In vivo quantitative 1H MRS of cerebellum and evaluation of quantitation reproducibility by simulation of different levels of noise and spectral resolution

Denoising single MR spectra by deep learning: Miracle or mirage?

PURPOSE

The inherently poor SNR of MRS measurements presents a significant hurdle to its clinical application. Denoising by machine or deep learning (DL) was proposed as a remedy. It is investigated whether such denoising leads to lower estimate uncertainties or whether it essentially reduces noise in signal-free areas only.

METHODS

Noise removal based on supervised DL with U-nets was implemented using simulated 1 H MR spectra of human brain in two approaches: (1) via time-frequency domain spectrograms and (2) using 1D spectra as input. Quality of denoising was evaluated in three ways: (1) by an adapted fit quality score, (2) by traditional model fitting, and (3) by quantification via neural networks.

RESULTS

Visually appealing spectra were obtained; hinting that denoising is well-suited for MRS. However, an adapted denoising score showed that noise removal is inhomogeneous and more efficient for signal-free areas. This was confirmed by quantitative analysis of traditional fit results as well as DL quantitation following DL denoising. DL denoising, although apparently successful as judged by mean squared errors, led to substantially biased estimates in both implementations.

CONCLUSION

The implemented DL-based denoising techniques may be useful for display purposes, but do not help quantitative evaluations, confirming expectations based on estimation theory: Cramér Rao lower bounds defined by the original data and the appropriate fitting model cannot be circumvented in an unbiased way for single data sets, unless additional prior knowledge can be incurred in the form of parameter restrictions/relations or applicable substates.

PCA denoising and Wiener deconvolution of 31 P 3D CSI data to enhance effective SNR and improve point spread function

Purpose

This study evaluates the performance of 2 processing methods, that is, principal component analysis‐based denoising and Wiener deconvolution, to enhance the quality of phosphorus 3D chemical shift imaging data.

Methods

Principal component analysis‐based denoising increases the SNR while maintaining spectral information. Wiener deconvolution reduces the FWHM of the voxel point spread function, which is increased by Hamming filtering or Hamming‐weighted acquisition. The proposed methods are evaluated using simulated and in vivo 3D phosphorus chemical shift imaging data by 1) visual inspection of the spatial signal distribution; 2) SNR calculation of the PCr peak; and 3) fitting of metabolite basis functions.

Results

With the optimal order of processing steps, we show that the effective SNR of in vivo phosphorus 3D chemical shift imaging data can be increased. In simulations, we show we can preserve phosphorus‐containing metabolite peaks that had an SNR < 1 before denoising. Furthermore, using Wiener deconvolution, we were able to reduce the FWHM of the voxel point spread function with only partially reintroducing Gibb‐ringing artifacts while maintaining the SNR. After data processing, fitting of the phosphorus‐containing metabolite signals improved.

Conclusion

In this study, we have shown that principal component analysis‐based denoising in combination with regularized Wiener deconvolution allows increasing the effective spectral SNR of in vivo phosphorus 3D chemical shift imaging data, with reduction of the FWHM of the voxel point spread function. Processing increased the effective SNR by at least threefold compared to Hamming weighted acquired data and minimized voxel bleeding. With these methods, fitting of metabolite amplitudes became more robust with decreased fitting residuals.

Quantitative analysis of spatial averaging effect on chemical shift imaging SNR and noise coherence with k-space sampling schemes

Spatial averaging of multiple voxels from high-resolution chemical shift imaging (hrCSI) is a common strategy for in vivo metabolic studies to achieve a better signal-to-noise ratio (SNR) for a region-of-interest. However, the mechanism about how the spatial averaging approach influences the respective spectral signal and noise and its relevance to the k-space sampling schemes remains unclear. Using three-dimension ¹⁷O CSI technique with the weighted k-space sampling method of Fourier series window, we performed quantitative SNR comparisons between a single low-resolution CSI (lrCSI) voxel (being 27 times larger than the hrCSI voxel size) and the spatially averaged hrCSI voxels with matched sampling volume and location. We demonstrated that the averaged hrCSI voxel spectrum had a large SNR loss (> 4 times) compared to the lrCSI voxel, which was resulted from unmatched increases in signal (~1.9 fold) and noise (~9.3 fold). The signal increase was caused by the spatial overlapping between the adjacent hrCSI voxels. The substantial noise increase was mainly attributed to the strong noise coherence among hrCSI voxels acquired with the weighted k-space sampling. This study presents a quantitative relation between the k-space sampling schemes to an apparent SNR penalty of the spatial averaging approach. The information could be useful for designing CSI acquisition method and determination of optimal spatial resolution for in vivo metabolic imaging studies.

Designing GABA-edited magnetic resonance spectroscopy studies: Considerations of scan duration, signal-to-noise ratio and sample size

BACKGROUND

The relationships between scan duration, signal-to-noise ratio (SNR) and sample size must be considered and understood to design optimal GABA-edited magnetic resonance spectroscopy (MRS) studies.

NEW METHOD

Simulations investigated the effects of signal averaging on SNR, measurement error and group-level variance against a known ground truth. Relative root mean square errors (measurement error) and coefficients of variation (group-level variance) were calculated. GABA-edited data from 18 participants acquired from five voxels were used to examine the relationships between scan duration, SNR and quantitative outcomes in vivo. These relationships were then used to determine the sample sizes required to observe different effect sizes.

RESULTS

In both simulated and in vivo data, SNR increased with the square root of the number of averages. Both measurement error and group-level variance were shown to follow an inverse-square-root function, indicating no significant impact of cumulative artifacts. Comparisons between the first two-thirds of the data and the full dataset showed no statistical difference in group-level variance. There was, however, some variability across the five voxels depending on SNR, which impacted the sample sizes needed to detect group differences in specific brain regions.

COMPARISON WITH EXISTING METHODS

Typical scan durations can be reduced if taking into account a statistically acceptable amount of variance and the magnitudes of predicted effects.

CONCLUSIONS

While scan duration in GABA-edited MRS has typically been considered in terms of SNR, it is more appropriate to think in terms of the amount of measurement error and group-level variance that provides sufficient statistical power.

Designing GABA-edited magnetic resonance spectroscopy studies: Considerations of scan duration, signal-to-noise ratio and sample size

BACKGROUND

The relationships between scan duration, signal-to-noise ratio (SNR) and sample size must be considered and understood to design optimal GABA-edited magnetic resonance spectroscopy (MRS) studies.

NEW METHOD

Simulations investigated the effects of signal averaging on SNR, measurement error and group-level variance against a known ground truth. Relative root mean square errors (measurement error) and coefficients of variation (group-level variance) were calculated. GABA-edited data from 18 participants acquired from five voxels were used to examine the relationships between scan duration, SNR and quantitative outcomes in vivo. These relationships were then used to determine the sample sizes required to observe different effect sizes.

RESULTS

In both simulated and in vivo data, SNR increased with the square root of the number of averages. Both measurement error and group-level variance were shown to follow an inverse-square-root function, indicating no significant impact of cumulative artifacts. Comparisons between the first two-thirds of the data and the full dataset showed no statistical difference in group-level variance. There was, however, some variability across the five voxels depending on SNR, which impacted the sample sizes needed to detect group differences in specific brain regions.

COMPARISON WITH EXISTING METHODS

Typical scan durations can be reduced if taking into account a statistically acceptable amount of variance and the magnitudes of predicted effects.

CONCLUSIONS

While scan duration in GABA-edited MRS has typically been considered in terms of SNR, it is more appropriate to think in terms of the amount of measurement error and group-level variance that provides sufficient statistical power.

Interleaved 31 P MRS imaging of human frontal and occipital lobes using dual RF coils in combination with single-channel transmitter-receiver and dynamic B0 shimming

In vivo 31 P magnetic resonance spectroscopy (MRS) provides a unique tool for the non-invasive study of brain energy metabolism and mitochondrial function. The assessment of bioenergetic impairment in different brain regions is essential to understand the pathophysiology and progression of human brain diseases. This article presents a simple and effective approach which allows the interleaved measurement of 31 P spectra and imaging from two distinct human brain regions of interest with dynamic B0 shimming capability. A transistor-transistor logic controller was employed to actively switch the single-channel X-nuclear radiofrequency (RF) transmitter-receiver between two 31 P RF surface coils, enabling the interleaved acquisition of two 31 P free induction decays (FIDs) from human occipital and frontal lobes within the same repetition time. Linear gradients were incorporated into the RF pulse sequence to perform the first-order dynamic shimming to further improve spectral resolution. The overall results demonstrate that the approach provides a cost-effective and time-efficient solution for reliable 31 P MRS measurement of cerebral phosphate metabolites and adenosine triphosphate (ATP) metabolic fluxes from two human brain regions with high detection sensitivity and spectral quality at 7 T. The same design concept can be extended to acquire multiple spectra from more than two brain regions or can be employed for other magnetic resonance applications beyond the 31 P spin.

Preliminary observation of dynamic changes in alcohol concentration in the human brain with proton magnetic resonance spectroscopy on a 3T MR instrument

PURPOSE

Our purposes were to establish suitable conditions for proton magnetic resonance spectroscopy (MRS) to measure dynamic changes in alcohol concentration in the human brain, to evaluate these changes, and to compare the findings with data from analysis of breath vapor and blood samples.

MATERIALS AND METHODS

We evaluated 4 healthy volunteers (mean age 26.5 years; 3 males, one female) with no neurological findings. All studies were performed with 3-tesla clinical equipment using an 8-channel head coil. We applied our modified single-voxel point-resolved spectroscopy (PRESS) sequence. Continuous measurements of MRS, breath vapor, and blood samples were conducted before and after the subjects drank alcohol with a light meal. The obtained spectra were quantified by LCModel Ver. 6.1, and the accuracy of the MRS measurements was estimated using the estimated standard deviation expressed in percentage (%SD) as a criterion.

RESULTS

Alcohol peaks after drinking were clearly detected at 1.2 ppm for all durations of measurement. Good correlations between breath vapor or blood sample and MRS were found by sub-minute MRS measurement. The continuous measurement showed time-dependent changes in alcohol in the brain and various patterns that differed among subjects.

CONCLUSIONS

The clinical 3T equipment enables direct evaluation of sub-minute changes in alcohol metabolism in the human brain.

Quantitative proton magnetic resonance spectroscopy and spectroscopic imaging of the brain: a didactic review

This article presents background information related to methodology for estimating brain metabolite concentration from magnetic resonance spectroscopy (MRS) and magnetic resonance spectroscopic imaging measurements of living human brain tissue. It reviews progress related to this methodology, with emphasis placed on progress reported during the past 10 years. It is written for a target audience composed of radiologists and magnetic resonance imaging technologists. It describes in general terms the relationship between MRS signal amplitude and concentration. It then presents an overview of the many practical problems associated with deriving concentration solely from absolute measured signal amplitudes and demonstrates how a various signal calibration approaches can be successfully used. The concept of integrated signal amplitude is presented with examples that are helpful for qualitative reading of MRS data as well as for understanding the methodology used for quantitative measurements. The problems associated with the accurate measurement of individual signal amplitudes in brain spectra having overlapping signals from other metabolites and overlapping nuisance signals from water and lipid are presented. Current approaches to obtaining accurate amplitude estimates with least-squares fitting software are summarized.

Quantitation of normal metabolite concentrations in six brain regions by in-vivoH-MR spectroscopy

This study examined the concentrations of brain metabolites visible to in-vivo¹H-Magnetic Resonance Spectroscopy (¹H-MRS) at 1.5 T in a sample of 28 normal subjects. Quantitation was attempted for inositol compounds, choline units, total creatine and N-acetyl moieties, using open-source software. Six brain regions were considered: frontal and parietal white matter, medial temporal lobe, thalamus, pons and cerebellum. Absolute concentrations were derived using tissue water as an internal reference and using an external reference; metabolite signal intensity ratios with respect to creatine were also calculated. The inter-individual variability was smaller for absolute concentrations (internal reference) as compared to that for signal intensity ratios. Significant regional variability in concentration was found for all metabolites, indicating that separate normative values are needed for different brain regions. The values obtained in this study can be used as reference in future studies, provided the same methodology is followed; it is confirmed that despite unsuccessful attempts in the past, smaller coefficients of variation can indeed be obtained through absolute quantification.

Improved myo-inositol detection through Carr-Purcell PRESS: a tool for more sensitive mild cognitive impairment diagnosis

A 3-T study is presented, comparing the ability of two (1) H spectroscopy pulse sequences, Carr-Purcell point resolved spectroscopy (CPRESS; TE = 45 msec), and conventional PRESS (TE = 35 msec), to separate between groups of 20 normal control (NC) and 20 mild cognitive impairment (MCI) subjects. Both sequences showed higher myo-inositol (mI) and mI/N-acetyl-aspartate (NAA) levels in the posterior cingulate gyrus of the MCI subjects. The increased intrasubject repeatability of mI and mI/NAA CPRESS measurements (∼ 6% vs. 9% for PRESS) translated into decreased intraclass variability. A 22% intraclass mI PRESS variability was reduced to 16% for CPRESS, and an 18% intraclass mI/NAA PRESS variability was reduced to 12% for CPRESS for the group of NC subjects. Similar results were observed for the MCI subjects. Decreased intraclass variability led to improved separation between NC and MCI subjects (P = 0.017 for PRESS and P < 0.0001 for CPRESS mI/NAA, the best NC/MCI discriminant for each method). Seventy-five percent sensitivity at eighty percent specificity was demonstrated by mI/NAA CPRESS measurements in separating NC from MCI subjects. High correlations were also observed between subject performance on a number of neuropsychological tests (probing verbal memory, visuoconstruction performance, and visual motor integration) and the mI/NAA ratio; higher correlation coefficients (with stronger statistical significance) were consistently evident for CPRESS than for PRESS data.

Magnetic resonance spectroscopy biomarkers in premanifest and early Huntington disease

OBJECTIVES

To evaluate in vivo brain metabolite differences in control subjects, individuals with premanifest Huntington disease (pre-HD), and individuals with early HD using ¹H magnetic resonance spectroscopy (MRS) and to assess their relationship with motor performance.

METHODS

Eighty-five participants (30 controls, 25 pre-HD, and 30 early HD) were recruited as part of the TRACK-HD study. Eighty-four were scanned at 3 T with single-voxel spectroscopy in the left putamen. Disease burden score was >220 among pre-HD individuals. Subjects underwent TRACK-HD motor assessment including Unified Huntington's Disease Rating Scale (UHDRS) motor scoring and a novel quantitative motor battery. Statistical analyses included linear regression and one-way analysis of variance.

RESULTS

Total N-acetylaspartate (tNAA), a neuronal integrity marker, was lower in early HD (∼15%) vs controls (p < 0.001). N-acetylaspartate (NAA), a constituent of tNAA, was lower in pre-HD (∼8%) and early HD (∼17%) vs controls (p < 0.05). The glial cell marker, myo-inositol (mI), was 50% higher in early HD vs pre-HD (p < 0.01). In early HD, mI correlated with UHDRS motor score (R² = 0.23, p < 0.05). Across pre-HD and early HD, tNAA correlated with performance on a tongue pressure task (R² = 0.30, p < 0.0001) and with disease burden score (R² = 0.17, p < 0.005).

CONCLUSIONS

We demonstrate lower putaminal tNAA in early HD compared to controls in a cross-section of subjects. A novel biomarker role for mI in early HD was also identified. These findings resolve disagreement in the literature about the role of MRS as an HD biomarker. We conclude that putaminal MRS measurements of NAA and mI are promising potential biomarkers of HD onset and progression.

Improving 1H MRSI measurement of cerebral lactate for clinical applications

Accurate measurement of cerebral lactate is critical to the understanding of brain function for psychiatric disorders such as panic disorder and bipolar disorder as well as mitochondrial dysfunction. Proton magnetic spectroscopic imaging (MRSI) techniques can be used to study lactate in vivo; however, accurate measurement of cerebral lactate, which is normally at low basal abundance, can be challenging. In this study, regional lactate measurements obtained with two different MRSI analytic approaches were evaluated using proton echo-planar spectroscopic imaging (PEPSI) data from 18 healthy adults participating in an in vivo sodium lactate infusion study. The results demonstrate that averaging data within a region of interest (ROI) before spectral fitting with LCModel results in significantly improved lactate measurement as compared to averaging chemical concentrations derived from the fitting of individual voxels in the ROI. Simulation results that confirm this finding are also presented. This study additionally outlines an atlas-based approach for the systematic computation of regional distributions of chemical concentrations in large MRSI data sets.

Metabolic correlatives of brain activity in a FOS epilepsy patient

The correlation and the interactions between neuronal activity and underlying metabolic dynamics are still a matter of debate, especially in pathological conditions. This study reports findings obtained on a subject suffering from fixation-off sensitivity (FOS) epilepsy, exploited as a model system of triggerable anomalous electrical activity. Functional Magnetic Resonance Spectroscopy was used to investigate the metabolic response to visual spike-inducing stimuli in a single voxel placed in the temporo-occipital lobe of a FOS epilepsy patient. MRS measurements were additionally performed on a control group of five healthy volunteers. The FOS patient also underwent an EEG session with the same stimulus paradigm. Uniquely in the FOS patient, glutamate and glutamine concentration increased during the first 10 min of stimulation and then returned to baseline. On the other hand, FOS-induced epileptic activity (spiking) endured throughout all the stimulation epoch. The observed metabolic dynamics may be likely linked to a complex interplay between alterations of the metabolic pathways of glutamate and modulation of the neuronal activity.

1H-MRS for the diagnosis of acute disseminated encephalomyelitis: insight into the acute-disease stage

BACKGROUND

Acute disseminated encephalomyelitis (ADEM) is a demyelinating disorder of the central nervous system (CNS). Differentiating ADEM from other inflammatory disorders, such as multiple sclerosis, is not always conclusive using conventional MRI.

OBJECTIVE

To evaluate longitudinal magnetic resonance spectroscopy (MRS) changes that distinguish ADEM from other inflammatory disorders.

MATERIALS AND METHODS

MRI/MRS scans were performed in seven patients with ADEM during the acute and chronic phases of the disease.

RESULTS

Partial recovery was detected between the acute and chronic phases in choline/creatine ratio. Major elevation of lipids and reduction in myo-inositol/creatine ratio was detected in all patients during the acute phase, followed by a reduction in lipids peak and elevation above normal in myo-inositol/creatine ratio during the chronic phase.

CONCLUSION

Consistent and unique MRS changes in metabolite ratios between the acute and chronic presentations of the disease were found. To the best of our knowledge, these patterns have not been described in other inflammatory disorders and might assist in the early diagnosis of ADEM.

Time-resolved functional 1H MR spectroscopic detection of glutamate concentration changes in the brain during acute heat pain stimulation

Non-invasive in vivo detection of cortical neurotransmitter concentrations and their changes in the presence of pain may help to better understand the biochemical principles of pain processing in the brain. In the present study acute heat pain related changes of the excitatory neurotransmitter glutamate were investigated in the anterior insular cortex of healthy volunteers by means of time-resolved functional proton magnetic resonance spectroscopy ((1)H-MRS). Dynamic metabolite changes were estimated with a temporal resolution of five seconds by triggering data acquisition to the time course of the cyclic stimulus application. An overall increase of glutamate concentration up to 18% relative to the reference non-stimulus condition was observed during the application of short pain stimuli.

Spectral resolution amelioration by deconvolution (SPREAD) in MR spectroscopic imaging

PURPOSE

To develop, implement, and evaluate a novel postprocessing method for enhancing the spectral resolution of in vivo MR spectroscopic imaging (MRSI) data.

MATERIALS AND METHODS

Magnetic field inhomogeneity across the imaging volume was determined by acquiring MRI datasets with two differing echo times. The lineshapes of the MRSI spectra were derived from these field maps by simulating an MRSI scan of a virtual sample whose resonance frequencies varied according to the observed variations in the magnetic field. By deconvolving the lineshapes from the measured MRSI spectra, the linebroadening effects of the field inhomogeneities were reduced significantly.

RESULTS

Both phantom and in vivo proton MRSI spectra exhibited significantly enhanced spectral resolutions and improved spectral lineshapes following application of our method. Quantitative studies on a phantom show that, on average, the full width at half maximum of water peaks was reduced 42%, the full width at tenth maximum was reduced 38%, and the asymmetries of the peaks were reduced 86%.

CONCLUSION

Our method reduces the linebroadening and lineshape distortions caused by magnetic field inhomogeneities. It substantially improves the spectral resolution and lineshape of MRSI data.

MRS signal quantitation: a review of time- and frequency-domain methods

In this paper an overview of time-domain and frequency-domain quantitation methods is given. Advantages and drawbacks of these two families of quantitation methods are discussed. An overview of preprocessing methods, such as lineshape correction methods or unwanted component removal methods, is also given. The choice of the quantitation method depends on the data under investigation and the pursued objectives.

Effect of signal-to-noise ratio and spectral linewidth on metabolite quantification at 4 T

The accuracy and precision of measurements of metabolite concentrations from short echo-time spectra has previously been characterized at l.5 T as a function of signal-to-noise ratio (SNR) and peak linewidth. The purpose of this study was to characterize the systematic error in quantification of metabolite concentrations associated with linewidth and SNR for the major metabolites of interest in the short echo-time 1H-MR spectrum at 4 T. Simulated 4 T LASER localized spectra (TE = 46 ms) were generated with full width at half maximum (FWHM) over the range 4-14 Hz, and SNR over the range 5-500 by adding 100 Gaussian-distributed noise realizations at each combination of SNR and linewidth. Linewidth and SNR were treated as independent parameters, and therefore an increase in linewidth at a constant SNR resulted in increased metabolite areas. All spectra were fitted in the time domain using identical prior-knowledge and relative parameter starting values. Six metabolites (N-acetylaspartate, glutamate, creatine, myo-inositol, glycerophosphocholine, phosphocholine) were quantified with >90% accuracy and <10% standard deviation at SNR = 10 for linewidths ranging from 8 to 14 Hz FWHM. These simulations did not consider additional sources of variation, including eddy current artifacts, incomplete macromolecule baseline removal, and incomplete water suppression. Regardless, the results show that metabolite quantification from 4 T short echo-time 1H-MRS is sensitive to SNR and linewidth.

Recent advances in magnetic resonance neurospectroscopy

Over the past two decades, proton magnetic resonance spectroscopy (proton MRS) of the brain has made the transition from research tool to a clinically useful modality. In this review, we first describe the localization methods currently used in MRS studies of the brain and discuss the technical and practical factors that determine the applicability of the methods to particular clinical studies. We also describe each of the resonances detected by localized solvent-suppressed proton MRS of the brain and discuss the metabolic and biochemical information that can be derived from an analysis of their concentrations. We discuss spectral quantitation and summarize the reproducibility of both single-voxel and multivoxel methods at 1.5 and 3-4 T. We have selected three clinical neurologic applications in which there has been a consensus as to the diagnostic value of MRS and summarize the information relevant to clinical applications. Finally, we speculate about some of the potential technical developments, either in progress or in the future, that may lead to improvements in the performance of proton MRS.

Metabolic alteration transients during paroxysmal activity in an epileptic patient with fixation-off sensitivity: a case study

The purpose of this study was to investigate short-time metabolic variations related to continuous epileptic activity elicited by fixation-off sensitivity (FOS). Time-resolved magnetic resonance spectroscopy was performed on a patient on whom previous clinical findings clearly indicated presence of FOS. The epileptic focus was localized with a simultaneous electroencephalographic and functional magnetic resonance imaging study. The results showed a linear increase of the sum of glutamate and glutamine with time of paroxysmal activity in epileptic focus and much greater concentration of choline-containing compounds in focus than in the contralateral side.

Error images for spectroscopic imaging by LCModel using Cramer-Rao bounds

The results of spectroscopic imaging (SI) measurements are often presented as metabolic images. If the spectra quality is not sufficient, the calculated concentrations are biased and the metabolic images show an incorrect metabolite distribution. To simplify the quality analysis of spectra measured by SI, an error image, reflecting the accuracy of the computed concentrations, can be displayed along with the metabolite image. In this paper the relevance of Cramer-Rao bounds (CRBs) calculated by the LCModel program to describe errors in estimated concentrations is validated using spectra simulations. The relation between the average CRBs and standard deviations (STD) of metabolite concentrations from 100 simulated spectra for various signal to noise ratio and line broadening conditions is evaluated. A parameter for calculating error images for metabolite ratios is proposed and an effective way to display error images is shown. The results suggest that the average CRBs are strongly correlated with the standard deviations and hence that CRB values reflect the relative uncertainty of the calculated concentrations. The error information can be integrated directly into a metabolite image by displaying only those areas of the metabolite image with corresponding CRBs below a selected threshold or by mapping CRBs as a transparency of the metabolite image. The concept of error images avoids extensive examination of each SI spectrum and helps to reject low quality spectra.