Two Recent Technical Innovations in MRI

Donald B. Twieg

Department of Biomedical Engineering, and Center for Development of Functional Imaging University of Alabama at Birmingham, Birmingham, Alabama, USA

MRI continues to evolve, continually improving in its power to reveal structural detail and function, and in its clinical diagnostic power. Discussed here will be two areas currently being developed.

Multiple-receiver technology to improve sensitivity and speed of MRI scanning

Recently MRI systems have achieved significant gains in speed and sensitivity by using multiple-coil arrays and multiple-receiver channels (1). These improvements are due to two physical facts. First, an array of small coils peripherally distributed around the head have greater sensitivity throughout the head, even in the center, than a conventional single volume coil.  Second, the multiple coils in the array produce separate signals simultaneously, which represent the different spatial patterns of sensitivity of the separate coils. This additional spatial information can be used to augment the spatial information from the gradient encoding process.  Thus the number of spatial phase encoding steps can be reduced, and acquisitions can be speeded up, often several-fold. The signals of multiple rf coils are combined to produce images by software algorithms such as SENSE (2).

In combination with the sensitivity of the newly available higher 3- and 4-Tesla MRI field strengths in commercial MRI systems, multiple-receiver technology facilitates very high resolution imaging. Higher field strengths make it possible to achieve adequate signal-to-noise ratios even for very small pixel sizes, but conventional single-channel signal acquisition requires lengthy spatial encoding times for very high spatial resolution. The acceleration provided by multiple-channel acquisition supplies the additional data acquisition speed to make high-resolution high-volume imaging possible.

The advantages of multiple-receiver methods come at a cost: additional receiver channels remain expensive, and the SENSE algorithm requires extensive additional computations beyond that required by conventional simple Fourier Transform reconstruction. Nonetheless, multiple-receiver technology has become incorporated into a number of clinical systems and continues to be developed.

A new approach for single-shot and functional MRI signal acquisition and reconstruction

In conventional MRI, the incoming MRI signal data are placed into a computational array, usually a square array. In single-shot methods such as EPI, a single excitation pulse produces a series of gradient echoes, which are sufficient to fill the entire array. After the array is filled, it is subjected to a 2-dimensional Discrete Fourier Transform (DFT). The new square array which results from this transform contains complex numbers which represent a spatial map – the image – of the object. The signal data in the original array represent a map of the frequency-domain, or “k-space” content of the object (3). Implicitly, current MRI methodology depends on the assumption that the signal data are measurements of the k-space content of the object; it makes this assumption when it Fourier transforms the data to produce an image.

Although this assumption makes the image reconstruction process simple, it is almost always wrong to some degree, because the object being imaged is changing, even during the signal acquisition: its local phase evolves where it is off-resonance, while its amplitude decreases due to transverse relaxation. These consequences are especially important in single-shot methods, used commonly in functional MRI (fMRI), because signal durations are relatively long, giving time for significant decay and phase evolution to occur. Thus in single-shot MRI, substantial off-resonance geometric errors are commonplace. In fMRI, these errors make it difficult to register functional maps with anatomic reference images, or to interpret spatial relationships within the images reliably.

Functional MRI (fMRI) relies on changes in blood oxygenation, flow, and volume, which occur when neural activity increases in cortical and subcortical gray matter in the brain. This set of hemodynamic changes, termed the Blood Oxygenation Level Dependent (BOLD) effect (4), produces changes in the effective transverse relaxation time, T₂*, which can be observed by obtaining a series of T₂*-weighted gradient-echo images. Most commonly, rapid single-shot methods such as echo-planar imaging (EPI) are used for this purpose. Sensitivity to T₂* is achieved with long gradient echo times TE (typical TEs are about 25 milliseconds for 4 Tesla MRI systems, and about 40 milliseconds for MRI systems operating at 1.5 Tesla).

We discuss here a new approach to functional MRI acquisition (5) which uses a more accurate signal model than the conventional MRI approach: instead of ignoring decay and off-resonance effects in the signal, it acknowledges them.   In this new approach, each signal sample is regarded as a sample of a distribution in (k,t) space, rather than a distribution in k-space. Thus the signal is modeled by a sum of decaying and precessing magnetization vectors, each described by an initial magnitude, a decay rate, and an angular rotation (precession) rate. These parameters are then estimated by finding the set of parameters (i.e., separate images of local magnetization amplitude, local frequency, and local decay rate) which give the best match to the signal observed. In mathematical terms, we obtain an inverse solution of the signal observation equation. This technique is a single-shot example of a broader set of methods which apply parameter estimation to the set of individual data, which we term PARSE, for Parameter Assessment by Retrieval from Signal Encoding. Thus the single-shot technique is known as SS-PARSE.

In order for this inverse solution approach to work well, the pattern of data acquisition in (k,t)-space must be appropriate. We have found that conventional multiple-echo EPI or spiral scans behave more poorly than a continuous rosette acquisition, because they lead to a mathematically poorly-conditioned inverse problem.

One of the major advantages of this approach is that it produces images completely free of geometric distortion or blurring in areas where there are usually large variations in resonance frequencies, such as in the lower portions of the brain adjacent to the auditory canals and sinuses.   Another advantage is that it produces direct estimates of local T2* and frequency. Initial results verify that SS-PARSE eliminates geometric off-resonance errors, and can more accurately measure relaxation rate changes such as those due to BOLD functional changes than can conventional multipl-echo single-shot techniques. Thus the method promises to permit increasing accuracy and robustness of BOLD relaxation measurements for functional MRI.

The SS-PARSE method (5) is now actively under development, as are related PARSE methods using other signal models.

(1)   P.B. Roemer et al., The NMR Phased Array, Magn Reson Med 16:192-225, 1990

(2)   K.P. Pruessmann et al., SENSE: Sensitivity Encoding for Fast MRI, Magn Reson Med 42:952-962, 1999

(3)   D. B. Twieg, The k-trajectory formulation of the NMR imaging process with applications in analysis and synthesis of imaging methods, Med Physics 10:610-621, 1983

(4)   S. Ogawa et al., Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging, Proc Natl Acad Sci USA, 89:5951-5955, 1992

(5)   D. B. Twieg, Parsing local signal evolution directly from a single-shot MRI signal: A new approach for fMRI, Magn Reson Med 50:1043-1052, 2003