By Michael Friebe

For about 45 years, MR has combined strong magnetic fields and mathematical models to provide excellent soft tissue contrast and, these days, even molecular information. But these systems have retained certain undesirable traits: they are massively heavy, require special environments, take a long time to perform scans and are hampered by high investment and running costs.

There are at least 25,000 units very unevenly installed in the world (depending on the country of installation between 1 MR for every 10,000 to 1 MR for every 10 million population) producing on the order of 100 million patient exams per year globally.

Research over the last few years has focused on increasing resolution, signal to noise ratio, and speeding up the examination time. Experts sought additional applications, faster contrast scans of the lungs, simplification of cardiac imaging workflows, and to address specific needs of changing demographics (aging population and multimorbid patients).

Economic pressure, the need for personalized treatment, and the increased imaging requirements of precision medicine are putting additional pressure on the future performance and features of MR systems. A lot of these issues are directly related to the MR systems magnet/magnetic field and therefore, one area of research in recent years was to implement higher and higher field strength magnets. This, however, comes with even stricter siting requirements and higher costs, as the entire environment has to be adapted to the constraints of the MR.

These days, innovation is being driven by new acquisition and dedicated imaging software (e.g., compressed sensing for reduced overall scan time, while maintaining resolution and image quality), and post-processing programs for better, faster cancer diagnosis using automation (which also has the benefit of increased reproducibility and high quality), artificial intelligence, deep learning, and big data.

Other developments have focused on adding MR to other modalities – combinations with linear accelerators were one of the RSNA 2017 highlights.

It is foreseeable that this will lead to a very streamlined and possibly almost completely automated acquisition and interpretation of the acquired MR images in the next 5-10 years. But are we ready yet to accept a diagnostic report that comes from a program rather than from a human? While there is scientific evidence that narrow artificial intelligence-based programs are already superior to the human created reports there are still many philosophical, moral, and ethical questions. For example the dual-use (good and evil) nature of research and development needs to be discussed and dealt with if we completely rely on programs in the future.

But will MR stay big, expensive, and limited to large shielded radiology-based suites? At least I hope that we will have many dedicated and small – maybe even portable – MR systems available in the next decade. This requires that we go back to lower field strength magnets, which means a reduction in sensitivities but also higher contrast for certain applications. These low-field magnets (on the order of several hundred mT) would not require superconductive coils and huge amounts of energy. They would be limited in their applications, but come with dramatically lower investment, siting, and running cost.

My prediction would be that we have, in 2028, around one to five percent very-high-field MR systems that have a limited benefit for few applications, lots of very dedicated small and inexpensive units with a limited scope (intervention, surgery guidance, emergency, intensive care, dedicated applications, pediatric, etc.), combined systems for diagnosis and therapy, and fully automated standard care units (1.5T/3T) that will be smaller and faster, using standardized and reproducible advanced automated software programs.

About the author: Michael Friebe is a research professor of image-guided therapies at the Otto von Guericke University in Magdeburg Germany (www.ovgu.de/friebe and www.inka-md.de)