Relaxation‐compensated APT and rNOE CEST‐MRI of human brain tumors at 3 T

Self Cite

Purpose Dynamic glucose‐enhanced (DGE)‐MRI based on chemical exchange‐sensitive MRI, that is, glucoCEST and gluco‐chemical exchange‐sensitive spin‐lock (glucoCESL), is intrinsically prone to motion‐induced artifacts because the final DGE contrast relies on the difference of images, which were acquired with a time gap of several mins. In this study, identification of different types of motion‐induced artifacts led to the development of a 3D acquisition protocol for DGE examinations in the human brain at 7 T with improved robustness in the presence of subject motion. Methods DGE‐MRI was realized by the chemical exchange‐sensitive spin‐lock approach based either on relaxation rate in the rotating frame (R1ρ)‐weighted or quantitative R1ρ imaging. A 3D image readout was implemented at 7 T, enabling retrospective volumetric coregistration of the image series and quantification of subject motion. An examination of a healthy volunteer without administration of glucose allowed for the identification of isolated motion‐induced artifacts. Results Even after coregistration, significant motion‐induced artifacts remained in the DGE contrast based on R1ρ‐weighted images. This is due to the spatially varying sensitivity of the coil and was found to be compensated by a quantitative R1ρ approach. The coregistered quantitative approach allowed the observation of a clear increase of the DGE contrast in a patient with glioblastoma, which did not correlate with subject motion. Conclusion The presented 3D acquisition protocol enables DGE‐MRI examinations in the human brain with improved robustness against motion‐induced artifacts. Correction of motion‐induced artifacts is of high importance for DGE‐MRI in clinical studies where an unambiguous assignment of contrast changes due to an actual change in local glucose concentration is a prerequisite.

Section: Discussionmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Dynamic glucose‐enhanced (DGE) MRI in the human brain at 7 T with reduced motion‐induced artifacts based on quantitative R_1ρ mapping

Boyd

Breitling

Zimmermann

et al. 2019

Self Cite

“…An overview of the presented data processing pipeline is given in Figure . In vivo Z‐spectra are acquired at 3T and evaluated by conventional methods (PCA denoising, Lorentzian fitting). The obtained results are used to train a deep neural network to directly map from raw Z‐spectra to Lorentzian parameters, yielding CEST contrast maps in ~1 s where conventional evaluation takes ~10 min.…”

Section: Methodsmentioning

confidence: 99%

“…The Z‐spectra were spectrally de‐noised by principal component analysis (PCA) following the approach of Breitling et al, keeping only the principal components suggested by the median criterion (Supporting Information Table ). To isolate CEST effects, a four‐pool Lorentzian fit model according to Goerke et al was used to fit the direct water saturation (DS), semi‐solid magnetization transfer (MT), amide proton transfer (APT), and relayed NOE peaks, using the model equation

Z (Δ ω) = c - L_{DS} - L_{APT} - L_{NOE} - L_{MT},

with a constant c, the direct saturation pool

L_{italicDS} = \frac{A_{DS}}{1 + {()(, \frac{Δ ω - {normalδ}_{DS}}{{normalΓ}_{DS} / 2})}^{2}},

and the other pools

L_{i} = \frac{A_{i}}{1 + {()(, \frac{Δ ω - {normalδ}_{DS} - {normalδ}_{i}}{{normalΓ}_{i} / 2})}^{2}}, i \in \{APT, NOE, MT\},

with amplitudes A i , full‐width‐at‐half‐maximum Γ i , and peak positions δ i . This model has 10 free parameters in total (4 amplitudes A i , 4 widths Γ i , water peak position δ DS , and constant c ), as the positions of APT, NOE, and MT peaks were fixed at +3.5 ppm, −3.5 ppm, and −2.5 ppm relative to the water peak position, respectively, and were therefore not treated as free parameters.…”

Section: Methodsmentioning

confidence: 99%

“…Numbers are given for the 3 subject data sets used for training, as well as for the healthy and the tumor patient test data set TABLE S2 Initial values and boundary conditions for the conventional least squares fit of the 4 pool Lorentzian model described in the methods section (Equations and ). Fit model, initial values, and boundary conditions are similar to the ones proposed by Goerke et al TABLE S3 Results of hyperparameter optimization combining the outcomes of the experiments addressing dropout, architecture, and activation function…”

mentioning

confidence: 99%

See 1 more Smart Citation

DeepCEST 3T: Robust MRI parameter determination and uncertainty quantification with neural networks—application to CEST imaging of the human brain at 3T

Glang

Deshmane

Prokudin

et al. 2019

Self Cite

Purpose Calculation of sophisticated MR contrasts often requires complex mathematical modeling. Data evaluation is computationally expensive, vulnerable to artifacts, and often sensitive to fit algorithm parameters. In this work, we investigate whether neural networks can provide not only fast model fitting results, but also a quality metric for the predicted values, so called uncertainty quantification, investigated here in the context of multi‐pool Lorentzian fitting of CEST MRI spectra at 3T. Methods A deep feed‐forward neural network including a probabilistic output layer allowing for uncertainty quantification was set up to take uncorrected CEST‐spectra as input and predict 3T Lorentzian parameters of a 4‐pool model (water, semisolid MT, amide CEST, NOE CEST), including the B0 inhomogeneity. Networks were trained on data from 3 subjects with and without data augmentation, and applied to untrained data from 1 additional subject and 1 brain tumor patient. Comparison to conventional Lorentzian fitting was performed on different perturbations of input data. Results The deepCEST 3T networks provided fast and accurate predictions of all Lorentzian parameters and were robust to input perturbations because of noise or B0 artifacts. The uncertainty quantification detected fluctuations in input data by increase of the uncertainty intervals. The method generalized to unseen brain tumor patient CEST data. Conclusions The deepCEST 3T neural network provides fast and robust estimation of CEST parameters, enabling online reconstruction of sophisticated CEST contrast images without the typical computational cost. Moreover, the uncertainty quantification indicates if the predictions are trustworthy, enabling confident interpretation of contrast changes.

A novel normalization for amide proton transfer CEST MRI to correct for fat signal–induced artifacts: application to human breast cancer imaging

Zimmermann

Korzowski

Breitling

et al. 2019

Self Cite

Purpose: The application of amide proton transfer (APT) CEST MRI for diagnosis of breast cancer is of emerging interest. However, APT imaging in the human breast is affected by the ubiquitous fat signal preventing a straightforward application of existing acquisition protocols. Although the spectral region of the APT signal does not coincide with fat resonances, the fat signal leads to an incorrect normalization of the Z-spectrum, and therefore to distorted APT effects. In this study, we propose a novel normalization for APT-CEST MRI that corrects for fat signal-induced artifacts in the postprocessing without the need for application of fat saturation schemes or water-fat separation approaches. Methods: The novel normalization uses the residual signal at the spectral position of the direct water saturation to estimate the fat contribution. A comprehensive theoretical description of the normalization for an arbitrary phase relation of the water and fat signal is provided. Functionality and applicability of the proposed normalization was demonstrated by in vitro and in vivo experiments. Results: In vitro, an underestimation of the conventional APT contrast of approximately −1.2% per 1% fat fraction was observed. The novel normalization yielded an APT contrast independent of the fat contribution, which was also independent of the | 921 ZIMMERMANN Et Al.