, our data suggest successful auditory categorical learning of music sounds is characterized by short-term functional changes (i.e., greater post-training efficiency) in sensory coding processes superimposed on preexisting structural differences in bilateral auditory cortex.

Multi-Tissue Time-Domain NMR Metabolomics Investigation of Time-Restricted Feeding in Male and Female Nile Grass Rats.

Metabolic disease resulting from overnutrition is prevalent and rapidly increasing in incidence in modern society. Time restricted feeding (TRF) dietary regimens have recently shown promise in attenuating some of the negative metabolic effects associated with chronic nutrient stress. The purpose of this study is to utilize a multi-tissue metabolomics approach using nuclear magnetic resonance (NMR) spectroscopy to investigate TRF and sex-specific effects of high-fat diet in a diurnal Nile grass rat model. Animals followed a six-week dietary protocol on one of four diets: chow ad libitum, high-fat ad libitum (HF-AD), high-fat early TRF (HF-AM), or high-fat late TRF (HF-PM), and their liver, heart, and white adipose tissues were harvested at the end of the study and were analyzed by NMR. Time-domain complete reduction to amplitude-frequency table (CRAFT) was used to semi-automate and systematically quantify metabolites in liver, heart, and adipose tissues while minimizing operator bias. Metabolite profiling and statistical analysis revealed lipid remodeling in all three tissues and ectopic accumulation of cardiac and hepatic lipids for HF-AD feeding compared to a standard chow diet. Animals on TRF high-fat diet had lower lipid levels in the heart and liver compared to the ad libitum group; however, no significant differences were noted for adipose tissue. Regardless of diet, females exhibited greater amounts of hepatic lipids compared to males, while no consistent differences were shown in adipose and heart. In conclusion, this study demonstrates the feasibility of performing systematic and time-efficient multi-tissue NMR metabolomics to elucidate metabolites involved in the crosstalk between different metabolic tissues and provides a more holistic approach to better understand the etiology of metabolic disease and the effects of TRF on metabolic profiles.

Early Time-Restricted Feeding Amends Circadian Clock Function and Improves Metabolic Health in Male and Female Nile Grass Rats.

Lengthening the daily eating period contributes to the onset of obesity and metabolic syndrome. Dietary approaches, including energy restriction and time-restricted feeding, are promising methods to combat metabolic disorders. This study explored the effect of early and late time-restricted feeding (TRF) on weight and adiposity, food consumption, glycemic control, clock gene expression, and liver metabolite composition in diurnal Nile grass rats (NGRs). Adult male and female Nile grass rats were randomly assigned to one of three groups: (1) access to a 60% high-fat (HF) diet ad-libitum (HF-AD), (2) time-restricted access to the HF diet for the first 6 h of the 12 h light/active phase (HF-AM) or (3) the second 6 h of the 12 h light/active phase (HF-PM). Animals remained on their respective protocols for six weeks. TRF reduced total energy consumption and weight gain, and early TRF (HF-AM) reduced fasting blood glucose, restored Per1 expression, and reduced liver lipid levels. Although sex-dependent differences were observed for fat storage and lipid composition, TRF improved metabolic parameters in both male and female NGRs. In conclusion, this study demonstrated that early TRF protocol benefits weight management, improves lipid and glycemic control, and restores clock gene expression in NGRs.

GSTM1 and Liver Iron Content in Children with Sickle Cell Anemia and Iron Overload.

Chronic blood transfusions in patients with sickle cell anemia (SCA) cause iron overload, which occurs with a degree of interpatient variability in serum ferritin and liver iron content (LIC). Reasons for this variability are unclear and may be influenced by genes that regulate iron metabolism. We evaluated the association of the copy number of the glutathione S-transferase M1 (GSTM1) gene and degree of iron overload among patients with SCA. We compared LIC in 38 children with SCA and ≥12 lifetime erythrocyte transfusions stratified by GSTM1 genotype. Baseline LIC was measured using magnetic resonance imaging (MRI), R2*MRI within 3 months prior to, and again after, starting iron unloading therapy. After controlling for weight-corrected transfusion burden (mL/kg) and splenectomy, mean pre-chelation LIC (mg/g dry liver dry weight) was similar in all groups: GSTM1 wild-type (WT) (11.45, SD±6.8), heterozygous (8.2, SD±4.52), and homozygous GSTM1 deletion (GSTM1-null; 7.8, SD±6.9, p = 0.09). However, after >12 months of chelation, GSTM1-null genotype subjects had the least decrease in LIC compared to non-null genotype subjects (mean LIC change for GSTM1-null = 0.1 (SD±3.3); versus -0.3 (SD±3.0) and -1.9 (SD±4.9) mg/g liver dry weight for heterozygous and WT, respectively, p = 0.047). GSTM1 homozygous deletion may prevent effective chelation in children with SCA and iron overload.

Prediction of final infarct volume on subacute MRI by quantifying cerebral edema in ischemic stroke.

Final infarct volume in stroke trials is assessed on images obtained between 30 and 90 days after stroke onset. Imaging at such delayed timepoints is problematic because patients may be lost to follow-up or die before the scan. Obtaining an early assessment of infarct volume on subacute scans avoids these limitations; however, it overestimates true infarct volume because of edema. The aim of this study was to develop a novel approach to quantify edema so that final infarct volumes can be approximated on subacute scans. We analyzed data from 20 stroke patients (median age, 75 years) who had baseline, subacute (fu5d) and late (fu90d) MRI scans. Edema displaces CSF from sulci and ventricles; therefore, edema volume was estimated as change in CSF volume between baseline and spatially coregistered fu5d ADC maps. The median (interquartile range, IQR) estimated edema volume was 13.3 (7.5-37.7) mL. The fu5d lesion volumes correlated well with fu90d infarct volumes with slope: 1.24. With edema correction, fu5d infarct volumes are in close agreement, slope: 0.97 and strongly correlated with actual fu90d volumes. The median (IQR) difference between actual and predicted infarct volumes was 0.1 (-3.0-5.7) mL. In summary, this novel technique for estimation of edema allows final infarct volume to be predicted from subacute MRI.