Sreenath Pruthviraj Kyathanahally, André Döring, Roland Kreis

Ghosting artifacts in clinical MR spectroscopy are problematic since they superimpose with metabolites and lead to inaccurate quantification. Here, we make use of “Deep Learning” (DL) methods to remove ghosting artifacts in MR spectra of human brain. The DL method was trained on a huge database of simulated spectra with and without ghosting artifacts, which represent complex variants of ghost-ridden spectra, transformed to time-frequency spectrograms. The trained model was tested on simulated and in-vivo spectra. The preliminary results for ghost removal show potential in simulated and in-vivo spectra, but need further refinement and quantitative testing.

Ipshita Bhattacharya, Ralph Noeske , Rolf Schulte, Mathews Jacob

Spectral interleaving is often used in echoplanar spectroscopic imaging (EPSI) sequences to achieve high spatial and spectral resolution, especially on high field scanners with larger chemical shift dispersion. Unfortunately, a major roadblock is the spurious Nyquist ghost artifacts, resulting from phase errors between interleaves. We introduce a novel framework, that simultaneously capitalizes on annihilation relation between the interleaves introduced by phase relations, as well as a linear predicability of the spectra, to remove the phase errors and to provide spectral denoising of the spectra. In addition, we also exploit on the low-rank structure of the EPSI data to provide additional spatial denoising, which will further improve the signal to noise ratio of the datasets.

Guilhem Pagès, Jean-Marie Bonny

Following MRS data fitting, absolute or relative Cramér-Rao bounds (CRB and rCRB, respectively) are often computed as indices of parameter uncertainties. However, the unknown true values of parameters are required to compute CRB. Here, we studied the effect of substituting the true values by the noisy estimates on the bounds (noted CRB* and rCRB*). We showed by simulations that both CRB* and rCRB* are particularly sensitive to noise. Also, the mode of rCRB* distribution is left-shifted, which leads to a significant false positive risk when rCRB* is thresholded. A threshold not exceeding 20% is recommended to limit this risk.

Michal Považan, Mark Mikkelsen, Adam Berrington, Peter B. Barker, Pallab K. Bhattacharyya, Maiken K. Brix, Pieter F. Buur, Kim M. Cecil, Kimberly L. Chan, David Y.-T. Chen, Alexander R. Craven, Koen Cuypers, Michael Dacko, Niall W. Duncan, Ulrike Dydak, David A. Edmondson, Gabriele Ende, Lars Ersland, Fei Gao, Ian Greenhouse, Ashley D. Harris, Naying He, Stefanie Heba, Nigel Hoggard, Tun-Wei Hsu, Jacobus F. A. Jansen, Alayar Kangarlu, Thomas Lange, R. Marc Lebel, Yan Li, Chien-Yuan E. Lin, Jy-Kang Liou, Jiing-Feng Lirng, Feng Liu, Ruoyun Ma, Celine Maes, Marta Moreno-Ortega, Scott O. Murray, Sean Noah, Ralph Noeske, Michael D. Noseworthy, Georg Oeltzschner, James J. Prisciandaro, Nicolaas A. J. Puts, Timothy P. L. Roberts, Markus Sack, Napapon Sailasuta, Muhammad G. Saleh, Michael-Paul Schallmo, Nicholas Simard, Stephan P. Swinnen, Martin Tegenthoff, Peter Truong, Guangbin Wang, Iain D. Wilkinson, Hans-Jörg Wittsack, Hongmin Xu, Fuhua Yan, Chencheng Zhang, Vadim Zipunnikov, Helge J. Zöllner, Richard A. E. Edden

In vivo ¹H MR spectroscopy (MRS) provides valuable information regarding various brain disorders. In order to enable further clinical acceptance of MRS, this technique needs to be robust across multiple sites and MR vendors. Therefore, we analyzed data from 288 healthy subjects from 25 research sites across the three major vendors and examined vendor-, site- and participant-related effects on metabolites detected by PRESS ¹H-MRS at 3T. Within-site and inter-site coefficients of variation were between 2-16%. Significant effects of vendor were found for Ins/tCr and Glx/tCr. Effect of sites contributed ~30% to total variance of all quantified metabolites.  

Kelley Swanberg, Hetty Prinsen, Abhinav Kurada, Katherine DeStefano, Mary Bailey, David Pitt, Robert Fulbright, Christoph Juchem

Water is a common internal reference for metabolite quantification by ¹H-MRS. We investigate potential influences on water-referenced metabolite quantification by differences in frontal cortex water T₂ in individuals with relapsing-remitting, progressive, and no multiple sclerosis. Water T₂ differed in monoexponential models, exhibiting highest values in progressive multiple sclerosis only when analyses were not age-controlled. Groupwise T₂ did not differ in biexponential models constrained by tissue and CSF partial volumes, suggesting that monoexponential T₂ differences reflected disparate proportions of water in tissue and CSF rather than differential behavior within them. Our results suggest stability of water T₂ within frontal cortex tissue and CSF with multiple sclerosis and emphasize the superiority of metabolite quantification with group-specific T₂ values when voxel composition may differ.

Dhritiman Das, Mike Davies, Jeremy Chataway, Siddharthan Chandran, Bjoern Menze, Ian Marshall

As an alternative to model-based spectral fitting tools, we introduce a machine-learning framework for estimating metabolite concentrations in MR spectra acquired from a homogeneous cohort of 42 patients with Secondary Progressive Multiple Sclerosis. Our framework based on random-forest regression performs a 42-fold cross validation on this dataset which involves (1) learning the spectral features from this cohort; (2) estimating concentrations and calculating relative error over the LCModel estimates. Compared to the LCModel, our method, after training, gives a low estimation error and a 60-fold improvement in estimation speed per patient.

Lorenz Pfleger, Emina Halilbasic, Martin Gajdošík, Marek Chmelík, Sigfried Trattnig, Michael Trauner, Martin Krššák

This study focuses on quantification of phosphatidylcholine (PtdC) in the gall bladder and investigates concentration differences between biliary pathologies and healthy subjects. Significant different PtdC content was detected in patients suffering from primary sclerosing cholangitis (PSC). Subjects with primary biliary cholangitis (PBC) showed high variances in PtdC concentration. Even though, the PSC and control group are relatively small our results justify for further ongoing studies on this topic.

Meiyappan Solaiyappan, Robert Weiss, Paul Bottomley

Myocardial energy demands are the highest in the body and cardiac metabolism is altered in common diseases. Only phosphorus magnetic resonance spectroscopy (MRS) can measure ATP and creatine-kinase (CK) metabolism, a primary reserve of ATP, noninvasively in the human heart. Here, neural-network analysis is used to test whether the combination of ³¹P MRS measurements of phosphocreatine and [ATP] concentrations, the CK reaction-rate and its ATP flux, can discriminate cardiac diseases among prior study data from 178 subjects. We find that a three-layer neural-network adequately discriminates diseases without over-training, suggesting that heretofore unidentified differences in CK metabolism may underlie cardiac disease.

Gil Farkash, Stefan Markovic , Lucio Frydman

Hyperpolarized ¹³C magnetic resonance spectroscopic imaging (MRSI) is a powerful metabolic technique, but it’s challenged by a rapid and irreversible decay of the signal that usually compromises its achievable spatial resolution. In this work we explore a way to improve this by utilizing a priori anatomical information derived from ¹H MRI. Enhanced HP-MRSI implementations based on Spectroscopy with Linear Algebraic Modeling (SLAM) were thus assayed, to enhance HP-MRSI’s spatial resolution without compromising SNR. ¹³C experiments were performed in-vivo and pyruvate/lactate images reconstructed for physiological compartments by SLAM; we compare these results to those arising by traditional Fourier analyses.

Yudu Li, Fan Lam, Bryan Clifford, Rong Guo, Xi Peng, Zhi-Pei Liang

Reconstruction for MR spectroscopic imaging (MRSI) is a challenging problem where incorporation of spatiospectral prior information is often necessary. While spectral constraints have been effectively utilized in the form of temporal basis functions, spatial constraints are often imposed using spatial regularization. In this work, we present a new kernel-based method to incorporate a priori spatial information, which was motivated by the success of kernel-based methods in machine learning. It provides a new mechanism for constrained image reconstruction, effectively incorporating a priori spatial information. The proposed method has been evaluated using both simulation and in vivo data, producing very impressive results. This new reconstruction scheme can be used to process any MRSI data, especially those from high-resolution MRSI experiments.