Kowalik, GT; (2016) Rapid online reconstruction of non-Cartesian magnetic resonance images using commodity graphics cards. Doctoral thesis , UCL (University College London).

Preview

Text Kowalik_GKowalik edited version.pdf.no signature.pdf Download (5MB) | Preview

Abstract

In Magnetic Resonance Imaging, energy of electromagnetic waves is used to excite protons placed in a static magnetic field. This generates a signal, which is further spatially encoded with linear magnetic field gradients. The signal exists in frequency domain called k-space. Traditionally, the signal is sampled in lines stored on a Cartesian grid. Next, Fast Fourier Transform is applied to generate images. However, the consecutive manner (line-by-line) of this strategy makes it very slow. Faster sampling strategies exist, but acquisitions with them require a more complex image reconstruction process. There is an obvious trade-off between acquisition time and complexity of image reconstruction. Real-time assessment protocols for day-to-day clinical work demand both data acquisition with rapid sampling trajectories and fast, robust image reconstructions. Computational solutions in form of parallel architectures can be used to aid image reconstruction, which has been proven to significantly speed-up reconstruction process. Regrettably, this is often done in off-line mode, where the data need to be downloaded from the scanner and reconstructed elsewhere. This process hinders the clinical workflow substantially. This work describes challenges entailed with translation of advanced imaging protocols into the clinical environment; (i) use of the advanced sequences is limited by their reconstruction time, and (ii) fast implementations exist but they still run in off-line mode. These were addressed and resolved with development of a novel online, heterogeneous image reconstruction system for Magnetic Resonance Imaging. The external platform was designed to support fast implementation of advanced reconstruction algorithms. An external computer equipped with a Graphic Processing Unit card was integrated into the scanner’s image reconstruction pipeline. This allowed direct access to high performance parallel hardware on which the rapid data reconstruction can be realised. Also, the automation of data transmission and reconstruction execution has preserved the non-interrupted assessment workflow.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as CSVJSONXML

Archive Staff Only

View Item