Synopsis

Keywords: Physics & Engineering: Low-Field MRI

Low-field MRI systems, historically seen as underperforming, are gaining renewed interest due to advancements in technology. These systems now enable techniques previously limited to high-field MRI, offering considerable clinical value with improved imaging modalities like 3D SWI at 0.55T. The application of AI and deep learning in image reconstruction and noise reduction enhances image quality and reduces imaging times. Despite inherent limitations in signal-to-noise ratio and spatial resolution, low-field MRI provides unique advantages, especially in reducing susceptibility artifacts near metal implants, making it a valuable tool in clinical diagnostics and MRI-guided interventions.

Abstarct

Historically, clinical MRI systems have evolved towards higher static magnetic field strengths. Initially, the performance of low-field MRI systems was perceived as inferior to their actual clinical utility, being considered low-cost yet underperforming devices. However, images obtained at low-field MRI systems around the year 2000 are now recognized to have considerable clinical value even by today's standards. Techniques such as diffusion-weighted imaging (DWI), despite challenges such as distortion in EPI methods, can be effectively compensated by employing alternative imaging techniques based on the spin-echo method (1). Nevertheless, due to inherently lower signal-to-noise ratios (SNR), low-field MRI systems tend to require longer imaging times compared to their 1.5T and 3T counterparts.
In recent years, advancements in technology, both hardware and software, have renewed interest in low-field MRI systems. These advancements have enabled imaging techniques previously exclusive to high-field MRI systems (1.5T, 3T) to be utilized in low-field settings (2). For instance, specific imaging modalities like T2*WI, traditionally seen as inferior in low-field systems for detecting microhemorrhages, have seen significant improvements. Imaging modalities such as 3D SWI at 0.55T are now possible, enhancing clinical utility beyond previous capabilities.
Furthermore, the application of deep learning and Artificial Intelligence in image reconstruction and noise reduction has not only improved image quality but also contributed to reducing imaging times. However, training deep learning models on high-field MRI images is not always advisable due to differences in relaxation times influenced by the magnetic field strength, which affect image contrast.
Moreover, high-performance low-field MRI systems propose unique advantages over high-field systems in various aspects of clinical application. One of the most significant advantages is the reduced susceptibility artifacts, particularly beneficial near metal implants, air, or bone, where high-field systems may fail to provide informative images. This advantage facilitates MRI-guided interventional procedures(3).
While low-field MRI systems may not match the performance capabilities of high-field systems in research settings, clinically, they can provide valuable information not obtainable through CT or ultrasound examinations, even with lower SNR and spatial resolution. This positions low-field MRI as a valuable tool in the diagnostic landscape, offering distinct benefits in terms of safety, accessibility, and specific imaging requirements.

Acknowledgements

This work was supported by JSPS KAKENHI Grant Number 24K10847.