TensorFit: A torch-based tool for ultrafast metabolite fitting of large MRSI data sets.

PURPOSE: To introduce a tool (TensorFit) for ultrafast and robust metabolite fitting of MRSI data based on Torch's auto-differentiation and optimization framework. METHODS: TensorFit was implemented in Python based on Torch's auto-differentiation to fit individual metabolites in MRS spectra. The underlying time domain and/or frequency domain fitting model is based on a linear combination of metabolite spectroscopic response. The computational time efficiency and accuracy of TensorFit were tested on simulated and in vivo MRS data and compared against TDFDFit and QUEST. RESULTS: TensorFit demonstrates a significant improvement in computation speed, achieving a 165-times acceleration compared with TDFDFit and 115 times against QUEST. TensorFit showed smaller percentual errors on simulated data compared with TDFDFit and QUEST. When tested on in vivo data, it performed similarly to TDFDFit with a 2% better fit in terms of mean squared error while obtaining a 169-fold speedup. CONCLUSION: TensorFit enables fast and robust metabolite fitting in large MRSI data sets compared with conventional metabolite fitting methods. This tool could boost the clinical applicability of large 3D MRSI by enabling the fitting of large MRSI data sets within computation times acceptable in a clinical environment.

Simultaneous multi-region detection of GABA+ and Glx using 3D spatially resolved SLOW-editing and EPSI-readout at 7T.

GABA+ and Glx (glutamate and glutamine) are widely studied metabolites, yet the commonly used magnetic resonance spectroscopy (MRS) techniques have significant limitations, including sensitivity to B0 and B1+-inhomogeneities, limited bandwidth of MEGA-pulses, high SAR which is accentuated at 7T. To address these limitations, we propose SLOW-EPSI method, employing a large 3D MRSI coverage and achieving a high resolution down to 0.26 ml. Simulation results demonstrate the robustness of SLOW-editing for both GABA+ and Glx against B0 and B1+-inhomogeneities within the range of [-0.3, +0.3] ppm and [40 %, 250 %], respectively. Two protocols, both utilizing a 70 mm thick FOV slab, were employed to target distinct brain regions in vivo, differentiated by their orientation: transverse and tilted. Protocol 1 (n = 11) encompassed 5 locations (cortical gray matter, white matter, frontal lobe, parietal lobe, and cingulate gyrus). Protocol 2 (n = 5) involved 9 locations (cortical gray matter, white matter, frontal lobe, occipital lobe, cingulate gyrus, caudate nucleus, hippocampus, putamen, and inferior thalamus). Quantitative analysis of GABA+ and Glx was conducted in a stepwise manner. First, B1+/B1--inhomogeneities were corrected using water reference data. Next, GABA+ and Glx values were calculated employing spectral fitting. Finally, the GABA+ level for each selected region was compared to the global Glx within the same subject, generating the GABA+/Glx_global ratio. Our findings from two protocols indicate that the GABA+/Glx_global level in cortical gray matter was approximately 16 % higher than in white matter. Elevated GABA+/Glx_global levels acquired with protocol 2 were observed in specific regions such as the caudate nucleus (0.118±0.067), putamen (0.108±0.023), thalamus (0.092±0.036), and occipital cortex (0.091±0.010), when compared to the cortical gray matter (0.079±0.012). Overall, our results highlight the effectiveness of SLOW-EPSI as a robust and efficient technique for accurate measurements of GABA+ and Glx at 7T. In contrast to previous SVS and 2D-MRSI based editing sequences with which only one or a limited number of brain regions can be measured simultaneously, the method presented here measures GABA+ and Glx from any brain area and any arbitrarily shaped volume that can be flexibly selected after the examination. Quantification of GABA+ and Glx across multiple brain regions through spectral fitting is achievable with a 9-minute acquisition. Additionally, acquisition times of 18-27 min (GABA+) and 9-18 min (Glx) are required to generate 3D maps, which are constructed using Gaussian fitting and peak integration.

Deep learning pipeline for quality filtering of MRSI spectra.

With the rise of novel 3D magnetic resonance spectroscopy imaging (MRSI) acquisition protocols in clinical practice, which are capable of capturing a large number of spectra from a subject's brain, there is a need for an automated preprocessing pipeline that filters out bad-quality spectra and identifies contaminated but salvageable spectra prior to the metabolite quantification step. This work introduces such a pipeline based on an ensemble of deep-learning classifiers. The dataset consists of 36,338 spectra from one healthy subject and five brain tumor patients, acquired with an EPSI variant, which implemented a novel type of spectral editing named SLOtboom-Weng (SLOW) editing on a 7T MR scanner. The spectra were labeled manually by an expert into four classes of spectral quality as follows: (i) noise, (ii) spectra greatly influenced by lipid-related artifacts (deemed not to contain clinical information), (iii) spectra containing metabolic information slightly contaminated by lipid signals, and (iv) good-quality spectra. The AI model consists of three pairs of networks, each comprising a convolutional autoencoder and a multilayer perceptron network. In the classification step, the encoding half of the autoencoder is kept as a dimensionality reduction tool, while the fully connected layers are added to its output. Each of the three pairs of networks is trained on different representations of spectra (real, imaginary, or both), aiming at robust decision-making. The final class is assigned via a majority voting scheme. The F1 scores obtained on the test dataset for the four previously defined classes are 0.96, 0.93, 0.82, and 0.90, respectively. The arguably lower value of 0.82 was reached for the least represented class of spectra mildly influenced by lipids. Not only does the proposed model minimise the required user interaction, but it also greatly reduces the computation time at the metabolite quantification step (by selecting a subset of spectra worth quantifying) and enforces the display of only clinically relevant information.

α-D-Glucose as a non-radioactive MRS tracer for metabolic studies of the brain.

Changes in brain glucose metabolism occur in many neurological disorders as well as during aging. Most studies on the uptake of glucose in the brain use positron emission tomography, which requires injection of a radioactive tracer. Our study shows that ultra-high-field 1H-MRS can be used to measure α-D-glucose at 5.22 ppm in vivo, and the α-D-glucose can be used as a radiation-free tracer in the human brain.

Accurate prediction of isocitrate dehydrogenase -mutation status of gliomas using SLOW-editing magnetic resonance spectroscopic imaging at 7 T MR.

Background: 2-hydroxy-glutarate (2HG) is a metabolite that accumulates in isocitrate dehydrogenase (IDH)-mutated gliomas and can be detected noninvasively using MR spectroscopy. However, due to the low concentration of 2HG, established magnetic resonance spectroscopic imaging (MRSI) techniques at the low field have limitations with respect to signal-to-noise and to the spatial resolution that can be obtained within clinically acceptable measurement times. Recently a tailored editing method for 2HG detection at 7 Tesla (7 T) named SLOW-EPSI was developed. The underlying prospective study aimed to compare SLOW-EPSI to established techniques at 7 T and 3 T for IDH-mutation status determination. Methods: The applied sequences were MEGA-SVS and MEGA-CSI at both field strengths and SLOW-EPSI at 7 T only. Measurements were performed on a MAGNETOM-Terra 7 T MR-scanner in clinical mode using a Nova 1Tx32Rx head coil and on a 3 T MAGNETOM-Prisma scanner with a standard 32-channel head coil. Results: Fourteen patients with suspected glioma were enrolled. Histopathological confirmation was available in 12 patients. IDH mutation was confirmed in 9 out of 12 cases and 3 cases were characterized as IDH wildtype. SLOW-EPSI at 7 T showed the highest accuracy for IDH-status prediction (91.7% accuracy, 11 of the 12 predictions correct with 1 false negative case). At 7 T, MEGA-CSI had an accuracy of 58.3% and MEGA-SVS had an accuracy of 75%. At 3 T, MEGA-CSI showed an accuracy of 63.6% and MEGA-SVS of 33.3%. The co-edited cystathionine was detected in 2 out of 3 oligodendroglioma cases with 1p/19q codeletion. Conclusions: Depending on the pulse sequence, spectral editing can be a powerful tool for the noninvasive determination of the IDH status. SLOW-editing EPSI sequence is the preferable pulse sequence when used at 7 T for IDH-status characterization.

SLOW: A novel spectral editing method for whole-brain MRSI at ultra high magnetic field.

PURPOSE: At ultra-high field (UHF), B1+ -inhomogeneities and high specific absorption rate (SAR) of adiabatic slice-selective RF-pulses make spatial resolved spectral-editing extremely challenging with the conventional MEGA-approach. The purpose of the study was to develop a whole-brain resolved spectral-editing MRSI at UHF (UHF, B0 ≥ 7T) within clinical acceptable measurement-time and minimal chemical-shift-displacement-artifacts (CSDA) allowing for simultaneous GABA/Glx-, 2HG-, and PE-editing on a clinical approved 7T-scanner. METHODS: Slice-selective adiabatic refocusing RF-pulses (2π-SSAP) dominate the SAR to the patient in (semi)LASER based MEGA-editing sequences, causing large CSDA and long measurement times to fulfill SAR requirements, even using SAR-minimized GOIA-pulses. Therefore, a novel type of spectral-editing, called SLOW-editing, using two different pairs of phase-compensated chemical-shift selective adiabatic refocusing-pulses (2π-CSAP) with different refocusing bandwidths were investigated to overcome these problems. RESULTS: Compared to conventional echo-planar spectroscopic imaging (EPSI) and MEGA-editing, SLOW-editing shows robust refocusing and editing performance despite to B1+ -inhomogeneity, and robustness to B0 -inhomogeneities (0.2 ppm ≥ ΔB0  ≥ -0.2 ppm). The narrow bandwidth (∼0.6-0.8 kHz) CSAP reduces the SAR by 92%, RF peak power by 84%, in-excitation slab CSDA by 77%, and has no in-plane CSDA. Furthermore, the CSAP implicitly dephases water, lipid and all the other signals outside of range (≥ 4.6 ppm and ≤1.4 ppm), resulting in additional water and lipid suppression (factors ≥ 1000s) at zero SAR-cost, and no spectral aliasing artifacts. CONCLUSION: A new spectral-editing has been developed that is especially suitable for UHF, and was successfully applied for 2HG, GABA+, PE, and Glx-editing within 10 min clinical acceptable measurement time.