To gain additional insights in the origin of the T 2 ∗ –weighted signal in ex- ercising muscle we consecutively performed dynamic measurements of T 2 ∗ – weighted BOLD signal and localized 31 P–spectroscopy in Section 2.3 . Corre- lation between the T 2 ∗ changes and high–energy phosphate kinetics revealed that during the exercise and early recovery the T 2 ∗ changes are caused mostly by changes of intracellular pH. The shape of EPI signal time courses and pH was very speciﬁc for each subject, even though both measurements were per- formed consecutively and not interleaved. Furthermore, it was found that the TTP of the BOLD signal was positive and highly signiﬁcant correlated with the time constant of PCr recovery (τ PCr recovery ) and the maximum ox- idative phosphorylation (Q max ). The discovery of these ﬁndings was possible due to the high SNR of 31 P–spectra at 7 T, which in turn enabled the quan- tiﬁcation of pH from single spectra without additional temporal averaging. Further SNR increase resulted from the custom–built, form–ﬁtted coil array. Also the use of localized MRS (semi–LASER sequence) ensured that the 31 P– spectra originated from the exactly the same muscle as the 1 H T ∗
Methods like spectrally selective 31 P MRI or chemical shift imaging (CSI) are rapidly advancing. With these methods, several muscles can be measured simultaneously and potential metabolic heterogeneities within a muscle can be detected, however, within their limits of point-spread function, spatial resolution, SNR and temporal resolution. Parasoglou et al.  showed the feasibility of 3D 31 P MRI to quan- tify PCr dynamics in the human calf. One serious drawback of this method is that localized pH is not accessible by PCr imaging, which is an important parameter for characterization of aerobic and glycolytic muscle metabolism. CSI, on the other hand, suffers from poor voxel definition and requires far longer acquisition times. While metabolic processes on the order of 20 – 30 seconds have been observed us- ing a gated CSI protocol , this approach depends on reproducibly reaching a steady-state (via a known dependence on time, e.g., mono-exponential) and imposes restrictions, e.g., acidosis should not be induced. The semi-LASER single-voxel MRS acquisition scheme is applicable to any exercise intensity and metabolic state, also any orientation and size of an individual muscle can be covered by the free positioning of the double oblique voxel in three dimensions, provided sufficient B 1 .
5 pH–Induced Chemical Shift Differ-
An accurate in–situ determination of pH in closed or inaccessible systems is challen- ging, but could facilitate a better characterization of biological, chemical, electrochem- ical, and biochemical reactions. Such inaccessible systems are for example the pH in catalytic or hydrothermal reactions [529,530], pH of confined liquids in porous media or reverse micelles [531–534], pH at electrochemical solid–liquid interfaces [49,50,535], and in vivo extracellular pH . A multitude of methods and techniques has been de- veloped to measure pH in those more challenging cases, encompassing positron emission tomography [537–539], fluorescence imaging microscopy , and absorption and emis- sion spectroscopy [541–543]. These methods rely on pH–sensitive, spectroscopically active probe molecules (tracer, fluorescence dye). NMR and MRI are among those techniques for non–invasive pH determination and are very powerful methods, since they allow for multi–nuclear measurements (e.g. 1 H [529,531,544] 14 N  19 F , 51 V [532,544] ,...).  Specific spectroscopic parameters, such as chemical shift, strongly depend on the local electronic structure and hence on the pH–dependent (de–)protonation of the probed molecule.  Often, NMR experiments can even provide additional information about mobility based on relaxation and line width  or about intermolecular distances exploiting the NOE .
The Sequence Response Kernel approach is based on the calculations of the k- space simulation, hence it needs to be prepared for an extended set of parameters to reflect the dynamic object properties. But since polynomial motion models cause a change of phase in signal space that is independent of the observed tissue, it can be expressed as an additional motion-dependent factor in signal space. It thereby becomes an extra convolution kernel in image space with behavior that can be investigated and applied independent of the tissue, geometry, and shift kernels. Similarly, diffusion and further advanced effects may be introduced with ease to the Sequence Response Kernel approach if it can be expressed as a multiplication in signal space. However, if the motion model affects the pulse effects, such as flow perpendicular to slice direction in the presence of slice selective pulses, then the tissue kernels still need to be calculated for the respective combinations of pulse effects.
Nine of the 14 pigs were investigated using MRI methods to estimate fractional blood vol- ume. To create relaxation-rate-change-time curves, baseline magnetization and relaxation rates were determined before each dynamic acquisition using a T1w 3D gradient echo sequence (TWIST, Siemens Healthcare, Erlangen, Germany) with different flip angles ( α = 5˚, 10˚, 20˚, and 30˚). The sequence parameters were: TR = 2.69 ms, TE = 0.86 ms, voxel size: 2.9 × 2.9 × 4.5mm 3 , 160 × 128 × 48 reconstruction matrix, frequency encoding in the axial direction, parallel imaging (GRAPPA) in 3D with 32 central k-space lines and an acceleration factor of 6. The central k-space region was 100%, i.e., k-space sharing was not used for the baseline acquisitions. To improve reproducibility of signal recordings, automatic sequence adjustments were turned off.
To demonstrate its practical use, the new method has been implemented in the 1000BRAINS population-based cohort study. Drawn from those data, two applications are presented. First, the dependence of the (semi-)quantitative parameters on an elevated body mass index was investigated. It could be shown that white and grey matter struc- tures are affected by adiposity, mainly including changes of the longitudinal relaxation time in the frontal, temporal and parietal grey matter, as well as the thalami, bilateral corona radiata, corpus callosum and left thalamic radiation. The lack of changes in the magnetisation transfer ratio suggests a developing low-grade inflammation with obesity. Second, the affect of white matter hyperintensities (WMH) on the (semi-)quantitative parameters could be demonstrated. WMH can be divided into deep and periventricular WMH, which differ in their pathological cause. The results indicate that myelin loss and gliosis due to a possible breakdown of the blood brain barrier and inflammation are the major pathological substrates turning white matter into deep WMH. In the case of periventricular WMH an altered fluid dynamic and CSF leakage add on top. A signif- icant change with severity could be observed especially in the free water content and bound proton fraction of the normal appearing white matter and deep WMH. These findings suggests that the pathology progresses faster in the deep WMH than in the periventricular WMH.
and 13.2 seconds of acquisition time per volume leading to 3.45 minutes for 17 measurements. A second set of high-spatial resolution T1-weighted imaging (repeated 3D- FLASH) was acquired after these 17 low-spatial VIBE res- olution images, as the peak enhancement of the lesion could be expected at the end of this time span ( and references therein). Finally, high-temporal resolution (repeated VIBE with 25 measurements, leading to an acquisition time of 5 minutes 35 seconds, and repeated 3D-FLASH for dy- namic assessment of lesion wash-out) was performed, and then high-spatial resolution T1-weighted images were recorded. The contrast agent used was Gd-DOTA (generic name: gadoterate meglumine; Dotarem, Guerbet, France), injected intravenously as a bolus (0.1 mmol per kilogram body weight) and administered with a power injector (Spectris Solaris EP; Medrad, Pittsburgh, PA) at 4 mL/s followed by a 20 mL saline ﬂush. The contrast agent was injected 75 seconds after starting the ﬁrst coronal T1- weighted VIBE.
As discussed in this work, the proposed modifications are in fact able to boost SNR signif- icantly while maintaining good resolution properties. However, they also make the sequence more prone to dynamic phase changes during signal acquisitoins (motion sensitisation, eddy currents, etc.). This renders the adapted spoiling scheme, which introduces a mixing between spin echo and stimulated echo pathways compared to the original, RF-spoiled sequence version, inappropriate for diffusion weighting applications. The novel signal shaping capabilities (more efficient flip angle algorithm and target shapes with a good trade-off between SNR increase and point spread function broadening) on the other hand are hypothesised to find their future ap- plications in diffusion imaging, if pathway mixing is again avoided. To be successfully applied at UHF, however, more effort has first to be put into current research topics such as flip angle homogenisation and reduction of RF power deposition in order to counteract UHF specific ex- citation field inhomogeneities and tissue heating. Novel parallel transmission techniques are promising developments with this regard. To the current state, however, DW Ss-STEAM is not an alternative to DW spin echo EPI for whole brain diffusion imaging at ultra high fields.
In order to diagnose structural changes consistent with Chiari-like malformation and SM, a standard MRI examination of the brain and spinal cord was performed prior to the perfusion studies. Imaging was performed using a 1 T MRI scanner (Gyroscan Intera, Phillips, Hamburg, Germany) and a two-part surface coil consist- ing of two elliptical elements, which were placed on the right and left sides of the head. Dogs were examined in sternal recumbency with their neck in extension sagittal, dorsal, and transverse images were obtained using T2-Turbospin echo sequences (TE: 120 ms, TR: 2,900 ms). Transverse FLAIR images and dorsal T1-weighted gradient echo images were acquired before and after contrast (i.e., after perfusion study) medium administration to exclude struc- tural brain abnormalities. Field of view was 180 mm × 180 mm, matrix was 288 × 288. Slice thickness varied from 2 to 3 mm. The cervical spine was examined until the first the first thoracic vertebra. Sagittal T2-weighted images were obtained. If the pres- ence of SM was confirmed, transverse gradient-echo images were obtained over the whole extension of the SM.
Dynamic susceptibility contrast magneticresonanceimaging (DSC-MRI) provides good contrast at short imaging times, but lacks commonly accepted robust methods to quantitatively assess cerebral perfusion. For instance, calculation of quantitative cerebral blood flow (CBF), mean transit time (MTT), but also the nowadays often used surrogate parameter time-to- maximum (Tmax), require deconvolution of a voxel’s time- concentration curve (TC) with a manually or automatically selected arterial input function (AIF) ( Carroll et al., 2002; Ostergaard, 2005; Boutelier et al., 2012 ). Selection of an AIF as well as deconvolution are prone to methodological bias potentially leading to significant variation of the results even when evaluating the same DSC-data ( Zaro-Weber et al., 2009, 2012 ). Moreover, optionally introduced TC-model fitting prior to deconvolution was also found to alter the appearance of the finally depicted lesion on the resulting perfusion map ( Christensen et al., 2009; Forkert et al., 2013 ).
Abstract— Building on a long tradition of developing robotic hands, we are developing robotic systems closely copying human hands in its kinematic and dynamic properties. To this end, we require an exact computational model of human hand kinematics in order to obtain optimal grasping properties. From a large number of MRI recordings of hand bones in various grasps, we construct a parametrisable kinematic model, of which optimal versions can be determined. In this paper we present the required image processing and modelling methods as well as a resulting model.
dynamic process and the critical thresholds for viability are dependent on its duration of the ischemia . The originally decreased MVD in the ischemic tissue could elevate again in the subacute stage due to angiogenesis  and the tissue may prevent the fate of infarction. It has been proposed that coupling CBV and CBF for thresholding will raise the accuracy for prediction . In future studies, the combination of Q with conventional perfusion parameters can be assessed for identification of lesion growth. Although the VSI was significantly different between all three regions defined in patients with the baseline measurement, it failed to provide useful information in predicting lesion growth giving a poor area under the curve of 0.59. Besides its heterogeneity in normal brain regions, the VSI is also very sensitive to the presence of large vessels and voxel contamination by CSF. When a voxel is dominated by a large vessel or the CSF, the measured VSI deviates towards much higher values (see Fig. 8.1). This explains the long tails of the VSI distributions in Fig. 8.3B. The shift in the OLI to larger VSI values compared to HEA is very likely due to the vasodilatation of arterioles and venules. In both INF and IGR, the VSI distributions had a wide range. To understand this, we have to keep in mind that the modelling and calculation of VSI is inevitably dependent on CBV. It has been found that CBV decreases significantly in the area of the initial infarct core and lesion growth since the tissue is badly perfused. This drives the decrease of the VSI. On the other hand, both edema and the absence of the CA cause an overestimation of VSI . These two opposite effects result in the broad distribution of the VSI. Moreover, the areas of INF and IGR in our patients were very small compared to other stroke studies [107, 117]. The possible miscoregistration of very small ROIs of INF and IGR could influence the results as well.
Textural features descriptors are widely used in the field of computer vision and medical image analysis [Castellano et al., 2004]. The texture of a region or window is a periodically repeating spatial pattern of gray value variations. By means of textural features, larger regions showing certain periodical patterns can be considered as homogenous. Textural features have been used for many medical problems such as texture-based image segmentation or lesion detection in image data obtained from modalities such as X-ray, ultrasound or MRI [Bankman, 2000]. In the domain of MR imaging, textural features have been used for instance for tissue classification in MR images of brains [Herlidou-Meme et al., 2003, Kmer et al., 1995], classification of lesions in high-resolution DCE-MRI images of breasts [Gibbs and Turnball, 2003] or for detecting simulated microcalcifications in DCE-MRI images [James et al., 2001]. The basic assumption behind textural approaches for tissue classification is that the varying structure of different tissue types is reflected in its visual appearance and exhibits tissue-specific texture patterns. If this assumption holds, the texture of e.g. rectangular windows can be assessed using dedicated mathematical techniques. The resulting vector of textural features is then assigned to the pixel at the centre of the window and provides a description of the pixel’s neighbourhood. By moving the window over the entire image, a feature vector is calculated for each pixel. Thereby, windows displaying the same tissue are likely to yield similar feature vectors.
Radial sampling with golden-ratio-based readout provided nearly optimal k-space coverage for flexible selection of the data and continuous acquisition allows for T1 reconstruction of any cardiac phase. Thus, reconstruction of dynamic T1 maps was possible. The stopping criterion was set to a fixed number of iterations. In our study, this stopping criterion was sufficient because of the convergence behavior of the T1 maps over the iterations, which was similar in all volunteers. However, other stopping criteria could have been chosen. For example, the algorithm could have been stopped if the difference in quantitative T1 map of two subsequent iterations was below a certain threshold as suggested previously [ 76 ]. During iterative T1 reconstruction, data consistency is only ensured by substitution of acquired data. Using substitution, also initial noise in the data will be included again during each iteration. To improve the reconstruction, the reconstruction problem could be reformulated to a least square minimization term. This might allow for additional regularization terms to further improve the achievable parameter estimation [ 90 , 92 ].
In both the institutes, breast MRI was performed according to the guidelines defined by the EUSOMA working group [ 15 ], on 1.5-T magnet (Magentom Avanto, Siemens, Erlan- gen, Germany in both institutes), using vendor-supplied dedicated bilateral breast coils (four channels). Exami- nations were performed with patient in the prone posi- tion, among the 7th and 14th day of the menstrual cycle for pre-menopausal women. The standard protocol used for the clinical evaluation consisted of an axial Short-Tau Inversion Recovery (STIR) T2-weighted sequence and an axial spoiled Gradient-Echo 3D (FLASH) T1-weighted sequence acquired before and five times after the injec- tion of contrast material for the dynamic study (Gado- benate Dimeglumine, Multihance, Bracco; 0.01 mmol/kg of body weight, injected at the rate of 2 ml/s, followed by a flush of 20 ml of saline solution). Technical parameters of the T1-weighted fast low-angle shot sequences were: TR 9 ms, TE 4.76 ms, FOV 340 × 340 mm, slice thickness 2 mm, matrix 512 × 512 at one institution; and 7.4, 4.7 ms, 340 × 340 mm, 1.3 mm, 384 × 369 at the other institution. Starting from 2012, an axial echo-planar imaging (EPI) dif- fusion-weighted sequence was also acquired.
Independent component analysis is a method for blind signal separation formed on the basis of assumed statistical independence of the source signals. The problem of blind source sep- aration or blind signal separation (BSS) appears in many contexts. Blind source separation is a class of explorative tools originally developed for the analysis of images and sound. BSS has received wide attention in various fields such as speech enhancement, geophysical data processing, data mining, wireless communications, image processing, and biomedical signal analysis and processing (EEG, MEG, fMRI). The method is called ’blind’ because it aims to recover source signals from mixtures with unknown coefficients. The most simple situation occurs for two speakers speaking simultaneously. Imagine that the mixture of their voices reaches two microphones, and one wants to separate both sources such that each detector registers only one voice. The problem is called the cocktail party problem which can also be extended to N people standing around and chatting with each other. This mixture of signals is recorded by N microphones. Again, the aim is to extract the voices of the speaker (the sources) from the mixture of speech signals without knowing the sources and the mixture process assuming that the voices are independent of each other. In this project the problem of BSS is applied to the field of functional magneticresonanceimaging (fMRI), especially to fMRI time series, For the fMRI time series it is assumed that the measured signal of neuronal activity are mixed linearly with multiple other signals like noise or movement artifacts, contributing to the measurement. The aim of blind signal sep- aration in fMRI is to detect the intrinsic signals, i.e. the neuronal activity, from the mixed signals measured during the fMRI study. ICA is a statistical approach of transforming multidimensional data into components that are as independent of each other as possible.
CHAPTER 1. INTRODUCTION
The inference of magnetic susceptibility distributions within a measurement volume implies certain requirements and preparatory steps. A model describing the field shifts induced by a susceptibility distribution has to be defined and algorithms for reconstruction, based on this model, have to be implemented. The actual field distribution must be derived from an appropriately configured measurement. This step includes three central topics of phase imaging: acquisition, unwrapping and fieldmap calculation. The acquisition has to be optimised for the best signal quality achievable within the time and resolution constraints [e.g. Deistung et al. , 2008 ]. From measured data only the principal value of the phase is determined (ranging from −π to π), whereas the true phase accumulation has an unconfined angular range. Thus, phase unwrapping – reconstruction of the true phase based on measured phase – is probably the most essential step when preparing the phase data and has been approached in various ways [see e.g. Jenkinson , 2003 ; Abdul-Rahman et al. , 2007 ]. The calculation of fieldmaps based on the phase is strongly intermingled with the two other steps as it depends on the acquisition protocol and the unwrapping strategy has to be chosen according to the characteristics of the data. However, the estimation of fieldmaps that properly represent the actual static magnetic field experienced by the subject does not yet suffice for a convenient reconstruction of susceptibility. The currently employed models relate susceptibility to the field distortion generated as a consequence thereof [e.g. Salomir et al. , 2003 ;
7.4. RECONSTRUCTING SUSCEPTIBILITY DISTRIBUTIONS
7.4 Reconstructing Susceptibility Distributions
All strategies for estimating susceptibility distributions in soft tissue are based on the a priori estimation of fieldmaps, and thus on phase imaging. Single-echo GRE measurements are not useful for this aim, since fieldmaps derived of such data may contain deviations from the true field due to unconsidered phase offsets (see Section 6.1). Fieldmaps used for susceptibility recon- struction should at least rely on double-echo, or offset-corrected phase data. Furthermore it is evident, that the estimation of susceptibility requires knowledge of the magnetic field in all spatial dimensions. Ideally, an isotropic sampling of space should be attempted. This is best achieved using an isotropically sampled, full 3D GRE sequence with slab- or non-selective excitation. Sev- eral strategies for the reconstruction of tissue magnetic susceptibility were introduced during the last years. Two main categories can be defined, single and multiple orientation measurements. Furthermore, reconstruction can be performed directly or by using a minimisation approach. All methods presented below use the model of scalar susceptibility (Section 7.3.1), and will be shown in application examples using optimised parametrisation. Parameter optimisation will be discussed later on in Section 7.6. A comprehensive discussion of the current reconstruction techniques and applications for susceptibility imaging in MRI can be found in Reichenbach [ 2012 ].
To develop an imaging modality for the quantification of brain tissue sodium concentration (TSC) calculated from magneticresonance electrical properties tomography (MR-EPT) based on the correlations between conductivity and sodium concentration in saline solutions. Conductivity maps were reconstructed using the transceive phase of the combined signal at 3T while sodium concentration scans were acquired at 4T both in phantom and in 8 healthy subjects. The brain conductivity and sodium concentration maps were co-registered and normalised to 1mm 152 MNI brain atlas. So-called pseudo tissue sodium maps (pTSC) were generated by performing a linear transform of conductivity maps based on saline solution model at 37°C. Statistical analysis was performed to investigate the discrepancy between pTSC and TSC. A strong linear correlation between pTSC and TSC was found when all brain regions were included (r=0.60, p<0.001). The same trend was found in gray matter (r=0.58, p<0.001) and in white matter (r=0.43, p<0.05), respectively. The slope of the overall linear regression was 0.72, which indicates that pseudo- sodium concentration tends to underestimate TSC values obtained by sodium MRI. Moreover, Bland-Altman analysis revealed that the mean difference between the two methods was ~4mMol/L. The underestimation of pTSC is likely due to the lower water content of brain tissue relative to the saline solution. Despite an overall underestimation compared to sodium MRI, pseudo sodium concentration correlated well with sodium MRI measurement. This provides evidence that sodium ion concentration is the dominant source of electrical conductivity, the latter being accessible with MR-EPT.
tion of a sequence, where the abscissa is the time axis of a sequence diagram. Along the ordinate, multiple axes are arranged: one for each event type of the sequence (RF pulses, gradients, readout, sig- nal). The sequence diagram of a gradient echo sequence is shown in Figure 2.6. After the echo readout as shown in Figure 2.6, residual magnetisation can be removed from the imaging process by applying additional gradients or modifying the phase of the subsequent RF pulses. This process is known as spoiling. GRE sequences are dif- ferentiated depending on the spoiling applied after the readout. The Fast Low Angle Shot (FLASH) sequence  is a spoiled gradient echo, whereas steady state free precession (SSFP) sequences refocus the available magnetisation in the transverse plane. This leads to higher signal–to–noise but may lead to a reduced T 1 contrast .