M R I Physics Course Multichannel Technology & Parallel Imaging Nathan Yanasak, Ph.D. Jerry Allison Ph.D. Tom Lavin, B.S. Department of Radiology Medical College of Georgia
References: 1) The Physics of Clinical MR, for Neuroradiology, Taught Through Images 2) The Physics of Clinical MR, Focusing on the Abdomen, Taught Through Images AUTHORS: VAL M. RUNGE1 MD, WOLFGANG R. NITZ2 PHD, STUART H. SCHMEETS2 BS, RT, WILLIAM H. FAULKNER, JR.3 BS, RT, NILESH K. DESAI1 MD 2
Multichannel Coil Technology (basics) Radiological Wish List for MR (and perhaps, other modalities as well): Higher spatial resolution Decreased acquisition time Higher signal to noise ratio (SNR) More images per patient for more diagnostic information Minimize SAR (problem with high-field MR) 3
Multichannel Coil (cont.) Some factors to help meet these needs: 1) Protocol/Pulse-sequence optimization 2) Faster image reconstruction hardware but also 3) Single-element coil: increasing SNR requires increased acquisition time 4) Receiving coil element size: decrease increased SNR per volume, but smaller tissue volume 5) Tissue proximity: decrease increased SNR But: Single-element coil + one receive channel = slow dataflow. Solution: Number of RF receive channels: increase decreased acquisition time 4
Multichannel Coil (cont.) Circularly-polarized (CP) coil led to a ~ 40% increase in SNR (two-element coil). Also called a quadrature coil. Recent development: Multichannel technology Coil uses multiple elements ( loops ) in phased array with overlapping anatomical coverage Each element acquires MR signals from the entire region. Highest signal in closest proximity to the element. Small element size higher received signal / higher overall signal RF hardware uses multiple channels for receiving signal from multiple elements. 5
Multichannel Coil (cont.) Putting the coil and RF system together Signal from each element is (ideally) transferred through its own high-bandwidth RF channel. Reconstruction corrects element-to-element signal variations before forming the final image. Advanced reconstruction and storage hardware necessary to process the rapid inflow of information. 6
Multichannel Coil (cont.) Figure 1 shows element arrangement for an 8element head coil, and images acquired from each element, together with the final single combined image. Note that images from each surface element show greater sensitivity near the element. 7
The scan illustrated is a fat suppressed FLAIR from a patient with brain metastases, with the edema from a metastasis just superior and posterior to the lateral ventricles visualized. Figure 1 (ref #1) 8
Multichannel Coil (cont.) Commercially available, eightchannel coil for brain imaging. (MRI Devices Corporation, Waukesha, WI). We use this coil with our GE 3T. Ref. #1 9
Multichannel Coil (cont.) Figure 2 shows element arrangement for a 12element body coil, with six of the elements activated, and images acquired from each element together with the final single combined image. Typically, the scanner software routinely reconstructs the signal from all elements and provides only the combined image. Note that images from each surface element show greater sensitivity near the element. 10
The scan sequence employed was truefisp with spectral fat saturation. This image depicts two large liver hemangiomas. Figure 2 (Ref #2) 11
Multichannel Coil Example #1 Case study: Axial T1- and T2-weighted images of the brain Figures 3A,C (on left) use a standard CP coil Figures 3B,D (on right) use an eight element, phased array coil MR system uses eight, high-bandwidth, RF receive channels All pulse-sequence parameters same between left and right figures. T1 comparison shows increase in SNR, improved gray versus white matter distinction, and improvement in anatomic definition (e.g., cortical gyri). T2 comparison shows increase in SNR, improved definition of gray matter, and improved visualization of the gray matter nuclei. 12
Figures 3A,B,C,D (Ref #1) 13
Multichannel Coil (cont.) Clinical benefit of multichannel technology: Higher SNR achieved with multichannel technology allows greater flexibility in sequence parameter selection. If SNR is higher than needed, we can afford to lose a little SNR to gain: An increase in spatial resolution A reduction in acquisition time (e.g., minimize motioninduced artifacts, increase number of images per exam). 14
Multichannel Coil (cont.) Advancements in multi-element/multichannel technology (to 32 elements and beyond) will continue to play a role in the development of imaging techniques with higher spatial resolution, faster scan times, and increased diagnostic quality. 1) MCG has an 8-element GE 3T scanner (Sept. 2005) 2) UGA has GE 3T research scanner (8-16 channels summer 2006). 15
Multichannel Coil (cont.) Advancements in multi-element/multichannel coils: New 96-channel head coil (Wald, MGH) High-field imaging with 8-channel coil 16
Parallel Imaging As shown previously, no image from a single surface coil element is optimally sensitive over the whole area. However, an image reconstructed from all coil elements leads to an increased SNR over a standard acquisition, because each region of the image is reasonably sampled by more than one element. If SNR is higher than needed, one can use the technique of parallel imaging to increase acquisition speed. How? We can decrease sampling of data by each element receiver. Also, reduced sampling less RF excitations per unit time lower SAR. Decrease sampling of data = decreased k-space sampling 17
Parallel Imaging Rather than fill all of k-space, parallel imaging acquires a fraction of k-space to save time. Because the anatomy is sampled by multiple coil elements, we can reconstruct the missing information (more or less). Less samples, of course, leads to decreased SNR. But, if our multi-channel SNR is better than we diagnostically require, so what? 18
Parallel Imaging How fast can we go? If we have M coil elements covering the FOV, we can skip up to M-1 lines for each line in k-space we sample. The number of lines skipped : acceleration factor (R). This can be fractional as well: # of phase-encodes to cover k-space R = # of phase-encodes used in acquisition Names for acceleration factors: ipat factor (Siemens) SENSE factor (Philips) ASSET factor (GE) Parallel systems: ipat (Siemens) SENSE (Philips) ASSET (GE) 19
Parallel Imaging Example #1: We have an 8-element phased-array head coil. We want an acquisition matrix of 256 x 256. What is the maximum acceleration factor we can achieve? Answer: If we have M elements, we can skip up to M-1 lines in k-space. So, M=8, and M-1=7. In the case of this acceleration, for each 8 lines within k-space, we are acquiring only 1 of these line. # of phase-encodes to cover k-space R = # of phase-encodes used in acquisition = 256 phase encodes / { (1 acquired line/8 lines) * 256 } = 8 20
Parallel Imaging Increasing acceleration leads to decreasing SNR. However, the benefits may be greater than saving time as well. For EPI images, which are greatly affected by susceptibility differences, parallel imaging can improve geometric distortion and/or image voids. Because the gradients are switching so quickly for an EPI image, one can accrue errors that lead to distortion. These are alleviated using parallel imaging, where the sequence requires less lines in k-space to be read out. 21
Parallel Imaging Example of Parallel Acceleration on the GE 3T: R=1 R=2.0 R=2.8 R=3.2 R=4.0 22
SNR vs. Acceleration Short-axis cardiac images 32-channel coil 1.5 T magnet Reeder SB et al. MRM 54:748,
Parallel Imaging (cont.) Spatial coil sensitivity = function describing the sensitivity of the coil element at any particular position in the FOV. (Ref. #2) R C1 Total C2 L Both types of parallel imaging techniques rely on this function. 24
Parallel Imaging (cont.) How is the spatial sensitivity measured? a b c d Method #1: Acquire quick images from each element (a) and reconstruct the full image using all elements (b). Image (a) divided by (b) gives a noisy sensitivity map (c). Filtering smoothes out the noise, yielding our sensitivity map (d). In short, with this method, one must acquire a map before running a parallel imaging sequence. Takes a minute or so. If one uses the summed image from all elements as a reference, this technique is called Auto-SENSE. 25
Parallel Imaging (cont.) Method #2: During the parallel scan, we can acquire extra data in the very center of k-space, using the number of phase encodes in this region that we would have used without parallel imaging. Because the center of k-space is responsible for low spatial resolution, this will also give you spatial sensitivity maps for each coil element. (This is the basis of AUTO-SMASH, VD-AUTOSMASH, and GRAPPA ). Key: White=filled part of k-space Black=unfilled k-space 26
Parallel Imaging Two main types of parallel imaging: image based reconstruction SENSE, msense, ASSET k-space based reconstruction SMASH, GRAPPA 27
Parallel Imaging (Image Based Reconstruction) Image-based reconstruction is, in principle, easier to understand than k-space-based reconstruction. If data is acquired with less phase encodes than will fill k-space, the reconstructed image will show aliasing. We ve seen this before: less phase encodes in the same region of k-space smaller FOV. If FOV is smaller than the object, we get aliasing. (from Boesiger & Pruessmann, http://www.mr.ethz.ch/sense/sense_method.html) # of encodes 256 128 107 85 28
Parallel Imaging (Image Based Reconstruction) Each pixel in the aliased image (Ialias) is comprised of overlapping (or summed) data from two or more pixels in the unaliased image (I1, I2, ). Use the spatial sensitivity function for each coil element to reconstruct the image intensity uniquely at each position. (from Boesiger & Pruessmann, http://www.mr.ethz.ch/sense/sense_method.html) 29
A Simplistic SENSE Example Ialias,1=s1,A IA + s1,b IB A s1 IA Ialias,1 B IB A s2 B Ialias,2 Ialias,2=s2,A IA + s2,b IB We know s1, s2 (sensitivity maps); we measure Ialias,1, Ialias,2; so we can calculate IA, IB. 30
Parallel Imaging (k-space Reconstruction) Let s review some topics quickly again: 1) What does k-space really represent (we know that MRI collects data in k-space before reconstructing an image)? 2) What is the relationship between spatial structure in an image and waves? 31
Parallel Imaging (k-space basics) Remember: we can decompose a complicated 1-D wave into a combination of simple waves of a given frequency. 32
Parallel Imaging (k-space basics) For each simple component, if we know the amplitude and the phase, we can construct a unique wave. Amplitude change Phase change 33
Here s the representation of the waveform as a plot of amplitudes and phases: Complicated Wave representation f amplitude/phase representation x phase Can transform back and forth. Now, let s look at 2D waves 34
Fourier Transform Basics y Complicated Wave representation K-space representation x Can transform back and forth. K-space is just a 2D (or 3D) version of the amplitude/phase representation. 35
Parallel Imaging (k-space method) Back to 1-D again, for simplicity (i.e., the A-P axis) A Coil 1 Coil 2 Coil 3 C2 C3 P Spatial sensitivities for each element of a multi-element coil are periodically distributed across the FOV. This picture shows three spatial sensitivity functions spanning our imaging FOV. So, each element is sensitive to signal in a particular location. C1 36
Parallel Imaging (k-space method) We can use each individual coil sensitivity to our advantage. We can examine the signals across a full field of view by combining the signals in some proportion from each coil. Depending on how we combine the signals (add or subtract) from each coil into the total, we can enhance or suppress our sensitivity to signals of different spatial variation. Each combination of coil signals would result in an effective sensitivity across the full FOV. 37
Parallel Imaging (k-space method) The black curves represent two of the effective sensitivities using elements in this example. The upper combination is sensitive to Fundamental: CA (add signal for all three) First Harmonic: CB (subtract middle signal) The upper combination is sensitive to these spatial variations: while the lower combination is sensitive to these spatial variations: 38
Parallel Imaging (k-space method) So, each combination is sensitive to spatial variations of different wavelength, or spatial harmonics. This modulation of sensitivity across the FOV mimics modulation of spatial sensitivity by phaseencode gradients. If our acceleration factor is R, we have to reconstruct R of these harmonics, by using R combinations of coil signals. 39
Parallel Imaging (k-space method) (from Sodickson, et al., MRM 41: 1009, 1999) 40
Parallel Imaging (k-space method) How are the missing k-space lines filled in? 1) Spatial sensitivities of each coil are determined. 2) Given an effective sensitivity that we wish to calculate (i.e., sensitivity across the FOV to a sinusoid of a particular periodicity), the actual sensitivities require summation in particular proportions. So, we use the desired effective sensitivity and the measured sensitivities to determine the necessary proportions. 3) The missing k-space lines are calculated by summing k-space data from each coil element, using the proportions determined in the first step. 41
Parallel Imaging (k-space method) Example: The fundamental effective sensitivity covers the original lines in k-space; Data for the first harmonic is shifted up a line in k-space. 42
Parallel Imaging and Noise Noise in parallel images is 1) increased, and 2) non-uniform. As shown in this SENSE example, unfolding the alias multiplies the noise within particular regions (non-uniform). A similar effect appears in k-space based methods. 43 Larkman DJ et al. Magn Reson Med 2006; 55:153-160
Parallel Imaging (k-space Example) K-space oriented parallel acquisition techniques: SMASH (Simultaneous Acquisition of Spatial Harmonics), AUTO-SMASH PILS (Parallel Imaging with Localized Sensitivities) GRAPPA (Generalized Autocalibrating Partially Parallel Acquisition). Figure 5 shows fast spin echo T2 weighted sagittal scans of the lumbar spine, without (A) and with (B) parallel imaging (the latter using GRAPPA). In (B), every second Fourier line has been skipped (an ipat factor of 2, or acceleration factor). Scan time is thus reduced by a factor of two (comparing B to A). 44
Figures 5A,B (Ref #2) 45
Parallel Imaging (Image Based Reconstruction) Image-based reconstruction parallel acquisition techniques: SENSE (SENSitivity Encoding). Figure 6 shows fast spin echo T2-weighted sagittal scan of the lumbar spine, without (A) and with (B) parallel imaging (using SENSE) In (B) every second Fourier line (parallel imaging with an IPAT factor of 2). Thus the scan time for (B) is half that of (A). Note that there are residual wrap around artifacts (arrow, B), a major drawback to the use of image-based reconstruction technique when anatomy is larger than FOV. 46
Figures 6A,B (Ref. #2) 47
Parallel Imaging (Drawbacks) Image-based reconstruction: If an aliasing artifact would be present in the chosen FOV for a non-parallel image sequence, then this aliasing will cause reconstruction problems if parallel imaging is attempted. K-space based reconstruction: The ability to construct effective sensitivities from the spatial sensitivities for each coil element depends on the sensitivity profile. This, in turn, depends on the coil element design; therefore, coil design is more critical with this technique. 48
GRAPPA (k-space) SENSE (image) 49
When Should You Use Parallel MR Imaging? To reduce total scan time To speed up single-shot MRI methods To reduce TE on long echo-train methods To mitigate susceptibility, chemical shift and other artifacts (may cause others) To decrease RF heating (SAR) by minimizing number of RF pulses
USE #1: Reduction of SAR in body imaging Case study: Breath-hold T2-weighted abdominal scans. Figure 4A: 17 second T2-weighted breath-hold acquisition, using 29 echoes Figure 4B: 17 second T2-weighted breath-hold acquisition, using 19 echoes. The missing Fourier lines for B were reconstructed using parallel imaging. Use of parallel imaging can be used to reduce the echo train length while keeping scan time the same (SAR reduction). Use of a shorter echo train more slices can be acquired within the same scan time, or overall time can be reduced. 51 Also, less T2-blurring and motion artifact.
Figures 4A,B (Ref #2) 52
Use #2: Reduction of T2-blurring Parallel imaging reduces T2-blurring because the readout time is shorter resolution is better. Augustin Me et al. Top Magn Reson Imag 2004; 15:207 53
Use #3: Reduction of Susceptibility Artifact (EPI) Parallel imaging reduces number of phase-encoding steps required per imaging time Top normal acquisition, Bottom R=2 acceleration
Other Current Uses Contrast enhanced MR (e.g., MRA) Improved spatial resolution for a given scan time. Cardiac MRI R=2 6 heartbeats R= 3 4 heartbeats R=4 3 heartbeats
(Not-so-distant) Future uses of parallel imaging: 2D acceleration 2D SENSE reconstruction (2X in L-R and 2X in A-P) from an 8channel head array coil conjugated gradient iterative solver after 10 iterations. http://www.nmr.mgh.harvard.edu/~fhlin/tool_sen