Deep learning can accelerate and quantify simulated localized correlated spectroscopy

Abstract

Nuclear magnetic resonance spectroscopy (MRS) allows for the determination of atomic structures and concentrations of different chemicals in a biochemical sample of interest. MRS is used in vivo clinically to aid in the diagnosis of several pathologies that affect metabolic pathways in the body. Typically, this experiment produces a one dimensional (1D) 1H spectrum containing several peaks that are well associated with biochemicals, or metabolites. However, since many of these peaks overlap, distinguishing chemicals with similar atomic structures becomes much more challenging. One technique capable of overcoming this issue is the localized correlated spectroscopy (L-COSY) experiment, which acquires a second spectral dimension and spreads overlapping signal across this second dimension. Unfortunately, the acquisition of a two dimensional (2D) spectroscopy experiment is extremely time consuming. Furthermore, quantitation of a 2D spectrum is more complex. Recently, artificial intelligence has emerged in the field of medicine as a powerful force capable of diagnosing disease, aiding in treatment, and even predicting treatment outcome. In this study, we utilize deep learning to: (1) accelerate the L-COSY experiment and (2) quantify L-COSY spectra. All training and testing samples were produced using simulated metabolite spectra for chemicals found in the human body. We demonstrate that our deep learning model greatly outperforms compressed sensing based reconstruction of L-COSY spectra at higher acceleration factors. Specifically, at four-fold acceleration, our method has less than 5% normalized mean squared error, whereas compressed sensing yields 20% normalized mean squared error. We also show that at low SNR (25% noise compared to maximum signal), our deep learning model has less than 8% normalized mean squared error for quantitation of L-COSY spectra. These pilot simulation results appear promising and may help improve the efficiency and accuracy of L-COSY experiments in the future.

Subnormothermic Ex Vivo Porcine Kidney Perfusion Improves Energy Metabolism: Analysis Using 31P Magnetic Resonance Spectroscopic Imaging

“… 10 , 33 Recent improvements using deep learning algorithms were used to reduce concentration estimation bias of metabolites with overlapping spectra. 34 pRMSI also suffers from a relative low sensitivity compared with liquid chromatography or 1H imaging at a constant magnetic field. 10 Thus, the acquisition is generally performed with higher voxel size to achieve enough signal to noise ratio while keeping an acceptable scan time.…”

Emerging MR Imaging and Spectroscopic Methods to Study Brain Tumor Metabolism

“…On conventional one-dimensional (1D) 1 H-MRS, spectral peaks due to methyl, methylene, and methane protons from the number of metabolites, namely, NAA, N-acetyl aspartyl glutamate (NAAG), Glu, Gln, GABA, and 2HG neurometabolic peaks severely overlap in the spectral region of 2–4 ppm, often confounding the detection and quantification of metabolite concentrations. In contrast, the 2D-COSY method offers the ability to identify potentially overlapping resonances of metabolites by dispersing the multiplet structure of scalar (J)-coupled spin systems into a second spectral dimension and by exploiting the unlikely possibility that two metabolites would share identical chemical shifts in two dimensions ( 52 – 55 ). A basic pulse sequence of a 2D-COSY consists of preparation time that allows the nuclei in the sample to reach equilibrium with the static external magnetic field environment and during which water suppression is performed.…”

“…Quantification of metabolic resonances detected on 2D-COSY could also be improved by implementing a prior-knowledge-based fitting approach analogous to linear combination (LC)-model fitting on 1D-MRS rather than using a peak integration method that is generally used. ProFit is one such 2D prior-knowledge-based fitting algorithm, and adopting this program could improve the reliable quantification of brain tumor metabolites ( 52 , 71 , 72 ). Future developments also include designing multivoxel-based 2D-COSY sequences using concentric circular echoplanar encoding or spiral encoding schemes for facilitating faster data with greater anatomical coverage and higher spatial resolution.…”

Improved signal‐to‐noise performance of MultiNet GRAPPA 1H FID MRSI reconstruction with semi‐synthetic calibration data

“…Relative to anatomical imaging, adaption to MRSI is more difficult due to the lower SNR of metabolites which are orders of magnitude lower than that of nuisance signals such as water and lipids. While machine‐learning approaches have been applied to MRS, these applications have mostly been limited to spectral fitting and artifact removal 26–30 …”

“…While machine-learning approaches have been applied to MRS, these applications have mostly been limited to spectral fitting and artifact removal. [26][27][28][29][30] While k-space-based NN reconstructions require less calibration data as a whole, it has long been documented that NNs perform better with more training data [31][32][33] which allows for greater accuracy. Hence, several strategies for better NN training for MultiNet GRAPPA MRSI reconstruction are investigated and compared against training with one image herein.…”