Parallel Imaging

Parallel imaging offers a number of advantages, like faster examination times, when applied to standard MR protocols. Rather than relying on rapid gradient switching to speed acquisition, parallel imaging uses clever algorithms and radiofrequency technology to acquire smaller data sets, specifically by undersam-pling fc-space [30, 54, 56, 66].

The basis of parallel imaging is the use of the spatial information of the radiofrequency field that is associated with the individual elements of an RF coil array and the use of this information in concert with conventional gradient-based image encoding. Therefore fewer phase-encoding steps have to be performed. It improves data acquisition speeds beyond what can be achieved with conventional non-parallel approaches without imposing additional stress on the gradients (Fig. 11.1). With this method, fully encoded images are reconstructed from undersampled data sets, yielding large savings in scan time. On the other hand, the reductions in scan time incur a cost in both the complexity of the scan and the signal-to-noise ratio (SNR) of the final image [45]. The complexity of the scan arises from the fact that all parallel MRI techniques rely on the knowledge of coil sensitivities, which necessitates some form of calibration. In addition, the SNR loss of every image takes place for two reasons. First, there is a reduction in the temporal averaging of noise associated with the fact that fewer fc-space points are used in the reconstruction. This leads to an SNR loss proportional to the square root of the reduction in scan time. This has the same effect as in conventional imaging with rectangular FOV [3]. Second, there is amplification of noise because, unlike in conventional Fourier imaging, the transformations used in coil-encoded image reconstruction are not unitary. This leads to an additional spatially dependent source of SNR loss that is quantified by the so-called geometry factor [49]. It is this geometry-associated loss that can be ameliorated through good coil array design and complex algorithms of reconstruction.

Parallel imaging (PI) methods can be categorized into two groups. In the first the missing fc-space lines are calculated before Fourier transforming the data and this kind of algorithm is called Simultaneous Acquisition of Spatial Harmonics (SMASH) [67]. It can be truly considered to be the start of a new approach in MRI technology. The third generation of this algorithm is the GeneRalized Auto-calibrating Partially Parallel Acquisition (GRAPPA) that is now available for clinical routine imaging. The second type of PI method first reconstructs images with reduced FOV for all receiver

Fig. 11.1. Basics of fc-space for faster MRI: MRI can be performed faster by acquiring fewer fc-lines per MR image. (Courtesy of GE Healthcare Technologies)

coil elements and then blends these different images, by utilizing the knowledge of the spatial distribution of individual coil sensitivity, into one image with full FOV. The algorithm of this second type is called SENSitive Encoding (SENSE) and the modified algorithm mSEN-SE is also available as packages on clinical scanners.

Parallel imaging, regardless of in which domain image reconstruction is performed, requires some additional information about the spatial coil sensitivities; in other words providing information about which part of the FOV is covered by each coil element or from which area the coil does receive its MR signal. A homogeneous sample will present itself with a different signal intensity depending on the distance from the coil (principle of reciprocity) [31]. For PI the coil sensitivity information can be acquired as a separate scan or, typical for the GRAPPA and mSENSE algorithm, by additionally acquiring some of the missing data lines in the centre of fc-space (so-called reference or auto-calibration lines) integrated into the acquisition [3].

In parallel imaging the acceleration of imaging is due to the reduced number of phase encoding lines that need to be acquired to form an image at a given FOV and resolution. The lines are replaced by exploiting the spatial information that is inherent in the spatially variable sensitivity of an array of surface coils. The number of receiver elements determines the maximum time reduction factor (R), which indicates the level of im provement to acquisition speed [38]. With an acceleration factor of 2, only every other line in fc-space is acquired and the imaging is virtually cut in half.

Several different techniques for parallel imaging have been proposed by different MR system vendors (ASSET, IPAT, SENSE); they currently allow a reduction of phase encoding steps by a reduction factor of 2 to 6 (and recently even higher). We can try to examine them separately.

Sensitivity encoding (SENSE) is a parallel imaging technique that trades signal-to-noise for speed. Just as S/N is reduced by 40 % whenever the number of excitations is halved in a conventional spin-echo image, the same is true of SENSE: A SENSE factor of 2 halves the acquisition time and reduces the S/N by 40%. SENSE is therefore most useful when there is excess S/N, such as at 3 T. It requires the use of phase-array coils, and the number of coil elements determines the theoretical limit on the SENSE factor. SENSE works by intentionally reducing the field-of-view and the number of phasing encoding steps, which in turn reduces the acquisition time [5]. This would normally lead to wraparound artefact, or „aliasing", but because the local sensitivity of each coil in the phase array is known, the aliasing can be unwrapped.

SENSE uses the knowledge of coil intensity (sensitivity) profiles to reconstruct undersampled data sets, post-Fourier transform [53]. Sensitivity assessment re quires that low-resolution, fully Fourier-encoded reference images are obtained with each array element and with a body coil prior to parallel imaging. Element-wise division of the array references by the body coil references yields raw sensitivity maps. This means that body coil homogeneity forms the basis of homogeneity correction accomplished by SENSE reconstruction [38]. Raw sensitivity maps are refined by a fitting procedure that performs noise elimination and sensitivity extrapolation. Regions that do not contribute signal, according to the references, are automatically excluded from SENSE reconstruction (Fig. 11.2). Another form of SENSE is Modified SENSE (mSENSE) The difference is that mSENSE uses autocalibration and does not require a reference scan to calculate sensitivity maps [74]. This is achieved by splitting fc-space into two regions: a central, fully sampled region from which information about coil sensitivities is derived, and an outer undersampled region as in generic SENSE.

The other technique is GRAPPA (generalized auto-calibrating partially parallel acquisition); it performs reconstruction in the fc-space domain [25]. GRAPPA acquires autocalibration signal (ACS) lines along with the reduced data set, and no reference scan is required. The scanner uses the signal lines to estimate a series of weighting functions, which are used to calculate the unacquired lines. When all lines are reconstructed for a particular coil, a Fourier transform can be used to generate the uncombined image for that coil. This process is repeated for each coil of the array to produce a full set of uncombined images, which can then be combined using a normal „sum of squares" reconstruction. GRAPPA uses several blocks of data to fit each missing line. Additional acquired calibration lines can also be included to improve image quality.

The advantages of using this kind of technique are considerable. The SENSE-mediated reduction of acquisition time can be traded for improved temporal and spatial resolution of any given pulse sequences, without change of image contrast. In addition to the mere increase in image acquisition speed, the reduction of phase-encoding steps brings two further advantages [35]: First, in single-shot EPI applications that are usually used for diffusion imaging, diffusion tensor imaging or functional BOLD-contrast MR studies, SENSE helps to shorten the echo train length in proportion to the reduction factor. The shorter echo train also reduces the phase errors during the EPI readout and the susceptibility effects such as image distortions and blurring. In addition, the shorter echo train translates into a significantly higher SNR.

PI helps to shorten the length of the echo train without loss of spatial resolution in single-shot turbo/fast spin echo (HASTE, SSFSE) too [25]. These sequences

Ssfse Asset

Fig. 11.2. ASSET, SENSE, andIPAT use the knowledge of coil intensity (sensitivity) profiles to reconstruct undersampled data sets, post-Fourier transform. Element-wise division of the array references by the coil references yields raw sensitivity maps. Raw sensitivity maps are refined by a fitting procedure that performs noise elimination and sensitivity extrapolation. Regions that do not contribute signal, according to the references, are automatically excluded from reconstruction. (Courtesy of GE Healthcare Technologies)

Fig. 11.2. ASSET, SENSE, andIPAT use the knowledge of coil intensity (sensitivity) profiles to reconstruct undersampled data sets, post-Fourier transform. Element-wise division of the array references by the coil references yields raw sensitivity maps. Raw sensitivity maps are refined by a fitting procedure that performs noise elimination and sensitivity extrapolation. Regions that do not contribute signal, according to the references, are automatically excluded from reconstruction. (Courtesy of GE Healthcare Technologies)

often suffer from image artefacts because of their long echo train and can appear blurred because of the T2-re-lated signal decay during the readout of echo train. Second, SENSE helps to reduce RF deposition (regular phase encoding requires an RF pulse for every step). This is extremely helpful for high-field imaging where specific absorption rate (SAR) increases with the square of B0. The higher the field strength, the more energy patients absorb with excessive heating of the patient. In fact SENSE can also decrease the number of echoes in TSE sequences at a given spatial resolution and thus reduce the specific absorption rate. Acceleration factors of up to six are feasible because of the intrinsically high SNR obtained at 3 T [21]. The combination of SENSE with 3 T makes an ideal partnership for optimized high-field MR protocols [77].

The shorter examination time obtained with parallel imaging is therefore useful for claustrophobic patients and patients who find it difficult to remain still.

Using GRAPPA it is possible to take cervical scans with fewer motion artefacts, because GRAPPA provides a greater signal and less ghosting than mSENSE [38].

While the maximum number of parallel receive channels on clinical scanners is currently in the vicinity of eight, Zhu et al. [79] presented a 32-element receive-coil array and a volumetric paradigm that address the SNR challenge at high accelerations. Each system represents the integration of multiple sets of MR electronics, including analog-to-digital converters and digital data pipelines, into a single system. All receivers in the system multiple sets of electronics were synchronized to each other, which in effect expanded the number of parallel receivers to a total of 32. Pulse sequences were adapted to work with the synchronization mechanism, and custom software was developed to facilitate scan control and data communication.

Parallel imaging facilitates tradeoffs between acquisition time, spatial coverage and image SNR, which has significant implications for the volumetric paradigm. While acquisition time tends to increase linearly with the expansion of volume coverage using a multislice approach, a more manageable time increase can be expected of the same expansion using a parallel imaging-based volumetric approach [79].

Parallel Imaging Applications

1024 MR Angiography

Another application of parallel imaging is in MR angiography, where the use of phase-array coils has the advantage of a higher S/N than standard quadrature RF coils. The greater S/N of these coils and the higher field strength of 3 T MR can be traded for speed or higher spatial resolution at the same acquisition time. This leads to 1024 MR angiography. The acquisition time for 3D time-of-flight MRA is TR x Np x Ns, and increasing the number of phase-encoding steps (Np) to 1024 would either lead to an excessively long acquisition time or mandate a reduction in TR. On the other hand, a very low TR limits the time available for inflow of un-saturated blood and reduces flow-related enhancement, which is the basis for TOF MRA. Increasing the matrix size from 512 to 1024 along one axis decreases S/ N by 50% (assuming FOV is held constant). Halving pixel dimension along both the phase and frequency axes reduces S/N to 25%. For all these causes the use of 3 T MR and the latest generation of phase-array coils with a rectangular FOV allows the same spatial resolution to be achieved, reducing the acquisition time. If we add a SENSE factor [5], we can halve the time acquisition from 14 min, 12 s to 7 min, 6 s (spoiled GRASS sequence with TR 31/TE 6). Although the spatial resolution of MRA can surpass that of DSA, the latter has the advantage of temporal information. Use of the techniques of Temporally Resolved 3D Contrast MRA allows temporal phases to be obtained from an MR an-giogram. They eliminate the traditional trade-off between spatial and temporal resolution in vascular imaging.

Temporally Resolved 3D Contrast MRA

We know that the major advantages of intra-arterial digital subtraction angiography are high-spatial-resolution images and temporal information regarding delayed filling of the vasculature of interest. Time-of-flight magnetic resonance (MR) angiography has proved to be accurate in the depiction of vascular pathologic conditions in the peripheral vasculature. However, long acquisition times and flow-related artefacts have hindered its widespread application [61]. Contrast-material enhanced MR angiography is an alternative that has been shown to be less susceptible to the saturation effects that are common with time-of-flight imaging. To optimize image contrast, the maximum concentration of gadolinium should be present during acquisition of central k-space lines during peak arterial enhancement [40]. This is normally achieved by using a test bolus injection and fluoroscopic or automated triggering [28]. A final approach has been to collect 3D image data sets rapidly enough (typically 1-10 s per

Fig. 11.3. Intracranial AVM. TRICKS images (3D TOF SPGR, TR 3.7, TE 1.3). Matrix 224x 192, Thk 1.8 mm, FOV 28, temporal res. 2.5 s, 8 phases. They show multiple vascular phases (contrast enhancement of arteries, capillaries and veins). No coordination of contrast agent arrival and image acquisition was required

acquisition) that at least one data acquisition is aligned with the arterial phase of the contrast material bolus [52]. Besides eliminating the timing problem, such a new „time resolved" approach provides temporal information about relative rates of contrast enhancement of arteries, parenchymal tissue, and veins [72]. With time-resolved techniques, bolus timing is no longer necessary, as multiple vascular phases are obtained without any predetermined timing, contrast injection and simultaneous beginning of scanning (Fig. 11.3). The operator simply selects the desired image set afterwards: pure arterial phase, maximum venous, etc. The most straightforward way to accelerate acquisition time is simply by scanning faster. This can be done using some combination of limiting the imaging volume, decreasing the resolution, decreasing TR, or using a parallel imaging technique such as SENSE [40]. Recent developments in gradient system allow TR's of < 2 ms in many systems. Even if the temporal resolution is insufficient for isolating the desired vascular phase, postprocessing techniques such as correlation analysis can be used to synthesize images of pure arterial versus pure venous phase.

Using 3D TRICKS (time resolved imaging of contrast kinetics) and its variants it is possible to obtain simultaneous high spatial resolution and high temporal resolution over a large field of view.

The basis of this technique is to combine the repeated sampling of the low-spatial-resolution fc-space views with temporal interpolation to produce a series of time-resolved imaging of contrast kinetics (TRICKS) three-dimensional MR angiographic images. Three-dimensional TRICKS images reveal the dynamics of contrast material arrival and obviate a timing image [34]. The prescribed fc-space is divided into a number of regions in a TRICKS acquisition [12]. Accordingto Caroll and Vigent [11,73], the regions can be three (A, B, C) or four (A, B, C and D), and the regions are divided along the phase-encoding (fcy) direction. The central region (A) contains the lowest one-third of spatial frequencies in the phase encoding direction, and successive regions correspond to higher spatial frequencies in this direction. During each TRICKS time frame, a segment of fc-space corresponding to one region is acquired. Central fc-space, which contributes most to overall image contrast, is repeatedly collected every 2 - 8 s in an alternating fashion, with other frames of more peripheral fc-space collected less frequently. Images are then „syn-thesized" at high temporal resolution by piecing together each unique block of central fc-space with a linear interpolation of the remaining fc-space frames acquired in closest temporal proximity. In this way a high-resolution reference image is collected with N phase encoding steps dimension. The fc-space segment acquisition time AT (and reconstructed frame period) is proportional to the acquired spatial resolution [73].

Temporal resolution is characterized by both the repetition time T of each fc-space segment and the width of the acquisition window necessary to reconstruct an image set [73].

Two parameters are more important to the understanding of the TRICKS acquisition: the frame rate and the temporal aperture [12]. The frame rate is the rate at which TRICKS segments are acquired and it is calculated as TRxNzx(Ny/Nr), where TR is repetition time, Ny is the number of phase-encoding values, Nz is the number of sections, and Nr is the number of TRICKS regions. It is possible to acquire the highest spatial resolution with a time between acquisition of a region A no greater than the time between arterial and venous enhancement. This ensures that region A is acquired vein-free and combined with regions, which are acquired later. The time during which data for a single frame are acquired is the temporal aperture. It is used to determine the length of the contrast agent injection. The largest temporal aperture, which corresponds to the acquisition of all regions of fc-space [12], is calculated using TR x Nz x (Ny/Nr) x (Nr+1). A high variation in the arrival time of the contrast agent curve can introduce ringing and blurring artefacts, increased by the temporal interpolation [13]. It is much better for a low injection rate, which provides a longer contrast agent bolus to increase imaging time in order to improve spatial resolution, to acquire a greater number of arterial phases and to reduce intravenous concentration. TRICKS does not need a very precise synchronization between acquisition and bolus arrival, providing that the mask is not corrupted by contrast. The TRICKS technique, in fact, uses the combination of an integrated mask and complex subtraction and the detection of the peak arterial frame for selective reconstruction [11]. Since the integrated mask image subsequently removes contrast material already present in the veins from the previous injection, a clean arterial frame can be obtained. In addition, with the application of the peak arterial time-frame technique, the signal intensity in the temporally acquired fc-space regions is analysed and those frames with increasing signal intensity are used for reconstruction of a clean arterial frame. This procedure is performed in fc-space and not in image space with reduction of post-processing time [27].

Therefore this technique allows greater temporal and spatial resolution/volume coverage than with a conventional sequence and it lends itself to the „video" format, where passage of contrast material can be directly viewed.

The biggest drawback, other than the large number of images generated, is that fc-space discontinuities, in conjunction with varying intravascular gadolinium concentration, can potentially lead to artefacts. A recent refinement of this strategy is to traverse fc-space using radial or spiral projections, which allow for slid

Fig. 11.4. Spinal MRA. Coronal TRICKS images (3D FAST SPGR, TR 4.3, TE 1,1). Matrix 384x 192, Thk 2 mm, FOV 35, temporal res. 0.04 s, 9 phases. Coronal MIP reconstruction: the images show different enhancing phases of a dural fistula

ing window reconstructions at very temporal and spatial resolutions (Fig. 11.4). One variation, known as isotropic projection reconstruction (VIPR), is particularly promising [40].

The advantages of temporal resolved 3D contrast MRA, like TRICKS, are: to eliminate timing and triggering, to isolate a pure arterial phase even with asymmetric flow, to eliminate venous contamination, to capture flow dynamics, to characterize filling like conventional angiography, and to enable multiple injections through automated complex subtraction. Moreover, the temporal resolution of 3D dynamic imaging is accelerated without sacrificing spatial resolution (Fig. 11.5). Spectroscopy

MR spectroscopic imaging (MRSI) can benefit from both high field strength and parallel imaging techniques. The acquisition of MR spectroscopy (MRSI) allows information to be obtained on the spatial distribution of metabolites in a lesion and surrounding tissue, the determination of the extent of the lesion, and the detection of contralateral lesion. Spectroscopy gains in multiple ways from increased field strength: besides a crucial increase in SNR, the chemical shift difference between different metabolites increases with field strength, allowing for better separation and identifica-

Fig. 11.5. Spinal MRA. Sagittal TRICKS images (3D FAST SPGR, TR 4.3, TE 1,1). Matrix 252X252, Thk 1 mm, FOV 35, temporal res. 0.15 s, 2 phases. Coronal MIP reconstruction: the image shows the Adamkiewicz artery

tion of metabolites. This increased chemical shift difference maybe used to obtain higher SNR or additional speed in advanced techniques such as multi-echo spectroscopic imaging (TSI), similar to RARE, FSE or TSE in imaging. This is because in TSI the echo spacing determines the achievable spectral resolution as well as the maximum tolerable number of echoes in the echo train. A field strength of 3 T enables a twofold shorter echo spacing and thus longer echo trains while maintaining the same spectral separation between metabolites as at 1.5 T [17]. The scan time required to achieve acceptable resolution and SNR for MRSI can become too long for practical use in a clinical examination. By using up to six spin-echoes in the echo train, acquisition times for MRSI are reduced significantly. Dydak et al. [15,16] in two recent works have proposed using the technique of parallel signal acquisition to reduce MRSI scan time and produce metabolite maps in patients with cerebral gliomas. They proposed undersampling the two in-plane phase-encoding dimension of a conventional 2D PRESS MRSI acquisition and using SENSE reconstruction. A reduction factor of 2 was used in each spatial direction, resulting in a net factor of 4 improvement in scan time to achieve a comparable spatial and spectral resolution. It is possible also to achieve a 3D SENSE-MRSI acquisition (24 X 24 X 8 slices) within 14 min. The flexibility of parallel imaging techniques allows the combination of two methods, enabling subminute scan times for single slice or a 3 min scan time for 6 slice MRSI scans with high spatial resolution. This will allow in the future the consideration of dynamic or functional MRSI (fMRSI) experiments with good tem poral resolution, or the imaging of dynamic metabolic processes covering the whole brain [17].

Was this article helpful?

0 0
Unraveling Alzheimers Disease

Unraveling Alzheimers Disease

I leave absolutely nothing out! Everything that I learned about Alzheimer’s I share with you. This is the most comprehensive report on Alzheimer’s you will ever read. No stone is left unturned in this comprehensive report.

Get My Free Ebook

Post a comment