The Taub Faculty of Computer Science Events and Talks

Magnetic Resonance Imaging has revolutionized diagnostic medicine. It is an excellent tool for disease diagnosis and monitoring, offering superb soft tissue contrast and high anatomic resolution; unlike computed tomography (CT), it lacks of ionizing radiation. However MRI suffers from several shortcomings, one of which is the inherently slow data acquisition. This has limited the penetration of MRI to applications that require sharp images of fast moving small body parts, such as angiography, cardiovascular imaging, imaging small children and fetal imaging.

To overcome this limitation, many methods for fast imaging by reduced sampling have been proposed. These are based on exploiting various redundancies in the data. Parallel Imaging is the most noteworthy. It is a well-established accelerated imaging technique based on the spatial sensitivity of array receivers. Compressed sensing is an emerging accelerated imaging based on the compressibility of medical images. Synergistic combination of parallel imaging and compressed sensing offers much higher acceleration and better quality imagery.

For the last two years, we have been experimenting with applying compressed sensing parallel imaging for MR body imaging of pediatric patients. It is a joint-effort by teams from UC Berkeley, Stanford University and GE Healthcare. The talk aims to summarize our experience so far. I will start with some background on MR imaging, parallel imaging and compressed sensing MRI. I will then turn to describe our unique approach of data acquisition and reconstruction, our implementation on parallel processors (multi-core and GPGPU), applications and clinical studies. Our approach is implemented and installed at Lucile Packard Children's Hospital at Stanford. So far, it is the first and only place in the world in which compressed sensing is routinely used in clinical practice. We can routinely achieve clinical quality reconstructions of body pediatric MR, at more than 8-fold acceleration. These are reconstructed and displayed in about a minute using our GPU-based parallel processing reconstruction.

Time permitting I will also describe some recent advances in parallel imaging that leads to interesting low-rank structured matrix completion problems.