Knowledge-based acceleration of MRI for metal object localization

In the last two decades Magnetic Resonance Imaging (MRI) has gained an important position in radiology mainly due to its unsurpassed performance in anatomical, functional, and physiological imaging of soft tissues in the human body. The favorable diagnostic properties of MRI have caused a steady rise in the number of MRI exams performed each year. Because of this growth, there is a need for efficiency in MRI exams, by accelerating the imaging process and/or by increasing the utility of the MRI scans.

In this work we address both acceleration of MRI and increasing utility of MRI by performing accurate localization of metal devices in MR images. We approached this by incorporating prior knowledge in the imaging process, following a basic principle: the more is known about a problem beforehand, the easier and faster it will be to solve the problem.

Acquiring less data than typically required is a common method to accelerate MRI. We found that prior knowledge is useful in selecting the data points that are acquired in an undersampled acquisition. Even when using only a single scan of a certain anatomy as prior knowledge this approach resulted in better MR images.

Simulation of MRI is an important tool in characterizing imaging artifacts resulting from the presence of metal objects in MRI. We show that with some assumptions about the MR signal, simulation of metal artifacts can be greatly accelerated. These fast simulations allowed us to characterize the artifacts around a metal device in varying orientations in reasonable time, with a model of the device as prior knowledge. By using these simulations as templates in a template matching method, this allowed us to accurately localize arbitrary metal devices in MRI scans. Automatically localizing metal devices has applications in detecting markers, tracking interventional devices, and possibly for correcting artifacts around larger implants.

We applied this method to the localization of brachytherapy seeds in the prostate for dosimetry in low dose rate brachytherapy. Despite the large number of seeds that are present in a relatively small region, we were able to correctly localize 96% of the seeds in 25 patients.

Finally, we combined the principle of undersampled MRI with metal device localization to achieve real-time tracking of interventional devices. By providing anatomical images simultaneous with the device localization, the device can be visualized in a real-time anatomical reference image. We demonstrated this method for tracking of multiple metal spheres and an MRI compatible biopsy needle, which showed the method accurately localized the devices, even when moving rapidly.

Together, these results show that the use of prior knowledge can help accelerate MRI and provide clinically relevant information about the location of metal devices in MRI.