Finally, a few authors [54, 49, 20] have borrowed ideas from statistical design for generating efficient sampling trajectories. In , the key point is to fix a set of feasible trajectories (e.g. pieces of spirals) and to select them iteratively by picking the one that brings the largest amount of information at each step. Hence, finding the most meaningful trajectory becomes com- putationally intensive and hardly compatible with a real-time acquisition. The main contribution of [49, 20] is to propose alternative approaches to reduce the computational burden, by working on training images. These adaptive approaches suffer from a few drawbacks. First, the whole versati- lity of MRI scanners is not exploited since fixed trajectories are imposed. Our formalism does not impose such a restriction. Second, even though adaptivity to the sampled image may seem appealing at the first glance, it still seems unclear whether this learning step is really helpful . Finally, these approaches strongly depart from existing sampling theories, whereas our contribution is still motivated by solid and recently established theories.
Reducing scan times in magneticresonanceimaging (MRI) is essential for attaining high spatial resolution, which could aid in diagnosing certain pathologies, such as Alzheimer’s disease. Methods to accelerate the time of segmented MR acquisitions commonly rely on simple sampling patterns such as straight lines, spirals or slight variations of these elemen- tary shapes. However, such geometrical approaches do not take full advantage ofthe degrees of freedom offered by the hardware and cannot be easily adapted to fit an arbitrary sampling distribution. Here, we report the use of a versatile method inspired from stippling techniques that automatically generates optimized sampling patterns compatible with MR hardware constraints on maximum gradient amplitude and slew rate. These non-Cartesian sampling curves are designed to comply with key criteria for optimal sampling: a controlled distri- bution of samples and a locally uniform k-space coverage. Combining sampling efficiency with compressed sensing, the resulting sampling patterns allowed up to 20-fold reductions in MR scan time (compared to fully-sampled Cartesian acquisitions) for two-dimensional
For different setups, prospective acquisitions were performed on an ex vivo baboon brain conserved in a fluorinert solution. All animal studies were conducted in accordance with the European convention for animal care and the NIHs Guide forthe Care and Use of Labora- tory Animals. First, for an isotropic resolution of 0.6 mm, the three proposed SPARKLING trajectories were com- pared: regular SOS, z-vd SOS and fully 3D SPARKLING. As reference, we also performed a standard Cartesian iPAT acquisition with GRAPPA reconstruction available on the scanner (Siemens product sequence) either for iPAT 4 (4x1 and 24 reference lines), iPAT 2 (24 references lines) with Partial Fourier 6/8 (phase and encode). Furthermore, the 3D SPARKLING strategy was com- pared to other 3D trajectories used in compressed sens- ing MRI. First, the 3D Poisson disk strategy introduced by Lustig et al  was considered. This method, which will be referred to as PD-lines, consists in acquiring along the partition direction cross-sections of 2D Poisson disk samples with a deterministic samplingofthe k-space center (see Fig. 4c). The size ofthe deterministically sampled region and the radially decaying rate ofthe density outside this region were selected using a grid- search on retrospectively subsampled reconstructions of a brain phantom image. Second, 3D radial trajectories were also acquired for comparison  (see Fig. 4b). Since  already investigated the performance of 2D SPARKLING against radial and variable-density spiral trajectories, we did not investigate here the stack-of- stars nor the stack-of-spirals. We expect the relative performance of 2D sampling patterns to remain the same when they are stacked into 3D trajectories.
Since the discovery, by chance, of X-ray potentiality by Roentgen in 1896, imaging techniques have been optimized and extended. Depending on the penetration depth, the spatial resolution and the sensitivity required, X-ray computed tomography (CT), Positron Emission Tomography (PET), MagneticResonanceImaging (MRI) or optical imaging, such as Near- Infrared (NIR) fluorescent imaging, may be selected as the most relevant imaging modality . In all cases, enhancement ofthe signal over noise ratio, and thus ofthe sensitivity of these imaging techniques, requires the introduction of contrast agents such as radioisotopes, paramagnetic molecules or more recently inorganic nanoparticles [ 2 ]. The latter have benefited from tremendous advances during the last two decades notably in terms of available synthetic routes. In particular, the improvements ofthe solution-phase approaches allow now an easy and low-cost access (compared to classical physical or vacuum methods) to well- controlled nano-probes. Prominent examples ofimaging modalities which have profited from this intense research activity are MRI and NIR luminescence: a new generationof sophisticated nano-objects has reached such stage of maturity, especially in terms of probe sensitivity, that their use has on one hand, become a standard practice in in vitro biology research and on the other hand, hold promise for in-vivo real-time detection.
When imaging tissues deeply inside the body is required, as microantennae have a limited field of view, they must be implanted . In this case, MRI becomes invasive but offers the possibility of an easy long-term monitoring. Besides flexibility, additional specifications forthe design and the packaging have to be defined to reduce as much as possible the invasive aspect of implantation. The antenna should be very small (i.e., about a few millimeters) and without wire connection. Our MTLR design is totally in agreement with these specifications. Furthermore, packaging the antenna with an adequate biocompatible polymer minimizes its impact on the surrounding environment. Reciprocally, the antenna must be protected from the undesirable effects ofthe biological tissues on its characteristics. Indeed, they have a high permittivity (about 80) and a high conductivity (about 0.7 S/m) thus induce by dielectric coupling a shift-down of f 0 and degradation of Q. This phenomenon has been
Application of this novel technology is illustrated in Figure 2,where the virtual phantom is positioned near the border ofthe field of view and the object of interest is a cylindrical tube filled with water (Tube 2). Both the magnitude (Figure 2a) and the phase (Figure 2b) MR imagesare shown. The k-space image (Figure 2c) indicates a slight shift between the k-space center ofthe object of interest and the k-space center ofthe virtual phantom. This shift has no effect on the magnitude image; on the other hand, it is responsible for a phase-wrapping, as illustrated in Figure 2b. With a fine tuning ofthe µ-ViP k-space transmission, it is possible to eliminate this shift (see the Materials and Methods section).Note that in the current study, MRµI was performed using pulse sequences with a Cartesian samplingofthe k-space. In the case of MRµI with a non-Cartesian k-space sampling (spiral, radial, etc.), it is still possible to construct a µ-ViP as long as the exact k-space trajectory ofthe MR pulse sequence is known.
H MRI. More interestingly, this application can be extended to MRI of nuclei other than 1 H, such as 19 F and 23 Na . As matter of fact, to determine the concentration ofthe total sodium in tissues, physical phantoms are routinely used ; in this case also, virtual phantoms could be used to provide a reference signal, so to replace the physical phantoms. Furthermore it should be also noted that, in analogy to the approach of Hanson et al.  where physiological recording were encoded in real time on the MRI images, other applications of ViP MRI could include the encoding of additional data such as text and graphs, for instance. Finally, as analytical phantoms are increasingly being used to test and validate MR algorithms, such as reconstruction algorithms for instance , the ViP MRI could be also used for generating images of complex analytical phantoms with realistic noise characteristics.
FIG. 4. (a) Schematic view ofthe rat inside the MRI bed, the resonators, and the loop coil. (b) In vivo demonstration ofthe sensitivity and homogeneity improvement of a single loop coil. Axial (a) and (b), coronal (c) and (d), and sagittal (e) and (f) views of a rat head from the same spin-echo acquisition at 7 T without (a), (c), and (e) and with (b), (d), and (f) hybridized rods positioned around the rat. The following settings were used for this turboRARE sequence: 128 128 matrix size, 30 slices with voxel size of 275 lm 235 lm 500 lm, RARE factor ¼ 8, TE/TR ¼ 24 ms/3000 ms, NA ¼ 8. The position ofthe vari- ous 3D regions of interest used to calculate mean signals is drawn as white dashed rectangles. The region of interest used to calculate the standard deviation ofthe noise is the red dashed square.
2001 ). In particular it has been reported to be activated both during motor ex- ecution (ME) and MI (Solodkin et al., 2004 ; Raffin et al., 2012 ; Lotze & Hals- band, 2006 ; Hétu et al., 2013 ; Confalonieri et al., 2012 ; Sharma & Baron, 2013b ; Fleming, Stinear, & Byblow, 2010 ) though greater activation has been observed during MI than ME (Gerardin et al., 2000 ; Hanakawa et al., 2002 ). The SPL is known to play a role in guiding motor activity in relation to spatial informa- tion (Buneo & Andersen, 2006 ; Wang et al., 2015 ; Culham & Kanwisher, 2001 ) and to be crucial in thegenerationof mental motor representations (Sirigu et al., 1996 ). Several studies have demonstrated that impairments to the parietal cortex reduced MI ability (Sirigu et al., 1996 ; Danckert et al., 2002 ; McInnes, Friesen, & Boe, 2016 ). A meta-analysis recently conducted to determine which neurologic disorders/lesions impair or restrict MI ability showed that patients with parietal lobe damage were most impaired (McInnes, Friesen, & Boe, 2016 ). In MI, the SPL is thought to play a role in facilitating the planning and coordi- nation of imagined movements and/or in indirectly inhibiting M1 through its connection with the SMA (McInnes, Friesen, & Boe, 2016 ; Kasess et al., 2008 ; Solodkin et al., 2004 ). Activations in the SPL have been shown to be more active during visual imagery than during kinaesthetic imagery (Guillot et al., 2009 ). However we found no significant activation in the occipital regions as would be expected during visual imagery. Therefore it is unlikely that the SPL activa- tion would indicate that participants in the BI_DIM performed a motor im- agery that would have been more visual than kinesthetic. The superior parietal cortex has also been demonstrated to be active during generalized neurofeed- back when feedback is presented visually (Sitaram, Ros, et al., 2016 ; Emmert, Kopel, Sulzer, et al., 2016 ; Ninaus et al., 2013 ). However the fact that the SPL was more significantly active in the BI_DIM group than in the UNI_DIM group suggest that it is more than a generalized NF effect. This activation could result from both the overlap ofthe motor imagery task and the self-regulation pro- cess (Sitaram, Ros, et al., 2016 ), both of which could be more intense under the bi-dimensional condition.
In this paper, forthe first time, we solve the SPARKLING optimization globally in 3 dimensions. First, in Sec II, we remind the optimization problem to be solved for generating SPARKLING trajectories. Then we focus on major computational bottlenecks that prevented us from scaling the original solution to 3D and provide detail on our main contributions. One key ingredient in SPARKLING lies in setting the right target sampling density. The latter may vary as a function ofthe resolution and acceleration factor. For that pur- pose, we parameterize radially decaying densities by two parameters (cut-off, decay) and find the optimal density through grid search over pairs of parameters. This study can be conducted on retrospective analysis and then the sought optimal density can be used further in prospective acquisitions. In Sec. IV, we present the experimental data set on which the numerical studies are performed later on for validation and comparison purposes. In this regard, we carry out retrospective analysis on NIST phantom data collected at 3 Tesla. Then we run prospective in vivo brain imaging acquisitions on a healthy adult volunteer still at 3 Tesla and compare the proposed full 3D SPARKLING with the existing spherical stack of 2D SPARKLING.
Gradient waveform design for variable density sampling in MagneticResonanceImaging
Nicolas Chauffert, Pierre Weiss, Jonas Kahn and Philippe Ciuciu.
Abstract—Fast coverage of k-space is a major concern to speed up data acquisition in MagneticResonanceImaging (MRI) and limit image distortions due to long echo train durations. The hardware gradient constraints (magnitude, slew rate) must be taken into account to collect a sufficient amount of samples in a minimal amount of time. However, sampling strategies (e.g., Compressed Sensing) and optimal gradient waveform design have been developed separately so far. The major flaw of existing methods is that they do not take thesampling density into account, the latter being central in sampling theory. In particular, methods using optimal control tend to agglutinate samples in high curvature areas. In this paper, we develop an iterative algorithm to project any parameterization of k-space trajectories onto the set of feasible curves that fulfills the gradient constraints. We show that our projection algorithm provides a more efficient alternative than existinf approaches and that it can be a way of reducing acquisition time while maintaining sampling density for piece-wise linear trajectories.
threshold is probably located in the Heschl gyrus). Because it is generally believed that the PT handles more complex computa- tions than the primary auditory cortex (Griffiths and Warren, 2002), it is reasonable to think that the complex process of pho- nological decoding takes place in the PT. Yet, the current state of knowledge does not allow to categorically claim that HG cannot support this process. Ja¨ncke et al. (2002) observed activations that also straddled the PT and HG when comparing unvoiced versus voiced consonants. Numerous studies have revealed in- creases of PT activations with the spectrotemporal complexity of sounds (for review, see Griffiths and Warren, 2002; Scott and Johnsrude, 2003). The present data indicate that PT activations do not simply depend on the acoustic complexity of speech sounds but also reflect processes tuned to the phonology ofthe native language. This result adds to the converging evidence in favor ofthe involvement ofthe PT in phonological processing. First, lesions in this region can provoke word deafness, the inabil- ity to process speech sounds with hearing acuity within normal limits (Metz-Lutz and Dahl, 1984; Otsuki et al., 1998), and sylla- ble discrimination can be disrupted by electrical interference in the left STG (Boatman et al., 1995). Second, activity in the PT has been observed in lip-reading versus watching meaningless facial movements (Calvert et al., 1997) when profoundly deaf signers process meaningless parts of signs corresponding to syllabic units (Petitto et al., 2000) and when reading (Nakada et al., 2001). Finally, PT activations have also been reported in speech produc- tion (Paus et al., 1996). These data are consistent with the notion that the PT subserves the computation of an amodal, abstract, phonological representation.
4 In hardwood species only vessels are considered to have sap conduction function. For poplar, tyloses that could block the flow are rarely presented, and its vessel elements are connected to each other by simple perforation plates (Wheeler et al. 2007). By this means the vessels in poplar are practically considered as long capillary tubes in parallel (a priori longer than the sample size, i.e. 10 cm) (see Figure 1). In addition poplar contains fiber parallel to vessels and of typical length between 900-1600 µm (data collected from InsideWood database; Wheeler 2011) (see Figure 1). There appear to be a few pits dispersed along the fibers (see Figure 1.C). The existence of some pits along the vessels cannot be excluded, but they were not observed in the RL and TL planes. Finally, there is a porous structure made of long and large channels through which the liquid can a priori rapidly progress, and a second porous network situated around these channels where the liquid may penetrate more slowly, either via pits or by diffusion through cell walls. The small size and numbers ofthe vessel-to-fiber pits substantially reduce the effectiveness of free liquid flow (Bonner and Thomas 1972). In this study, the secondary penetration will be a subject of discussion. Anyway, one is dealing with a porous medium, i.e. a solid structure with voids, for which it is natural to expect a Washburn imbibition process (Washburn 1921; Gruener et al. 2012), namely a penetration driven by capillary (Laplace) pressure due to menisci formed by the air-liquid interface inside the porous medium, and a viscous resistance due to liquid flow through the lumens.
Developing statistical methods based on detailed modeling ofthe fMRI process opens the door to using more direct, informative inference paradigms based on estimated effect sizes, confidence intervals, and Bayesian posterior assessments rather than more indirect approaches based on significance tests and P values. Link- ing statistical methodology development and fundamental fMRI research is crucial for developing more accurate analysis methods, attributing accurate scientific interpretations to results, and ensur- ing the reliability and reproducibility of fMRI studies. These points have been made before. However, their significance has perhaps not been considered to the extent required.
gradient descent can be used to attain a minimum ofthe error function. With recent successes in real world applications with important economic interest, development in the field of artificial neural nets has dramatically in- creased and many variants on architecture and refinements ofthe learning algorithm have been proposed. Many changes in architecture boil down to constraining the linear transformations at each layer to adhere to a certain form. In [LeCun, 1985] the first network trained by gradient descent using convolutions as the linear operations was introduced to classify hand-written digits. Imposing that the linear transformations be convolutions is a transla- tion of mathematical assumptions into the architecture: Convolutions imply spatial covariation ofthe output with respect to the input and carry the mes- sage that most translated versions of images are still images and to be treated in a similar way. A further, practical aspect is the typically very restricted size ofthe filter footprint, forcing the linear transformation to extract local- ized filter responses, while preserving spatial organization ofthe signal. For natural images, this typically results in the learning of edge, texture boundary and blob detectors. Having several layers of convolution operations followed by pointwise nonlinearities or localized nonlinearities like maximum pooling leads to a parallel treatment of all image patches by the same operations. This reflects biological processing in the sense that visual neurons perform mostly local computations and are organized in a retinotopic manner.
Prvotní zobrazení srdce magnetickou re- zonancí (cardiac magneticresonance ima- ging – CMR) trpělo špatnou kvalitou obra- zu, tedy nedosahovalo požadované úrovně kvality. K velkému zlepšení došlo zavedením segmentované akvizice dat k-prostoru spolu se synchronizací akvizice s EKG pacienta, ať už při zadrženém dechu, nebo při volném dýchání s dechovým navigátorem. V sou- časné době se CMR používá nejčastěji pro hodnocení srdeční funkce, charakterizaci tkání, hodnocení zánětlivých nebo post-in- farktových změn a také pro hodnocení střá- davých onemocnění, kde své uplatnění na- chází pozdně postkontrastní skeny (tzv. late gadolinium enhancement – LGE). Své uplat- nění však CMR nachází také při hodnoce- ní chlopní, kdy je možné provádět i měření průtoku danou oblastí, nebo u hodnocení srdečních dysynchronií. Mimo získaná one- mocnění se využívá i u pacientů s vrozenými vadami srdce, kde umožňuje objektivně sle- dovat dlouhodobý vývoj nemoci.
Water in wood material plays a major role with regard to its mechanical or physical properties in various situations, in particular in outdoor applications, leading to a possible deterioration in the performance ofthe wood elements (eg. structural integrity, thermal insulation). Therefore, it is crucial to understand the water uptake and release as well how the water transports in the wood material to extend the lifetime ofthe timber buildings. The mechanisms of liquid transport are related to the hierarchical and multi-scale structure of wood and to the presence of free and bound water. Despite numerous studies, the water migration process in wood material is not well understood.
structured certification in Europe. 1 , 3 – 5 It provides a core knowledge
summary for cardiology trainees, cardiologists, and professionals with an interest in CMR. The syllabus can serve as a guide for devel- oping educational material and educational course content. Trainees may use the CMR syllabus to guide their preparation for knowledge- based assessments (in particular the European CMR Exam). Levels 2 and 3 CMR-certified practitioners might identify topics in which they would benefit from Continuous Medical Education (CME). Harmo- nized imaging education and assessment facilitate the delivery of stan- dardized high-quality cardiac imaging, which in turn leads to improved diagnosis and management of cardiovascular disease across Europe.
Abstract—MagneticResonanceImaging (MRI) is an imaging technique exploiting themagneticresonanceof specific nuclear spins, like protons. In this paper, MRI probes based on dielectric ring resonators are investigated from a theoretical point of view. We take advantage ofthe high-permittivity and low- losses properties ofthe ceramic material used for manufacturing these probes for microscopy applications. MagneticResonance Microscopy (MRM) aims at imaging tiny samples with a sufficient resolution to distinguish small details. In this framework, com- pact resonators, called volume probes, contain the investigated sample and are used for both signal transmission and reception. The new developed semi-analytical model enables estimation ofthe frequency ofthe first transverse electric mode of a cylindrical resonator. It also provides a method to compute the corresponding magnetic field distribution, the dielectric losses contributions from the probe and the sample, and Signal-to-Noise Ratio (SNR). The proposed approach aims at providing design guidelines for dielectric probes.