1. Burst: Applications in Ultra-Rapid Imaging and Quantitative Diffusion Measurement Simon J Doran Department of Physics, University of Surrey S Dr. S. J. Doran Department of Physics, University of Surrey, Guildford, GU2 5XH, UK

2. Acknowledgements Marc Bourgeois (ICR) Claudia Domenig (UniS) Andy Dzik-Jurasz (ICR) Martin Leach (ICR) David Collins (ICR) Claudia Wheeler-Kingshott (IoN) Roger Ordidge (UCL)

3. Summary Introduction to Burst and background  Basic concept  Historical survey Single-shot Burst imaging  Burst variants, SNR comparison and choice made  Problems to overcome  Comparison of techniques in phantoms and in vivo Quantitative diffusion imaging extra-cranially  Application of Burst for diffusion measurements  Early results and analysis  Comparison of Burst and other techniques at 1.5 T

4. Introduction: Basic concept Burst is a rapid imaging technique, first proposed by Hennig in 1988. A series of low angle pulses creates a train of echos, which can be used to form an image. Burst pulse train (64 low-angle pulses) 180 o Train of 64 echos G read G phase

5. Advantages of Burst Using a slice-selected Burst sequence, all the signals can come from pure spin-echoes.  Little geometric distortion or “signal drop-out” in regions of large susceptibility change  Better off-resonance properties than EPI — no need for fat-sat Less rapid gradient switching than EPI  dB/dt issues not a problem from a safety point of view  Can be acoustically very quiet Lower RF power deposition than HASTE Extremely robust  no shimming or set-up period required dB

6. Disadvantages of Burst Low signal-to-noise  Intrinsically low SNR due to low flip angle pulses  (Relatively) high acquisition bandwidth (e.g., 10  s / point), but still much lower than EPI at 4.7 T Signal Decay  Diffusion and T 2 during both excitation and read periods  This is both a disadvantage (for single-shot imaging) and a valuable feature (for diffusion imaging)

7. Literature survey: (1) Basic sequence This type of pulse sequence has been extensively studied in a non-imaging context.  Kaiser, Bartholdi and Ernst, J. Chem. Phys. 60, 2966 (1974)  Hennig et al. MRM, 3, 823 (1986) First images published in 1993 by Hennig and Hodapp (MAGMA, 1, 39-48) and Lowe and Wysong (JMR 101, 106) Burst is a variant on the DANTE sequence  spectroscopy: Morris and Freeman, JMR 29, 433 (1978)  cardiac tagging: Mosher and Smith, MRM 15, 334 (1990) Data: McVeigh and Atalar

8. Literature survey: (2) Linear approximation Easiest theoretical treatment by assuming linear approximation, i,e., each pulse causes one echo. However, this works only for very low pulse angles. In practice, the Bloch equations are non-linear and higher order echoes occur. Interference between spin and stimulated echoes reduces the echo amplitudes. Simulations from Zha and Lowe, MRM 33, 377 (1995)

9. Literature survey: (3) Phase modulation We can also look at the problem in the frequency domain. We get a small signal because only a small fraction of the sample is excited. One pixel Zha and Lowe (MRM 33, 377 (1995)) showed that by suitable phase- modulation of the low-angle pulses, one can excite the sample almost completely and obtain the desired echo train. One pixel

10. Literature survey: (4) Optimisation Several authors have considered optimisation of the Burst excitation pulse train.  Le Roux et al. Chirp pulses Proc. 10th SMRM, 238 (1991)  Zha and Lowe, OUFIS, MRM 33, 377 (1995)  van Gelderen et al. JMR B, 107, 78 (1995)  Heid, MRM 38, 585 (1997) The bottom line is that for N excitation pulses, i.e., N echoes, the pulse flip angle should be at most    2  N as opposed to  / 2N for the non-optimised pulse train. For 64 pulses, this equates to  11.25°  still poor SNR

11. Literature survey: (5) Burst variants Burst can be seen as “simply” a means of generating multiple echoes. As such it can be incorporated into many standard sequences.  Radial imaging: Jakob et al., 36, 557 (1996)  SSFP: Heid, Proc. 8th ISMRM, 1499 (2000)  STEAM: Cremillieux et al. MRM 38, 645 (1997) (6  64  64 images in 210 ms) Burst pulse train (16 low-angle pulses) 180 o Train of 16 echos G read G phas e  4  HASTE (BASE): van Gelderen et al. MRM 33, 439 (1995); Zha et al. Proc 5th ISMRM, 1820 (1997) Burst pulse train (9 low-angle pulses) Multiple trains of 9 echos G read RF  N / 18  EPI (URGE-EVI): Heid, Proc. 3rd ISMRM, 98 (1995) Burst pulse train (9 low-angle pulses) Train of 9 echos G read RF N / 9  FLASH (URGE): Heid et al., MRM 33, 143 (1995)

12. Multi-refocusing Burst: (1) Excitation The original design of Burst sequence has two major problems with its excitation scheme:  The entire sample is excited by the train of hard pulses, so multi- slice acquisitions are not possible. Image profile Theoretical profile from  -pulse frequ- ency response Pixel Number Signal intensity / arb. units  Although overall RF energy deposition is relatively small, the peak power required is excessive, because it needs to be applied as a short pulse.

13. Multi-refocusing Burst: (2) Selective excitation A solution to both problems is found by using selective excitation (van Gelderen et al. MRM 33, 439 (1995)) However, this removes several of the key advantages of Burst. Now the sequence becomes noisy, is highly demanding on the gradients and we get some artifacts. G read G slice  N N RF G read G slice  N / 2 RF Unipolar schemeBipolar scheme RF frequency offset inverted

14. Multi-refocusing Burst: (3) Excitation artifacts In the presence of B 0 -inhomogeneities, the bipolar scheme gives rise to slice-definition inconsistencies. Not noticeably a problem. However, we do see significant differences in echo phase. Echo number Echo phase / rad Echo number Echo phase / rad Slice offset = 0 Slice offset = 60 mm Raw Corrected

15. Multi-refocusing Burst: (4) Echo phase The most significant problem in developing the multi-refocusing Burst sequence at 1.5 T on the Siemens Vision is the unwanted variation in echo phase. A standard FT reconstruction algorithm assumes that, in the absence of the phase-encoding, all echoes have the same phase. In fact, we observe the phase to change in the following ways:  continuously during an echo train  discontinuously between echo trains  alternating when we use the bipolar slice selection  with an amplitude of variation that depends on the slice offset from isocentre The cause of these phase variations is still uncertain, but may be an eddy current effect.

16. Multi-refocusing Burst: (6) Echo phase examples Continuous phase change during a single readout of 64 echoes Alternating phase during a single readout of 64 echoes (bipolar slice gradient), small slice offset Alternating phase during a single readout of 64 echoes (bipolar slice gradient), large slice offset (NB Phase needs unwrapping!) Discontinuous phase change during a multi- refocusing readout of 12  6 echoes

17. Multi-refocusing Burst: (6) Echo phase artifact Uncorrected, the echo phase problem gives rise to a serious artifact. With suitable correction, using a non-phase encoded echo train, the artifact can be mostly removed, but the remaining artifacts still degrade the performance of the sequence. The major unsolved problem is to achieve the correction in regions of the body that move between the non-phase encoded scan and the “image” scan.

18. Early results at 1.5 T Comparison of original OUFIS with “off-the-shelf” EPI on Siemens Vision. Slice deliberately chosen to highlight problems with EPI. Poor resolution and SNR, but excellent geometric fidelity, particularly around air spaces Note the difference in contrast. 64-pulse OUFIS 128  128 EPI

19. Very (!) early results at 4.7 T Results after approximately two days on 4.7 T system in factory environment Single-shot 64 2 image (partial Fourier, reconstructed to 64  112) acquired at 4.7 T Poor resolution and SNR, but excellent geometric fidelity, particularly around air spaces Note: no need to shim Compare the EPI acquired at the same time (shimmed to get best results on top slice).

20. Recent comparison at 1.5 T “Original” Burst (OUFIS) 100 ms, 3.6  3.6 mm 2, TE eff ~10 ms, SNR N =2.3, SL=7mm Refocussed Burst 238 ms, 1.8  1.8 mm 2, TE eff ~25 ms, SNR N =8.3, SL=7 mm EPI 248 ms, 1.8  1.8 mm 2, TE eff ~90 ms, SNR N =27, SL=7 mm HASTE 344 ms, 1.25  1.25 mm 2, TE eff =?, SNR N =48, SL=7 mm Refocused Burst“Original” BurstEPI HASTE SNR N = SNR / ( (acq. time) 1/2. (pixel area) )

21. Extra-cranial imaging at 1.5 T: pelvis HASTE EPI + “fat sat” “New” Burst Pelvic imaging is important for diagnosing rectal and prostate cancers. HASTE is currently the method of choice for single-shot imaging, but RF power deposition is a potential problem. EPI is not widely used because of the presence of fat. Burst works “quite well”.

22. SNR: Comparison with EPI Ignoring artifacts, the key relationship is SNR  sin  / BW 1/2. On the Siemens Vision at 1.5 T, we have shown that the SNR of EPI is approximately a factor of 3 higher than our best Burst. At higher field, a spin-echo based Burst sequence could be read out at the same BW, whereas the EPI sequence would be likely to require a much higher bandwidth.

23. Image contrast (1) What sort of information can we get out of Burst images? Contrast properties of Burst images very little studied so far. Sequence is inherently T 2 and D weighted. For low flip angles By adjusting TE and the read gradient, we can emphasise either T 2 or diffusion decay.

24. 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0510152025303540 Echo Number A / A0 Data for CuSO 4 T 2 and D double fit Typical Spectroscopic Data Typical Image Data Can we use the decay to get D and T 2 ?

25. ADC and T 2 Burst sequence A data array of n echoes is acquired for each PE step. Each echo, j, corresponds to the same k-space line of the same slice, but with a different ADC and T 2 weighting. Corresponding echoes in successive arrays are used to reconstruct a given image. In one scan we collect n images weighted by ADC and T 2. Phase encode Readout D, T 2 j = 0 j = n-1

26. Burst images SE images ME images First in vivo images (8 T)

27. Quality of double-exponential data fit Echo number A j / A 0 ROI1 ROI2 ROI3 Typical single pixel fit

28. First human study (1.5 T) Multi-functional rectal carcinoma study Images had very poor SNR, so analysis performed on ROIs in tumour. Remarkable correlation between tumour ADC and treatment success.

29. But are we really measuring diffusion? The data fit moderately well to a bi-exponential function. There are several possible explanations:  genuine IVIM perfusion effect  incorrect T 2 correction  motion  many different ADC values in the ROI

30. Burst single-shot imaging: Conclusions Burst has been around for a “long” time (10 years), but has never really caught on. It has a number of attractive features, most notably that it can be made almost impervious to susceptibility, giving undistorted images. The SNR has been improved by a factor of approximately 30 since the original introduction of the sequence, but is still quite low. The contrast of the sequence needs investigating further. Our 3-year EPSRC project came to the conclusion that Burst is “almost competitive”, but not quite on the hardware we used. Application at higher fields remains an attractive possibility.

31. Measurement of diffusion with Burst: Conclusions Burst gives us a potentially exceedingly time-efficient way of obtaining many b-values in the same measurement. The SNR in the original measurements was low, but we have researched a number of ways of improving this. There are still a number of technical difficulties with the approach, the most serious of which is motion. This makes it as yet unclear whether the values we are getting from Burst are correct or not. We have performed extensive phantom and initial in vivo comparisons with three other diffusion imaging sequences: split-echo HASTE, PSIF and segmented EPI.

32. Choice of Burst sequence to develop Aim: medium resolution, single-shot, multi-slice dataset Choice made on basis of expected SNR. Chosen sequence “Original” Burst

33. Multi-refocusing Burst: (4) Unwanted echoes The need for crusher gradients around the 180° pulse can be understood by the use of an extended phase graph.  180 ... Everywhere that magnetisation crosses the central axis, an echo is formed. (Not all paths from the original  - pulses are shown.) Higher order echoes are superimposed on the desired spin echoes.

34. Multi-refocusing Burst: (5) Unwanted echoes We can separate out the unwanted echoes by changing the gap between the last  - pulse and the first 180°.  180 ... The unwanted echoes are small, but can be significant. A very sensitive test of the efficiency of spoiling is to acquire a non- phase-encoded dataset and FT all the echoes. Depending on where the unwanted echoes occur, the effect on the image may be slight or extremely serious.

35. Multi-refocusing Burst: (7) Segment offset artifacts The current implementation of the sequence uses mosaic tiling of the k-space segments. Eddy currents and poor performance of the Vision gradient system lead to offsets in the phase-encoding blip gradient of small fractions of  k phase. These again lead to complicated multiple ghosting artifacts in the phase-encoding direction. The process can be simulated and, in principle, corrected. Segment 1 Segment 2 Segment 4 Segment 3 kxkx kyky Segment 1 Segment 2 Segment 4 Segment 3 kxkx kyky

36. Phantom comparison at 1.5 T SNR=19, SL=10mmSNR=72, SL=10mmSNR=75, SL=5mm “Original” Burst“New” Burst“Best” EPI

37. Image contrast (2) Do we need T 2 contrast to see the activations?  E.g., Hutchinson et al. JMRI 7(2), 361-364 (1997) Potential method for getting T 2 contrast … Burst pulse train (64 low-angle pulses) 180 o Train of 64 echos G read G phase T 2 delay In practice, this works with the basic Burst sequence, but we have not had any success in achieving T2* weighting with multiply refocused Burst.