Reading notes from Sarracanie et al 2015 and additional refs.
Low field MRI presents significant advantages. It is known that the relaxation times are dependent on the magnetic field strength. In low fields, the relaxation time, especially T1, is shorter and the relative differences of T1 between different tissues are larger (Sepponen RE et al 1985).
The potential of low field MRI had been demonstrated by the pioneering work of Sepponen RE et al (Sepponen RE et al 1985) who conducted brain MRI in 20 mT. I
n 1993, the concept of pre-polarized MRI (PMRI) was introduced which consists of the use of a strong, inhomogeneous pulsed magnetic field to generate increased nuclear polarization and a much weaker homogeneous magnetic field for signal detection. The PMRI technique was utilized for all low field MRI systems since then.
Ultra low field MRI consists of the use of a detection field below 10 mT. Numerous PRMI ultra low field MRI studies have been conducted using SQUID-magnetometers for detection. A significant effort is represented by the ULF MRI project by the Los Alamos National Laboratory (cf. Zotev et al 2009 and media article) which performs brain MRI with prepolarization at 0.01–0.1 T followed by SQUID-detection.
A Berkeley National Lab/Berkeley University group conducted in 2013 brain MRI using proton prepolarization at 80 mT and SQUID magnetometers for detection at 130 μT (Inglis et al 2013). An atomic magnetometer study in the PMRI regime was published by the Los Alamos National Laboratory (Savukov I et al 2013) with pre-polarization at 80 mT and detection at 4 mT.
In 2015, researchers from the Low-Field MRI lab of the MGH Martinos Center for Biomedical Imaging, a leading center which introduced fMRI and MRI contrast agents, presented brain MRI in the ULF regime without prepolarization or cryogenics combining undersampling strategies and a high performance fully refocused steady-state-based acquisition i.e. specific MRI sequences (b-SSFP) (Sarracanie et al 2015).
Investigation of algorithmic pattern recognition for matching to predicted signal evolutions (Magnetic Resonance Fingerprinting - MRF)
Reading notes from above and references
The technique uses an ultra-low field of 6.5 mT while no pre-polarization or cryogenics are utilized. It is noted that the Larmor frequency at this magnetic field strength it 276 KHz. It also uses balanced Steady-State-Free-Precession (b-SSFP)* sequences which dynamically refocus the spins after measurement, thereby eliminating the delays associated with T2 decay and T1 recovery (Sarracanie M et al 2013). This reduces acquisition time and provides the highest signal-to-noise ratio (SNR) per unit time of all imaging sequences. These sequences are very sensitive to the spin dephasing which occurs between consecutive RF pulses and therefore are highly susceptible to inhomogeneities of the magnetic field, major cause of spin decoherence. A significant technical requirement associated with high magnetic fields is the need for high homogeneity as even small heterogeneities can introduce very important artifacts. This requirement is less important at low magnetic fields and as a result at 6.5 mT, b-SSFP sequences are largely immune to heterogeneity artifacts.
Additionally, an undersampling stategy is pursued meaning that sparse samples are obtained for the reconstruction of the image. This favors the selection of samples representing image features (large coefficients) while reducing the samples representing noise and artifacts (small coefficients).
Also, a technique termed "Magnetic Resonance Fingerprinting", which is similar to "compressed sensing" is investigated (Ma D et al 2013). This consists of pseudorandomized acquisition which causes signals from different tissues to have a unique signal evolution of "fingerprint" that is simultaneously a function of the multiple constituents properties. Following acquisition, processing with a pattern recognition algorithm allows for matching of the fingerprints to a predefined dictionary of predicted signal evolutions. These can be translated to quantitative maps of MR parameters. It must be mentioned that the technique allows for the simultaneous examination of many MR parameters thereby enabling computer-aided multiparametric MR analyses, similar to genomic or proteomic analyses, which can detect important complex changes across a large set of MR parameters simultaneously. Also, use of appropriate pattern recognitions algorithms allows to minimize the effect of noise and artifacts, practically suppressing these factors.
Furthermore, there are theoretical frameworks for image reconstruction of highly undersampled datasets using multiple channel acquisition and parallel imaging i.e. different coils. Finally, parallelized computing enables to address computationally demanding tasks.