Advanced acquisition and reconstruction techniques in magnetic resonance imaging
- Montesinos Suárez de la Vega, Paula
- Juan Felipe Pérez-Juste Abascal Director/a
- Manuel Desco Menéndez Director/a
Universidad de defensa: Universidad Carlos III de Madrid
Fecha de defensa: 25 de septiembre de 2015
- Elfar Adalsteinsson Presidente/a
- Juan Miguel Parra Robles Secretario/a
- Pedro Ramos Cabrer Vocal
Tipo: Tesis
Resumen
Magnetic Resonance Imaging (MRI) is a biomedical imaging modality with outstanding features such as excellent soft tissue contrast and very high spatial resolution. Despite its great properties, MRI suffers from some drawbacks, such as low sensitivity and long acquisition times. This thesis focuses on providing solutions for the second MR drawback, through the use of compressed sensing methodologies. Compressed sensing is a novel technique that enables the reduction of acquisition times and can also improve spatiotemporal resolution and image quality. Compressed sensing surpasses the traditional limits of Nyquist sampling theories by enabling the reconstruction of images from an incomplete number of acquired samples, provided that 1) the images to reconstruct have a sparse representation in a certain domain, 2) the undersampling applied is random and 3) specific non-linear reconstruction algorithms are used. Cardiovascular MRI has to overcome many limitations derived from the respiratory and cardiac cycles, and has very strict requirements in terms of spatiotemporal resolution. Hence, any improvement in terms of reducing acquisition times or increasing image quality by means of compressed sensing will be highly beneficial. This thesis aims to investigate the benefits that compressed sensing may provide in two cardiovascular MR applications: The acquisition of small-animal cardiac cine images and the visualization of human coronary atherosclerotic plaques. Cardiac cine in small-animals is a widely used approach to assess cardiovascular function. In this work we proposed a new compressed sensing methodology to reduce acquisition times in self-gated cardiac cine sequences. This methodology was developed as a modification of the Split Bregman reconstruction algorithm to include the minimization of Total Variation across both spatial and temporal dimensions. We simulated compressed sensing acquisitions by retrospectively undersampling complete acquisitions. The accuracy of the results was evaluated with functional measurements in both healthy animals and animals with myocardial infarction. The method reached accelerations rates of 10-14 for healthy animals and acceleration rates of 10 in the case of unhealthy animals. We verified these theoretically-feasible acceleration factors in practice with the implementation of a real compressed sensing acquisition in a 7 T small-animal MR scanner. We demonstrated that acceleration factors around 10 are achievable in practice, close to those obtained in the previous simulations. However, we found some small differences in image quality between simulated and real undersampled compressed sensing reconstructions at high acceleration rates; this might be explained by differences in their sensitivity to motion contamination during acquisition. The second cardiovascular application explored in this thesis is the visualization of atherosclerotic plaques in coronary arteries in humans. Nowadays, in vivo visualization and classification of plaques by MRI is not yet technically feasible. Acceleration techniques such as compressed sensing may greatly contribute to the feasibility of the application in vivo. However, it is advisable to carry out a systematic study of the basic technical requirements for the coronary plaque visualization prior to designing specific acquisition techniques. On simulation studies we assessed spatial resolution, SNR and motion limits required for the proper visualization of coronary plaques and we proposed a new hybrid acquisition scheme that reduces sensitivity to motion. In order to evaluate the benefits that acceleration techniques might provide, we evaluated different parallel imaging algorithms and we also implemented a compressed sensing methodology that incorporates information from the coil sensitivity profile of the phased-array coil used. We found that, with the coil setup analyzed, acceleration benefits were greatly limited by the small size of the FOV of interest. Thus, dedicated phased-arrays need to be designed to enhance the benefits that accelerating techniques may provide on coronary artery plaque imaging in vivo.