Nomenclature for real‐time magnetic resonance imaging

With virtually no exception, the situations described in the literature meet the first criterion. However, there is a large disparity among the cases in meeting the second, and the authors believe this is an important distinction affecting both acquisition and reconstruction methods that needs to be captured in terminology. While a range of works using the RT-MRI label in the past decades have satisfied both the above criteria,(1-6) many do not, resulting in potential for confusion. For example, applications where imaging is used for on-the-fly treatment guidance will call for RT-MRI techniques. However, many studies often label their technique as “RT-MRI,” but lack any indication of reconstruction time or latency for the entire reconstruction pipeline.(7-11) We propose to the community that techniques satisfying the first criterion alone adhere to the established description of “dynamic MRI.” Techniques that satisfy both use the terminology “real-time MRI,” because these are capable of real-time use, not just high frame rate dynamic visualization. Another distinction often used in the literature as an automatic rationale for the use of the RT-MRI label is that of online reconstruction. Image reconstruction can be conducted either offline or online, the latter being reconstructed at the console. An important requirement in our proposed definition of RT-MRI is the ability to visualize the acquired data with minimal latency. Specifically, the term latency is used here to describe the delay between a time-frame of interest represented by a particular data acquisition, and its final reconstruction and visualization. Several offline reconstruction techniques use k-space acquired from the entire dynamic data set (in the k-t domain). The k-t reconstruction algorithms are a perfect demonstration of these (often time consuming) offline reconstruction techniques.(12-15) However, strictly speaking RT-MRI can still be achieved frame-by-frame using offline reconstruction methods, provided the combined acquisition, data transfer, and reconstruction time is fast enough to constitute RT (i.e., low latency). As such these authors believe that the offline/online distinction is an independent concern to that of real-time imaging, and that one should not be implicitly linked to the other. It has become apparent that researchers may claim their reconstruction technique is capable of RT-MRI, although the reconstruction time may take tens of seconds,(16, 17) or several minutes,(18, 19) which violates the second proposed criterion (low latency). This misuse (in our opinion) is due to researchers considering only the acquisition time, also known as the true temporal resolution,(20) as opposed to the total imaging time (including reconstruction).(7-11, 16-19, 21-23) There have been a few recent papers that aim to define RT-MRI for specific experiments. Lingala et al recommends a definition for RT-MRI in application to speech imaging,(1) which proposes a maximum latency criterion of 500 ms. Campbell-Washburn et al outlines RT-MRI for cardiac interventions, which recommends a latency of <200 ms for tolerable device navigation.(24) While the value for minimum latency will clearly be heavily dependent on the anatomic site and intended image use (and hence require careful and rigorous determination), these publications underscore the importance of the latency criterion in real-time applications. RT-MRI is a heavily researched area in MR, and as such, we propose that care must be taken as to what experiments are considered RT-MRI. It should be noted that RT-MRI is inherently coupled to computational hardware (i.e., CPUs, GPUs) used for image reconstruction. Techniques that currently have reconstruction times that result in a long delay between data acquisition and visualization may one day be reasonably considered real-time based on improved hardware and algorithm development alone. However, predictions of increased acquisition/reconstruction times alone do not warrant a claim of RT-MRI, although predictions of that nature could well be included in discussion. We look forward to hearing responses from the MR community regarding the definition of RT-MRI. The authors are affiliated with MagnetTx Oncology Solutions, of which B.G.F is co-founder and CEO.