Application en neuro-imagerie de techniques IRM en encodage spiralé de l'espace de Fourier

par Hardik Doshi

Projet de thèse en PCN - Sciences cognitives, psychologie et neurocognition

Sous la direction de Laurent Lamalle.

Thèses en préparation à l'Université Grenoble Alpes , dans le cadre de École doctorale ingénierie pour la santé, la cognition, l'environnement , en partenariat avec IRMaGE (laboratoire) depuis le 06-10-2014 .

  • Titre traduit

    Neuro-imaging applications of spiral trajectory based k-space encoding MRI techniques.


  • Résumé

    We have explored potential improvements of advanced neuro-imaging applications of Magnetic Resonance Imaging (MRI) to measure blood perfusion and / or detect neuronal activation, associated with the use of a specific approach to encode the spatial origin of signals. MRI is a well-established and commercially available technique. It is non-invasive and gives access to a wide palette of contrasts by using specially developed acquisition sequences and appropriate reconstruction algorithms, which have proven never ending and very active fields of research and development. For measuring blood perfusion, we have considered Arterial Spin Labelling (ASL) sequences, which rely on measuring subtle (~4%) signal difference between magnetically labelled blood image and control image. For detecting neuronal activation in functional MRI (fMRI) experiments, Blood Oxygenation Level Dependent (BOLD) contrast is used, which arises from small local variations in blood oxygenation of cortical microvasculature resulting from neural activation. The signal changes are also very small (~4%). A further application of interest originally considered was Diffusion Weighted Imaging (DWI) for tractography applications. In DWI the signal losses associated which Brownian motion of water molecules in a systematically varied diffusion-weighting gradient have to be measured to reveal macroscopic effects of brain tissue microstructure. All these sequences suffer from low signal to noise ratio (SNR) and impose to scan over relatively long times series, during which patient motion can lead to severe difficulties. In MRI, the signal is usually sampled in the presence of a spatial encoding magnetic field gradient. When spatial encoding gradients are homogeneous, implying a linear relation between spatial location and local magnetic field, the relationship between signal and image is of Fourier transform. Signal is sampled in spatial frequency domain or k-space. The k-space coordinates are directly related to the time integral of the magnetic field gradient across the object being imaged. By changing the gradient over time, the k-space data are sampled in a trajectory through k-space. Generation of trajectory k(t) is dependent on applied gradient strength and slew rate. Their maximum values impose time constraints on the trajectory and thus a trajectory-dependent minimum time to reach k-space center. The effective contrast of an image is mainly determined by the effects of physical parameters influencing the signal at the time of sampling k-space center. The intrinsic signal variations during the spatial-encoding trajectory should be small to reduce widening the Point Spread Function (PSF). Typically, the effective transverse relaxation time T2* determines the time scale of signal duration after excitation. If T2* is not the source of the targeted contrast, k-space sampling should start as soon as possible to avoid unwanted signal decay. Echo-planar Imaging (EPI) is a fast spatial encoding technique commonly used for the ASL, BOLD fMRI and DWI sequences. Traditional EPI sampling follows cartesian trajectory, starts from corner of the k-space and samples it line by line. From samples acquired in cartesian trajectory images can be reconstructed using Fast Fourier Transformation (FFT) with relative ease. With EPI trajectory, time to reach k-space centre is half of the complete trajectory duration. In case of ASL or DWI acquisitions, no T2* signal weighting is desired and T2* signal decay during first half of trajectory reduces SNR. Spiral-EPI corresponds to variants of EPI where the readout trajectory are spiral-shaped. It is worth noting that the k-space coordinates sampled along the trajectory are no longer uniformly distributed on a cartesian grid and straightforward FFT reconstruction is not possible anymore. A “spiral-out” trajectory starts sampling at k-space centre and proceeds to reach radially remote coordinates. A “spiral-in” trajectory starts at k-space periphery and progressively gets closer to and ends at k-space center. Various combinations (“in-out”, “out-in”, “out-out”, etc) are possible, with different properties. Using spiral-out trajectory, ASL and DWI can benefit from very short delay and reduced SNR losses of T2* origin at the time of sampling k-space center. In ASL, using a spiral-out-out trajectory, two signals can be acquired after a single excitation: a first one with very small T2* weighting, and a second, T2*-weighted one, sensitive to BOLD contrast. The simultaneous excitation of both collected signals avoids physiologic variations between them, an inherent property desirable for the prospect of quantitative signal analysis. Amongst other interesting properties of spiral vs cartesian EPI encoding are a higher intrinsic robustness to motion during acquisition readout (due to periodic nulling of gradient waveforms first moment) and a more symmetric use of spatial encoding gradient hardware. To reduce PSF degradation, the readout duration should be short with respect to the time scale set by the shortest T2* affecting the signal of interest. For large image matrix sizes, the readout can be shortened by distributing the acquisition over multiple shots. This approach however introduces an added sensitivity to motion or physiologic variations between shots, which we preferred to avoid. Single shot EPI or spiral readouts were thus considered, but they can be long and lead to increased sensitivity not only to T2* decay during encoding but also to global and local off-resonance effects, associated with B0 field inhomogeneities. As a result, the Fourier transform relationship between image and frequency domains gets compromised and specific reconstruction algorithms are necessary. Another approach to shorten readout duration is to apply parallel acquisition techniques. The latter exploit the spatial sensitivities of multiple coil elements to provide some degree of signal localization, allowing to apply the Nyquist sampling constraint to effectively reduced image field-of-views. This translates to a sparser k-space trajectory, which can be sampled in a shorter time. Parallel acquisition techniques also require specific reconstruction algorithms to combine signals acquired from coil elements. The goal of this study was to develop a perspective motion correcting perfusion imaging sequence using spiral encoding trajectory which could be used in clinical protocol. We implemented spiral k-space encoding trajectories to improve ASL and BOLD sequences. We merged our spiral encoding developments with motion correction functionality (motion correction patch developed by Michael Helle, Philips Healthcare). We further merged these functionalities with a time encoded ASL (te-ASL) functionality (patch provided by M. P. J. Van Osch, Leiden University Medical Center). We also evaluated original, locally developed approaches for: - trajectory calibration, allowing to take into account the effects during readout of residual eddy currents associated with time-varying gradient waveforms, - image reconstruction from non-uniform sampling schemes and parallel acquisitions - B0 mapping - Off-resonance deblurring