Metabolite selection for machine learning in childhood brain tumour classification

MRS can provide high accuracy in the diagnosis of childhood brain tumours when combined with machine learning. A feature selection method such as principal component analysis is commonly used to reduce the dimensionality of metabolite profiles prior to classification. However, an alternative approach of identifying the optimal set of metabolites has not been fully evaluated, possibly due to the challenges of defining this for a multi-class problem. This study aims to investigate metabolite selection from in vivo MRS for childhood brain tumour classification. Multi-site 1.5 T and 3 T cohorts of patients with a brain tumour and histological diagnosis of ependymoma, medulloblastoma and pilocytic astrocytoma were retrospectively evaluated. Dimensionality reduction was undertaken by selecting metabolite concentrations through multi-class receiver operating characteristics and compared with principal component analysis. Classification accuracy was determined through leave-one-out and k-fold cross-validation. Metabolites identified as crucial in tumour classification include myo-inositol (P < 0.05, $\text{◂,▸}\text{AUC}=0.81\pm 0.01$), total lipids and macromolecules at 0.9 ppm (P < 0.05, $\text{◂,▸}\text{AUC}=0.78\pm 0.01$) and total creatine (P < 0.05, $\text{◂,▸}\text{AUC}=0.77\pm 0.01$) for the 1.5 T cohort, and glycine (P < 0.05, $\text{◂,▸}\text{AUC}=0.79\pm 0.01$), total N-acetylaspartate (P < 0.05, $\text{◂,▸}\text{AUC}=0.79\pm 0.01$) and total choline (P < 0.05, $\text{◂,▸}\text{AUC}=0.75\pm 0.01$) for the 3 T cohort. Compared with the principal components, the selected metabolites were able to provide significantly improved discrimination between the tumours through most classifiers (P < 0.05). The highest balanced classification accuracy determined through leave-one-out cross-validation was 85% for 1.5 T ¹H-MRS through support vector machine and 75% for 3 T ¹H-MRS through linear discriminant analysis after oversampling the minority. The study suggests that a group of crucial metabolites helps to achieve better discrimination between childhood brain tumours.     

Synthetic MRI improves radiomics-based glioblastoma survival prediction

Glioblastoma is an aggressive and fast-growing brain tumor with poor prognosis. Predicting the expected survival of patients with glioblastoma is a key task for efficient treatment and surgery planning. Survival predictions could be enhanced by means of a radiomic system. However, these systems demand high numbers of multicontrast images, the acquisitions of which are time consuming, giving rise to patient discomfort and low healthcare system efficiency. Synthetic MRI could favor deployment of radiomic systems in the clinic by allowing practitioners not only to reduce acquisition time, but also to retrospectively complete databases or to replace artifacted images. In this work we analyze the replacement of an actually acquired MR weighted image by a synthesized version to predict survival of glioblastoma patients with a radiomic system. Each synthesized version was realistically generated from two acquired images with a deep learning synthetic MRI approach based on a convolutional neural network. Specifically, two weighted images were considered for the replacement one at a time, a T2w and a FLAIR, which were synthesized from the pairs T1w and FLAIR, and T1w and T2w, respectively. Furthermore, a radiomic system for survival prediction, which can classify patients into two groups (survival >480 days and $\le$ 480 days), was built. Results show that the radiomic system fed with the synthesized image achieves similar performance compared with using the acquired one, and better performance than a model that does not include this image. Hence, our results confirm that synthetic MRI does add to glioblastoma survival prediction within a radiomics-based approach.     

Relations between property versus molar mass relationships, 1. Sedimentation coefficient versus molar mass and diffusion coefficient versus molar mass relationships

The general validity of the “Exponents' Rule”, i.e. the interdependencies of the exponents from exponential property versus molar mass M relationships, is based on the general definition for the average of a molar mass dependent property. From this, the numerical value of the product K_sD · K_s · K_D^-1 of the pre-exponential constants of the sedimentation constant (s) versus molar mass relationship, K_s, the diffusion coefficient (D) versus molar mass relationship, K_D, and the constant of Svedberg's equation \documentclass{article}\pagestyle{empty} is found to be dimensionally and numerically unity (\documentclass{article}\pagestyle{empty}: partial specific volume of polymer; ρ₁ : density of solvent): The experimental validity of this “Product of Constants” to be unity is proven by evaluating 26 different s versus M and D versus M relationships of 8 different polymer/solvent (temperature) systems from literature, which are based on mutually independent measurements of s, D, and M.     

Correlation of hyperpolarized 13C‐MRI data with tissue extract measurements

Hyperpolarized (HP) MRI provides the means to monitor lactate metabolism noninvasively in tumours. Since ‐lactate signal levels obtained from HP imaging depend on multiple factors, such as the rate of substrate delivery via the vasculature, the expression level of monocarboxylate transporters (MCTs) and lactate dehydrogenase (LDH), and the local lactate pool size, the interpretation of HP metabolic images remains challenging. In this study, ex vivo tissue extract measurements (i.e., NMR isotopomer analysis, western blot analysis) derived from an MDA‐MB‐231 xenograft model in nude rats were used to test for correlations between the in vivo data and the ex vivo measures. The lactate‐to‐pyruvate ratio from HP MRI was strongly correlated with [1‐ ]lactate concentration measured from the extracts using NMR (R = 0.69, p 0.05), as well as negatively correlated with tumour wet weight (R =  0.60, p 0.05). In this tumour model, both MCT1 and MCT4 expressions were positively correlated with wet weight ( = 0.78 and 0.93, respectively, p 0.01). Lactate pool size and the lactate‐to‐pyruvate ratio were not significantly correlated.     

Measurement of brain lactate during visual stimulation using a long TE semi‐LASER sequence at 7 T

Estimation of metabolic changes during neuronal activation represents a challenge for in vivo MRS, especially for metabolites with low concentration and signal overlap, such as lactate. In this work, we aimed to evaluate the feasibility of detecting lactate during brain activation using a long (144 ms) semi‐LASER sequence at 7 T. spectra were acquired on healthy volunteers ( ) during a paradigm with 15 min of visual stimulation. Outer‐volume signals were further attenuated by the use of saturation slabs, and macromolecular signals in the vicinity of the inverted lactate peak were individually fitted with simulated Lorentzian peaks. All spectra were free of artefacts and highly reproducible across subjects. Lactate was accurately quantified with an average Cramér‐Rao lower bound of 8%. Statistically significant ( , one‐tailed ‐test) increases in lactate ( 10%) and glutamate ( 3%) levels during stimulation were detected in the visual cortex. Lactate and glutamate changes were consistent with previous measurements. We demonstrated that quantification of a clear and non‐contaminated lactate peak obtained with a long TE sequence has the potential of improving the accuracy of functional MRS studies targeting non‐oxidative reaction pathways.     

Demonstration and suppression of respiration‐related artifacts in Bloch–Siegert shift‐based B1+ maps of the human brain

Respiration‐induced movement of the chest wall and internal organs causes temporal B₀ variations extending throughout the brain. This study demonstrates that these variations can cause significant artifacts in maps obtained at 7 T with the Bloch–Siegert shift (BSS) mapping technique. To suppress these artifacts, a navigator correction scheme was proposed. Two sets of experiments were performed. In the first set of experiments, phase shifts induced by respiration‐related B₀ variations were assessed for five subjects at 7 T by using a gradient echo (GRE) sequence without phase‐encoding. In the second set of experiments, maps were acquired using a GRE‐based BSS pulse sequence with navigator echoes. For this set, the measurements were consecutively repeated 16 times for the same imaging slice. These measurements were averaged to obtain the reference map. Due to the periodicity of respiration‐related phase shifts, their effect on the reference map was assumed to be negligible through averaging. The individual maps of the 16 repetitions were calculated with and without using the proposed navigator scheme. These maps were compared with the reference map. The peak‐to‐peak value of respiration‐related phase shifts varied between subjects. Without navigator correction, the interquartile range of percentage error in varied between 4.0% and 8.3% among subjects. When the proposed navigator scheme was used, these numbers were reduced to 2.5% and 2.9%, indicating an improvement in the precision of GRE‐based BSS mapping at high magnetic fields.     

High‐resolution in vivo MR‐STAT using a matrix‐free and parallelized reconstruction algorithm

MR‐STAT is a recently proposed framework that allows the reconstruction of multiple quantitative parameter maps from a single short scan by performing spatial localisation and parameter estimation on the time‐domain data simultaneously, without relying on the fast Fourier transform (FFT). To do this at high resolution, specialized algorithms are required to solve the underlying large‐scale nonlinear optimisation problem. We propose a matrix‐free and parallelized inexact Gauss–Newton based reconstruction algorithm for this purpose. The proposed algorithm is implemented on a high‐performance computing cluster and is demonstrated to be able to generate high‐resolution (1 mm 1 mm in‐plane resolution) quantitative parameter maps in simulation, phantom, and in vivo brain experiments. Reconstructed and values for the gel phantoms are in agreement with results from gold standard measurements and, for the in vivo experiments, the quantitative values show good agreement with literature values. In all experiments, short pulse sequences with robust Cartesian sampling are used, for which MR fingerprinting reconstructions are shown to fail.     

Mapping the human brainstem: Brain nuclei and fiber tracts at 3 T and 7 T

Structural high‐resolution imaging of the brainstem can be of high importance in clinical practice. However, ultra‐high field magnetic resonance imaging (MRI) is still restricted in use due to limited availability. Therefore, quantitative MRI techniques (quantitative susceptibility mapping [QSM], relaxation measurements [ , R₁ ], diffusion tensor imaging [DTI]) and T₂ ‐ and proton density (PD)‐weighted imaging in the human brainstem at 3 T and 7 T are compared. Five healthy volunteers (mean age: 21.5 ± 1.9 years) were measured at 3 T and 7 T using multi‐echo gradient echo sequences for susceptibility mapping and relaxometry, magnetization‐prepared 2 rapid acquisition gradient echo sequences for R₁ relaxometry, turbo‐spin echo sequences for PD‐ and T₂ ‐weighted imaging and readout‐segmented echo planar sequences for DTI. Susceptibility maps were computed using Laplacian‐based phase unwrapping, V‐SHARP for background field removal and the streaking artifact reduction for QSM algorithm for dipole inversion. Contrast‐to‐noise ratios (CNRs) were determined at 3 T and 7 T in ten volumes of interest (VOIs). Data acquired at 7 T showed higher CNR. However, in four VOIs, lower CNR was observed for at 7 T. QSM was shown to be the contrast with which the highest number of structures could be identified. The depiction of very fine tracts such as the medial longitudinal fasciculus throughout the brainstem was only possible in susceptibility maps acquired at 7 T. DTI effectively showed the main tracts (crus cerebri, transverse pontine fibers, corticospinal tract, middle and superior cerebellar peduncle, pontocerebellar tract, and pyramid) at both field strengths. Assessing the brainstem with quantitative MRI methods such as QSM, , as well as PD‐ and T₂ ‐weighted imaging with great detail, is also possible at 3 T, especially when using susceptibility mapping calculated from a gradient echo sequence with a wide range of echo times from 10.5 to 52.5 ms. However, tracing smallest structures strongly benefits from imaging at ultra‐high field.     

Cross‐relaxation parameters in cortical bone assessed with different MR sequences

This study aimed to show evidence of MR cross‐relaxation effects in cortical bone and to compare different MR sequences for the quantification of cross‐relaxation parameters. Measurements were performed on bovine diaphysis samples with spectroscopic methods (inversion‐recovery, off‐resonance saturation) and with a variable flip angle (VFA) UTE imaging method on a 4.7 T laboratory‐assembled scanner. Cross‐relaxation parameter assessment was carried out via a two‐pool model simulation with a matrix algebra approach. A proton signal amplitude of 28 Mol/L was observed (equivalent water fraction of 25%). It was attributed to collagen‐bound water, with values of ~ 0.3 ms, a “long‐T₂” proton pool, in exchange with protons from the collagen macromolecules ( of 10–20 μs). Magnetization transfer (MT) effects were detected with all sequences. The best precision of model parameters was obtained with off‐resonance saturation; the fraction of collagen methylene protons was found in the range of 22–28% and the transverse relaxation time for collagen methylene protons was 11 μs (1% precision). The model parameters obtained were compatible with VFA‐UTE results but could not be assessed with acceptable accuracy and precision using this method. In vivo MT quantification using off‐resonance saturation with a single B₁ amplitude and offset frequency may provide information about the relative amount of collagen per unit volume in cortical bone.     

Exercising calf muscle changes correlate with pH,PCr recovery and maximum oxidative phosphorylation

Skeletal muscle metabolism is impaired in disorders like diabetes mellitus or peripheral vascular disease. The skeletal muscle echo planar imaging (EPI) signal (S_EPI) and its relation to energy metabolism are still debated. Localised 31P MRS and S_EPI data from gastrocnemius medialis of 19 healthy subjects were combined in one scanning session to study direct relationships between phosphocreatine (PCr), pH kinetics and parameters of time courses. Dynamic spectroscopy (semi‐LASER) and EPI were performed immediately before, during and after 5 min of plantar flexions. Data were acquired in a 7 T MR scanner equipped with a custom‐built ergometer and a dedicated ³¹P/¹H radio frequency (RF) coil array. Using a form‐fitted multi‐channel ³¹P/¹H coil array resulted in high signal‐to‐noise ratio (SNR). PCr and pH in the gastrocnemius medialis muscle were quantified from each ³¹P spectrum, acquired every 6 s. During exercise, S_EPI(t) was found to be a linear function of tissue pH(t) (cross‐correlation r = –0.85 ± 0.07). Strong Pearson's correlations were observed between post exercise time‐to‐peak (TTP) of S_EPI and (a) the time constant of PCr recovery τ_{PCr recovery} (r = 0.89, p < 10^{? 6}), (b) maximum oxidative phosphorylation using the linear model, Q_{max, lin} (r = 0.65, p = 0.002), the adenosine‐diphosphate‐driven model, Q_max,ADP (r = 0.73, p = 0.0002) and (c) end exercise pH (r = 0.60, p = 0.005). Based on combined accurately localised 31P MRS and weighted MRI, both with high temporal resolution, strong correlations of the skeletal muscle S_EPI during exercise and tissue pH time courses and of post exercise S_EPI and parameters of energy metabolism were observed. In conclusion, a tight coupling between skeletal muscle metabolic activity and tissue signal weighting, probably induced by osmotically driven water shift, exists and can be measured non‐invasively, using NMR at 7 T. © 2014 The Authors. NMR in Biomedicine published by John Wiley & Sons, Ltd.