Low frequency MRI

Andrew Webb, Thomas O’Reilly, Kirsten Koolstra, Wouter Teeuwisse
Funding: NWO Simon Stevin Prijs, NWO WOTRO 

Magnetic resonance imaging (MRI) is one of the most common clinical imaging techniques to diagnose disease and assess the success of follow-up surgery or therapy. It has the advantages of being non-invasive, fully three-dimensional, offers a range of image contrasts, and does not involve ionizing radiation which is critical when considering pediatric applications. Conventional MRI systems consist of superconducting magnets, high power gradient and radiofrequency amplifiers, and proprietary software. Typical costs are 1 million euros per Tesla, meaning that clinical systems cost between 1 and 3 million euros, with annual service contracts of 200 000 euros, and high expertise required for operation and repair. The overall aim of this project is to design an MRI system which is ten times less expensive than existing technology, while maintaining the ability to produce diagnostically useful images.

In a conventional MRI system the vast majority of the costs are associated with a high level of sophistication of the hardware components of an MRI system, with relatively simple software algorithms being used for image reconstruction. Our research approach is to completely reverse this paradigm, in other words to produce a low field MRI system from inexpensive hardware components, and by exactly characterizing the system to shift the focus onto more sophisticated image reconstruction algorithms to use the relatively inexpensive yet powerful capabilities of personal computers. 

Our envisioned system will consist of a small Halbach permanent magnet able to image the brain of young children or the extremities of adults. Multiple-frequency, multi-channel RF transmit and receive systems will be used to encode spatial information from the subject. The transmit-coils will be powered by multiple switched-mode low-cost RF amplifiers. Low-cost low current gradient amplifiers and gradient coils will also be constructed to increase the spatial resolution. Image processing algorithms based on machine learning and training data sets will be used to overcome the high degree of spatial non-linearities associated with the measurement system.