2D Magnetic Resonance Spectroscopy Research Articles

Progressive myoclonic epilepsy type 1 (EPM1) patients present with abnormal 1H MRS brain metabolic profiles associated with cognitive function

PurposeProgressive myoclonic epilepsy, type 1A (EPM1, Unverricht-Lundborg disease), is a rare neurodegenerative autosomal recessive disorder characterized by stimulus-sensitive and action myoclonus and tonic-clonic epileptic seizures. Patients develop neurological symptoms, including ataxia, intention tremor, and dysarthria, over time, with relatively limited and nonspecific MRI atrophy findings. The effects of the disease on brain metabolism are largely unknown. MethodEighteen EPM1 patients (9 M, 9F) underwent clinical evaluation and neuropsychological testing, which included the assessment of intellectual ability, verbal memory, and psychomotor and executive functions. Magnetic resonance spectroscopy (MRS) and imaging (MRI) were performed on a 1.5 T MRI system. 2D MRS chemical shift imaging (CSI) maps (TE = 270) were obtained from the following regions of the brain: basal ganglia, thalamus, insula, splenium, and occipital white and gray matter, and N-acetyl-aspartate (NAA)-, choline (Cho)-, and lactate (Lac)-to-creatine (Cr) ratios were analyzed. Ten healthy age-and sex-matched subjects (5M, 5F) were used as controls for MRS. ResultsWe found significant brain metabolic changes involving lactate, NAA, and choline, which are widespread in the basal ganglia, thalamic nuclei, insula, and occipital areas of EPM1 patients. Changes, especially in the right insula, basal ganglia, and thalamus, were associated with intellectual abilities and impairment of the psychomotor and executive functions of EPM1 patients. ConclusionMultiple brain metabolic alterations suggest the presence of neurodegeneration associated with EPM1 progression. The changes in metabolite ratios are associated with the neurocognitive dysfunction caused by the disease. However, the role of MRS findings in understanding pathophysiology of EPM1 warrants further studies.

Simultaneous non-convex low rank regularization for fast magnetic resonance spectroscopy reconstruction

2D magnetic resonance spectroscopy (MRS) is extensively used to analyze the components, structures, and interactions of substances in chemistry and bioengineering. To reduce data acquisition time, the spatiotemporally encoded ultrafast (STEU) MRS employs a fast means to acquire data in the hybrid time and frequency (HTF) plane, but relies on non-uniform sampling (NUS) technique. After that, a proper reconstruction method is essential to recover a high quality 2D MRS from undersampled HTF data. In this work, each column and row of 2D MRS are converted into Hankel matrices, by which a simultaneous low rank regularization model is proposed to exploit the bivariate exponential structure of 2D MRS signal. Additionally, a non-convex surrogate function for rank is integrated in the proposed model to more precisely enforce the low rank property of Hankel matrices. To ensure computational accuracy and convergence, an efficient numerical algorithm is further deduced based on alternating direction method of multipliers (ADMM) iteration. The experimental results have shown that the proposed method significantly outperforms existing 2D MRS reconstruction methods, and presents a strong robustness to low sampling rates, various sampling patterns, and measurement noise with a favorable complexity.

Glioblastoma Stem-Like Cells (GSCs) with Mesenchymal Signature: Lipid Profiles of Mobile Lipids Obtained with MRS before and after Radio/Chemical Treatments.

Glioblastoma is the most common and lethal primary malignant brain tumor in adults. Glioblastoma stem cells (GSCs) promote and are responsible for glioblastoma intratumoral heterogeneity and therapy resistance, due to their two main features: self-renewal and differentiation. Lipids have important biological and physiological functions that are critical for understanding the regulation and control of stem cell fate; lipid metabolism and related unsaturation levels play a possible role as the target of therapeutics to overcome glioblastoma radioresistance. This paper aimed at an in-depth analysis of 13 GSC mesenchymal (MES) lines, two subclones, and a stabilized glioblastoma line (T98G) by magnetic resonance spectroscopy (MRS). Particularly, 2D MRS was used to investigate lipid unsaturation behavior during growth in culture and after treatment with etomoxir and photon beams. MES lines, although belonging to the same genetic and metabolic cluster, showed metabolic heterogeneity when observed by MRS, focusing on lipid signals. Nonetheless, the observed unsaturation level stability for two representative lines after stressful treatments suggests unusual robustness of the unsaturation levels for each line, as a peculiar and intrinsic characteristic of GSCs.

Post-traumatic stress disorder affects fucose-\u03b1(1\u20132)-glycans in the human brain: preliminary findings of neuro deregulation using in vivo two-dimensional neuro MR spectroscopy

Post-traumatic stress disorder (PTSD) is triggered by experiencing terrifying event(s) for which there is currently no objective test for a definitive diagnosis. We report a pilot study where two-dimensional (2D) neuro magnetic resonance spectroscopy (MRS), collected at 3 T in a clinical scanner with a 64-channel head coil, identifies neuro deregulation in the PTSD cohort. The control subjects (n = 10) were compared with PTSD participants with minimal co-morbidities (n = 10). The 2D MRS identified statistically significant increases in the total spectral region containing both free substrate fucose and fucosylated glycans of 31% (P = 0.0013), two of multiple fucosylated glycans (Fuc IV and VI) were elevated by 48% (P = 0.002), and 41% (P = 0.02), respectively, imidazole was increased by 12% (P = 0.002), and lipid saturation was increased by 12.5% (P = 0.009). This is the first evidence of fucosylated glycans, reported in animals to be involved in learning and memory, to be affected in humans with PTSD.

Obstructive sleep apnea is associated with altered midbrain chemical concentrations

Obstructive sleep apnea (OSA) is accompanied by altered structure and function in cortical, limbic, brainstem, and cerebellar regions. The midbrain is relatively unexamined, but contains many integrative nuclei which mediate physiological functions that are disrupted in OSA. We therefore assessed the chemistry of the midbrain in OSA in this exploratory study. We used a recently developed accelerated 2D magnetic resonance spectroscopy (2D-MRS) technique, compressed sensing-based 4D echo-planar J-resolved spectroscopic imaging (4D-EP-JRESI), to measure metabolites in the midbrain of 14 OSA (mean age±SD:54.6±10.6years; AHI:35.0±19.4; SAO2 min:83±7%) and 26 healthy control (50.7±8.5years) subjects. High-resolution T1-weighted scans allowed voxel localization. MRS data were processed with custom MATLAB-based software, and metabolite ratios calculated with respect to the creatine peak using a prior knowledge fitting (ProFit) algorithm. The midbrain in OSA showed decreased N-acetylaspartate (NAA; OSA:1.24±0.43, Control:1.47±0.41; p=0.03; independent samples t-test), a marker of neuronal viability. Increased levels in OSA over control subjects appeared in glutamate (Glu; OSA:1.23±0.57, Control:0.98±0.33; p=0.03), ascorbate (Asc; OSA:0.56±0.28, Control:0.42±0.20; (50.7±8.5years; p=0.03), and myo-inositol (mI; OSA:0.96±0.48, Control:0.72±0.35; p=0.03). No differences between groups appeared in γ-aminobutyric acid (GABA) or taurine. The midbrain in OSA patients shows decreased NAA, indicating neuronal injury or dysfunction. Higher Glu levels may reflect excitotoxic processes and astrocyte activation, and higher mI is also consistent with glial activation. Higher Asc levels may result from oxidative stress induced by intermittent hypoxia in OSA. Additionally, Asc and Glu are involved with glutamatergic processes, which are likely upregulated in the midbrain nuclei of OSA patients. The altered metabolite levels help explain dysfunction and structural deficits in the midbrain of OSA patients.

Accelerated 2D magnetic resonance spectroscopy of single spins using matrix completion.

Two dimensional nuclear magnetic resonance (NMR) spectroscopy is one of the major tools for analysing the chemical structure of organic molecules and proteins. Despite its power, this technique requires long measurement times, which, particularly in the recently emerging diamond based single molecule NMR, limits its application to stable samples. Here we demonstrate a method which allows to obtain the spectrum by collecting only a small fraction of the experimental data. Our method is based on matrix completion which can recover the full spectral information from randomly sampled data points. We confirm experimentally the applicability of this technique by performing two dimensional electron spin echo envelope modulation (ESEEM) experiments on a two spin system consisting of a single nitrogen vacancy (NV) centre in diamond coupled to a single 13C nuclear spin. The signal to noise ratio of the recovered 2D spectrum is compared to the Fourier transform of randomly subsampled data, where we observe a strong suppression of the noise when the matrix completion algorithm is applied. We show that the peaks in the spectrum can be obtained with only 10% of the total number of the data points. We believe that our results reported here can find an application in all types of two dimensional spectroscopy, as long as the measured matrices have a low rank.

SU-D-303-07: A Lorentzian-Function-Sparsity Approach for Fast High-Dimensional Magnetic Resonance Spectroscopy

Purpose: High-dimensional magnetic resonance spectroscopy (MRS) is challenging, even with the state-of-art L1-sparsity based compressive reconstruction method. For example, the reconstructed spectroscopy quality deteriorates dramatically with 25% undersampled 2D MRS data. In this work, using the prior that the MRS consists of Lorentzian functions, we aim to develop a robust Lorentzian-function-sparsity based spectroscopy reconstruction method for high-dimensional MRS. Methods: The proposed method utilizes the sparsity via Lorentzian functions. That is, instead of reconstructing thousands of pixel-wise unknowns, the Lorentzian-function based approaches considers the parameter reconstruction with only tens of unknowns, i.e., the center, magnitude and the shape for each Lorentzian function. And the simulated annealing algorithm is developed to find a global minimizer for this small-scale nonlinear and nonconvex optimization problem, in which the 1D Fourier transform of the first k-line is used to identify the initial guesses of centers and magnitudes. Results: The proof-of-concept simulations were performed for 2D MRS. First we compared FFT method, L1-sparsity method and Lorentzian-function-sparsity method with 25% undersampled data for a simple single-peak case, during which FFT and L1 results contained severe artifacts while the Lorentzian method was able to provide nearly perfect reconstructed image. Then we tested a more practical case with seven peaks of widely ranged peak intensities, during which Lorentzian method was still able to provide almost perfect reconstruction Result even with 1% undersampled data. Conclusion: A new MRS reconstruction method is proposed using the Lorentzian-function-based sparsity, with significantly reduced number of unknown variables. The new method can achieve significantly better MRS reconstruction results than FFT method or L1-based sparsity method, e.g., even with 1% k-space data. Boyu Jiang and Hao Gao were partially supported by the NSFC (#11405105), the 973 Program (#2015CB856000) and the Shanghai Pujiang Talent Program (#14PJ1404500)

Ultrafast acquisition of localized two-dimensional magnetic resonance correlated spectra of inhomogeneous biological tissues with resolution improvements

Long acquisition time and strict requests of homogeneous magnetic fields restrain wide applications of localized 2D magnetic resonance spectroscopy. Here, a pulse sequence utilizing the spatial encoding technique and spin echo is proposed to obtain localized 2D correlated spectra with resolution improvements under inhomogeneous fields within greatly reduced times. The localized correlated spectrum of pig marrow obtained with the proposed scheme retained more information compared to the conventional spectrum. The method proposed herein offers important perspectives for fast in vivo analyses of metabolites in living inhomogeneous organisms and in situ characterization of intermediates of inhomogeneous chemical reaction systems.

A Post-processing Framework for Localized 2D MR Spectroscopy in Vivo

We propose a post-processing framework for localized two-dimensional (2D) magnetic resonance spectroscopy (MRS) in vivo. Our framework consists of corrections on eddy current and subject motion along with the framework used in conventional analytical 2D nuclear magnetic resonance (NMR) spectroscopy. In the eddy current correction, the phases of the free induction decays (FIDs) of the metabolite (1)H are corrected along the t₂ direction by the phase of the FID of water (1)H. The corrected FIDs are Fourier transformed along the t₂ direction, and interferograms of F(t₁, ω₂) are calculated. In the motion correction, the zero-order phase of the N-acetyl aspartate (NAA) singlet peak for each t₁ axis is corrected after correction of frequency drift. We applied this framework in phantom and human brain measurements in a 4.7T whole-body MR system. Two-dimensional data were collected by the localized 2D constant-time correlation spectroscopy (CT-COSY) sequence. We used a phantom containing a brain metabolite mixture of NAA, creatine (Cr), glutamate (Glu), glutamine (Gln) and γ-amino butyric acid (GABA). We demonstrated the eddy current correction procedure in the phantom experiments and the subject motion correction in human measurements. Though asymmetric patterns of the singlets of NAA and Cr were shown around the peak along the F₂ direction in the reconstructed phantom spectra without eddy current correction, symmetric patterns arose after the correction. The t1 noise caused by those singlets was found in the human brain spectra without motion correction. The t₁ noise was sufficiently suppressed by the motion correction. Our proposed post-processing framework for localized 2D MRS can improve the quality of in vivo 2D spectra and may allow improved quantitation and robustness of in vivo 2D spectroscopy.

3D localized 2D ultrafast J-resolved magnetic resonance spectroscopy: In vitro study on a 7T imaging system

2D Magnetic Resonance Spectroscopy (MRS) is a well known tool for the analysis of complicated and overlapped MR spectra and was therefore originally used for structural analysis. It also presents a potential for biomedical applications as shown by an increasing number of works related to localized in vivo experiments. However, 2D MRS suffers from long acquisition times due to the necessary collection of numerous increments in the indirect dimension (t1). This paper presents the first 3D localized 2D ultrafast J-resolved MRS sequence, developed on a small animal imaging system, allowing the acquisition of a 3D localized 2D J-resolved MRS spectrum in a single scan. Sequence parameters were optimized regarding Signal-to-Noise ratio and spectral resolution. Sensitivity and spatial localization properties were characterized and discussed. An automatic post-processing method allowing the reduction of artifacts inherent to ultrafast excitation is also presented. This sequence offers an efficient signal localization and shows a great potential for in vivo dynamic spectroscopy.

Combined DCE‐MRI and single‐voxel 2D MRS for differentiation between benign and malignant breast lesions

Dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) and proton (1H) magnetic resonance spectroscopy (MRS) provide structural and biochemical information, including vascular volume, vascular permeability and tissue metabolism. In this study, we performed analysis of the enhancement characteristic from DCE-MRI and the biochemical information provided by two-dimensional (2D) Localized Correlated Spectroscopy (L-COSY) MRS to determine the sensitivity and specificity of using DCE-MRI alone compared to the combination with 2D MRS. The metabolite ratios from the 2D MRS spectra were analyzed using multivariate statistical analyses to determine a method capable of automatic separation of the patient cohort into malignant and benign lesions. A total of 24 lesions were studied with 21 diagnosed accurately using the enhancement characteristics alone resulting in sensitivity and specificity of 100% and 73%, respectively. Analysis of the 2D MRS data demonstrated a significant difference (p < 0.05) in 12 of 18 metabolite ratios analyzed for malignant compared to benign lesions. Previous research focused on utilizing the choline signal to noise ratio (SNR) as a marker for malignancy has been verified using 2D MRS in this study. Using Fisher's linear discriminant test using water (WAT)/olefinic fat diagonal (UFD), choline (CHO)/fat (FAT), CHO/UFD, and FAT/methyl fat (FMETD) as predictors the sensitivity and specificity increased to 92% and 100%, respectively. Using the Classification and Regression Tree (CART) statistical analysis the resulting sensitivity and specificity were 100% and 91%, respectively, with the most accurate predictor for differentiating malignant and benign determined to be FAT/FMETD. The cases within the study that presented a indeterminate diagnosis using DCE-MRI alone were able to be accurately diagnosed when the metabolic information from 2D MRS was incorporated. The results suggest improved breast cancer detection through the combination of morphological and enhancement information from DCE-MRI and metabolic information from 2D MRS.

Magnetic resonance T2‐relaxometry and 2D L‐correlated spectroscopy in patients with minimal hepatic encephalopathy

To evaluate T(2)-relaxation changes in patients with minimal hepatic encephalopathy (MHE) using T(2) relaxometry and to correlate T(2) values with brain metabolites evaluated using 2D magnetic resonance spectroscopy (MRS). Eight MHE patients and 13 healthy subjects were evaluated using T(2) relaxometry, and eight patients and nine healthy subjects underwent 2D MRS in right frontal and left occipital regions. Whole-brain T(2)-relaxation maps were compared between MHE and control subjects using analysis-of-covariance, with age and gender included as covariates. T(2) values derived from the right frontal and left occipital lobes were correlated with the metabolite ratios. Multiple brain regions including anterior and mid cingulate cortices, right anterior and left posterior insular cortices, right prefrontal, medial frontal, and right superior temporal cortices showed significantly increased T(2) values in MHE patients compared to control subjects. MRS showed significantly increased ratios of glutamine/glutamate (Glx) and decreased ratios of myo-inositol, taurine, choline, and myo-inositol/choline (mICh) with respect to creatine (Cr_d) in patients compared to controls. Frontal Glx/Cr_d showed significantly positive correlation with T(2) values. MHE patients showed significantly increased T(2) values in multiple brain regions reflecting increased free water content and T(2) values in frontal lobe correlated with the increased Glx/Cr_d ratio.

TH-D-304A-03: Applications of Two Dimensional MR Spectroscopy in Breast Cancer Detection

Purpose:Magnetic Resonance Spectroscopy (MRS) allows noninvasive measurements of the concentrations of different metabolites in human breast in vivo. Two‐dimensional correlated spectroscopy (2D COSY) of breast tissues showed that in addition to Choline, several other metabolite ratios are different between malignant tumor and healthy tissues. The purpose of this study is to investigate the possibility of using Choline/Noise ratio (CNR) and other metabolite ratios from 2D COSY of breast tissues to detect breast cancer.Method and Materials: A total of 31 women participated in this study, including 13 healthy women, 9 subjects with malignant tumors, 9 women with benign tumors. Each 2D spectrum was acquired from malignant tumor, benign tumor, healthy fatty tissue or healthy glandular tissue. The size of each volume of interest was 1×1×1cm3, the total scan time was 12 minutes. All 2D MRS data files were processed using Felix2000 software package. Each 2D spectrum contains : Water (4.8, 4.8)ppm, Fat (1.4, 1.4)ppm ; Methyl Fat (0.9, 0.9) ppm ; Olefinic Fat (5.4, 5.4)ppm; Choline (3.3, 3.3)ppm; UFR(2.1, 5.4)ppm ; UFL (2.9, 5.4) ppm; TGFR (4.3, 5.3)ppm. CNR, 9 diagonal and 9 cross peak metabolite ratios were calculated for each spectrum. CNR and two statistical analysis methods (Fisher's linear discriminate, Classification and regression tree) were used to distinguish between malignant tumors from other breast tissues.Results: The CNR and 9 metabolite ratios are significantly different (p<=0.05) between malignant tissue and other breast tissues. Using both statistical analysis methods can get better detection accuracy than using only CNR. Conclusion: 2D MR spectroscopy can identify the major metabolites in human breast in vivo. CNR and different metabolite ratios can be used as biomarkers to distinguish between malignant tumors from other breast tissues. This method needs to be evaluated using a large cohort of patients.

Molecular classification of brain tumor biopsies using solid-state magic angle spinning proton magnetic resonance spectroscopy and robust classifiers.

Brain tumors are one of the leading causes of death in adults with cancer; however, molecular classification of these tumors with in vivo magnetic resonance spectroscopy (MRS) is limited because of the small number of metabolites detected. In vitro MRS provides highly informative biomarker profiles at higher fields, but also consumes the sample so that it is unavailable for subsequent analysis. In contrast, ex vivo high-resolution magic angle spinning (HRMAS) MRS conserves the sample but requires large samples and can pose technical challenges for producing accurate data, depending on the sample testing temperature. We developed a novel approach that combines a two-dimensional (2D), solid-state, HRMAS proton (1H) NMR method, TOBSY (total through-bond spectroscopy), which maximizes the advantages of HRMAS and a robust classification strategy. We used approximately 2 mg of tissue at -8 degrees C from each of 55 brain biopsies, and reliably detected 16 different biologically relevant molecular species. We compared two classification strategies, the support vector machine (SVM) classifier and a feed-forward neural network using the Levenberg-Marquardt back-propagation algorithm. We used the minimum redundancy/maximum relevance (MRMR) method as a powerful feature-selection scheme along with the SVM classifier. We suggest that molecular characterization of brain tumors based on highly informative 2D MRS should enable us to type and prognose even inoperable patients with high accuracy in vivo.

Area-selective signal parameter estimation for two-dimensional MR spectroscopy data

We consider the problem of parametric spectral analysis of two-dimensional (2D) magnetic resonance spectroscopy (MRS) data. Estimating the signal components from 2D MRS data is becoming common practice in many clinical MR applications. The most frequently used signal processing tool for this estimation problem is the non-parametric 2D-FFT. There are several alternative parametric methods available to perform this analysis, yet their computational complexity is generally rather high and it becomes prohibitive when the number of points in the measured data matrix is large. In this paper, we propose a novel signal parameter estimation technique which operates on a pre-specified sub-area of the 2D spectrum. This area-selective approach can be used either to estimate only the signal components of main interest in the data, or to compute signal parameter estimates of all present signal components as the computational burden for each sub-area is low. In the numerical example section we consider both simulated data and in vitro 1H data acquired from a 1.5 T MR scanner.

In vivo 2D magnetic resonance spectroscopy of small animals

Localized in vivo NMR spectroscopy, chemical shift imaging or multi-voxel spectroscopy are potentially useful tools in small animals that are complementary to MRI, adding biochemical information to the mainly anatomical data provided by imaging of water protons. However the contribution of such methods remains hampered by the low spectral resolution of the in vivo 1D spectra. Two-dimensional methods widely developed for in vitro studies have been proposed as suitable approaches to overcome these limitations in resolution. The different homonuclear and heteronuclear sequences adapted to in vivo studies are reviewed. Their specific contributions to the spectral resolution of spectroscopic data and their limitations for in vivo investigations are discussed. The applications to experimental models of pathological processes or pharmacological treatment in mainly brain and muscle are presented. According to their combined sensitivity, acquisition duration and spatial resolution, the heteronuclear 2D experiments, which are mainly used for 1H detected-13C spectroscopy after administration of 13C-labeled compounds, appear to be less efficient than 1H detected-13C 1D methods at high field. However, the applications of 2D proton homonuclear methods show that they remain the best tools for in vivo studies when an improved resolution is required.

Assessment of GABA concentration in human brain using two-dimensional proton magnetic resonance spectroscopy

A quantitative method to assess in vivo brain gamma-aminobutyric acid (GABA) levels is proposed using a J-resolved, two-dimensional (2D) magnetic resonance spectroscopy (MRS) technique. Localized, J-resolved 2D MR spectra were obtained from a 12-cm 3 voxel in the occipital lobe of 36 healthy volunteers (18 male and 18 female, age: 25.1±4.8 years). Based on phantom measurements, a GABA resonance peak located at 2.94 ppm, 7.45 Hz, in J-resolved 2D MRS overlaps the least with other resonance peaks which arise from N-acetylaspartate, choline, creatine, glutamate and glutamine. Measurements of this resonance peak yield in vivo GABA concentrations of 1.01±0.36 μmol/cm 3 for male and 1.16±0.43 μmol/cm 3 for female volunteers, without correction for T1 and T2 relaxation effects. These results are in good agreement with previously reported data and suggest that, with further development, 2D MRS may provide a practical means to estimate the concentration of this important neurotransmitter.

Necrotic tumor versus brain abscess: importance of amino acids detected at 1H MR spectroscopy--initial results.

To assess the usefulness of the 0.9-ppm peak from amino acids (-CH3 moieties from valine, leucine, and isoleucine) for the differentiation of brain abscesses and tumors at in vivo hydrogen 1 magnetic resonance (MR) spectroscopy. Amino acid concentrations were determined in vitro in 13 purulent samples from brain and nonbrain tissues and in nine aseptic fluids from necrotic brain tumors at two-dimensional (2D) 1H MR spectroscopy and liquid chromatography. Thirty-four patients with cystic intracerebral mass lesions (28 tumors, six abscesses) were examined at 1H MR spectroscopy in vivo. Amino acids were identified in vitro in both purulent and aseptic samples. Amino acid concentrations measured in the aseptic fluids at both liquid chromatography and 2D MR spectroscopy were far below the detection threshold of in vivo 1H MR spectroscopy. Quantitative results obtained at 2D MR spectroscopy showed no overlap in the ranges of amino acid concentrations in purulent and aseptic samples. In vivo, the proton spectra obtained with a 136-msec echo time (TE) revealed amino acids (inverted peak at 0.9 ppm) in only the abscesses. The detection of amino acid resonance at 0.9 ppm at in vivo 1H MR spectroscopy (136-msec TE) is a promising tool for distinguishing bacterial abscesses and cystic brain tumors.

Molecular classification of brain tumor biopsies using solid-state magic angle spinning proton magnetic resonance spectroscopy and robust classifiers

Brain tumors are one of the leading causes of death in adults with cancer; however, molecular classification of these tumors with in vivo magnetic resonance spectroscopy (MRS) is limited because of the small number of metabolites detected. In vitro MRS provides highly informative biomarker profiles at higher fields, but also consumes the sample so that it is unavailable for subsequent analysis. In contrast, ex vivo high-resolution magic angle spinning (HRMAS) MRS conserves the sample but requires large samples and can pose technical challenges for producing accurate data, depending on the sample testing temperature. We developed a novel approach that combines a two-dimensional (2D), solid-state, HRMAS proton (1H) NMR method, TOBSY (total through-bond spectroscopy), which maximizes the advantages of HRMAS and a robust classification strategy. We used ≈2 mg of tissue at -8°C from each of 55 brain biopsies, and reliably detected 16 different biologically relevant molecular species. We compared two classification strategies, the support vector machine (SVM) classifier and a feed-forward neural network using the Levenberg-Marquardt back-propagation algorithm. We used the minimum redundancy/maximum relevance (MRMR) method as a powerful feature-selection scheme along with the SVM classifier. We suggest that molecular characterization of brain tumors based on highly informative 2D MRS should enable us to type and prognose even inoperable patients with high accuracy in vivo.