mc DESPOT at 3 T Jason Su Sept

mc. DESPOT at 3 T Jason Su Sept. 12, 2011

mc. DESPOT • Two component mapping technique, signal is due to a fast and slow relaxing pool • Vitally depends on accurate knowledge of flip angle • Maps produced: – T 1 and T 2 for each pool – Fast volume fraction (also MWF), amount of a voxel that’s taken up by the fast pool – Off-resonance (B 0 map) – Residence time, how long water stay in its pool

Current Developments • 2 mm isotropic protocols for 8 ch and Nova-32 ch – 4 -6 min • 1 mm isotropic protocol for Nova-32 ch – ~30 minutes • Key remaining decisions: – Bloch-Siegert pulse – Acceleration method and factor • “DISCOPOT” – view sharing with variable flip angle data

Protocol Goals • Applicability to various studies: MS, AD, thalamus segmentation, etc. – Ease of use – Need to maintain 20 x and 22 x versions of PSD • Future compatibility for 7 T • Robustness, especially with B 0 and B 1 inhomogeneity • Post-processing needs to be well packaged

Protocol Outline 1. 2. 3. 4. 5. 9 SPGR angles up to 18 deg. IR-SPGR for DESPOT 1 -HIFI (B 1 map) Bloch-Siegert B 1 map 9 SSFP angles with phase cycling up to 72 deg. Clinical/structural – – MPRAGE for WM/GM segmentation FLAIR for lesion segmentation DTI for tract-based analysis? Imports from the ADNI protocol? 6. AFI?

Hard Pulses • Bloch-Siegert B 1 maps do not account for excitation slab profile • Pulses that were used at 1. 5 T can no longer reach high enough angles at 3 T • Solution: hard pulses – No slab profile – Lower max RF amplitude allows higher angles – Enables shorter TR -> faster acquisition – Our acquisitions are almost always whole-brain anyway (the major selling point of mc. DESPOT)

Testing • Phantom – Agar Gel Sphere: only good for looking at parallel imaging comparisons – Half & Half Cream Cylinder: water and fat emulsion provides multi-component signal • Allows us to test the performance of our B 1 correction • Can we build one with B 1 inhomogeneity as severe as 7 T brain? • Fast volume fraction map has shown ability to estimate fat percentage • In vivo – Repeatability study – 2 scans sessions already performed on 2 volunteers, a variety of accelerations for each

B 1 Maps • Bloch-Siegert – Complex – Pulse width • 0. 5 ms pulse width: 56 deg. max, TR = 450 ms • 1 ms pulse width: 75 deg. , TR = 450 ms • 2 ms pulse width: 112 deg. max, TR = 500 ms • DESPOT 1 -HIFI with ASSET-2. 5 • Processing: – Maps cannot be registered – Bloch-Siegert map turned into a “kappa map” (flip angle scale factor) by normalizing by the mean B 1 over the brain

DESPOT 1 -HIFI

Bloch-Siegert: 2 ms

Bloch-Siegert: 1 ms

Bloch-Siegert: 0. 5 ms

Bloch-Siegert (2 ms) – DESPOT 1 -HIFI

Bloch-Siegert (2 ms) – DESPOT 1 -HIFI

Bloch-Siegert (1 ms) – DESPOT 1 -HIFI

Bloch-Siegert (0. 5 ms) – DESPOT 1 -HIFI

B-S (1 ms) – B-S (2 ms)

B-S (1 ms) – B-S (2 ms)

B-S (0. 5 ms) – B-S (2 ms)

B-S (0. 5 ms) – B-S (2 ms)

Comments • Normalization of B-S needs to be fixed • In general, I don’t trust HIFI very much but would rather see these corrections in H&H phantom • 0. 5 ms pulse width for B-S produces a L/R shifted homogeneity pattern compared to 1 ms and 2 ms

Parallel Imaging • Parallel imaging artifacts on mc. DESPOT maps? • How much acceleration can we use? • In vivo – – – ARC-2 x 2 (baseline) ARC-2. 5 x 2. 5 (reasonable) ARC-3 x 2 ASSET-2. 5 (reasonable) ASSET-4 (aggressive) • Images are registered to the ARC-2 x 2 baseline after proc. • Some attempts at a g-factor map but only based on collected data, should use fully sampled raw P-files – Want to determine slice and acquisition orientation effects

ARC-2 x 2: SPGR, 12 deg.

ARC-2. 5 x 2. 5: SPGR, 12 deg.

ARC-3 x 2: SPGR, 12 deg.

ASSET-2. 5: SPGR, 12 deg.

ASSET-4: SPGR, 12 deg.

ARC-2 x 2: T 1

ARC-2. 5 x 2. 5: T 1

ARC-2. 5 x 2. 5: T 1

ARC-3 x 2: T 1

ARC-3 x 2: T 1

ASSET-2. 5: T 1

ASSET-2. 5: T 1

ASSET-4: T 1

ASSET-4: T 1

Comments • Registration issues need to be re-evaluated – Ideally resample same k-space data • Higher ASSET acceleration biases T 1 upwards • Parallel imaging artifacts are subtle on T 1 maps – This is a bad thing, hard to figure out what voxels are corrupted

ARC-2 x 2: MWF

ARC-2. 5 x 2. 5: MWF

ARC-2. 5 x 2. 5: MWF

ARC-3 x 2: MWF

ARC-3 x 2: MWF

ASSET-2. 5: MWF

ASSET-2. 5: MWF

ASSET-4: MWF

ASSET-4: MWF

Comments • ASSET may skew MWF distribution downwards, no long symmetric error distribution • MWF is more sensitive than T 1 to recon changes • ARC-3 x 2 gives the closest match to maps produced by ARC-2 x 2

1. 5 T: MWF

3 T: ASSET-2. 5 HIFI MWF

MWF Difference

3 T: ARC-2 x 2 BS 1 ms MWF

MWF Difference

Comments • Values in the corpus callosum have changed the least • The peak in 0 s is due to included CSF which has very low MWF in both • Systematic overestimation of MWF at 3 T, this has been observed in mc. T 2 studies as well • A global bias correction for B 1 has been suggested at 1. 5 T (Aviv Mezer) • B 1 inhomogeneity is not severe in this example

1. 5 T: T 1

3 T: ASSET-2. 5 HIFI T 1

3 T: ARC-2 x 2 BS 1 ms T 1

Comments • This indicates that there’s something going very wrong • The T 1 values are shortening at 3 T, which is incorrect

DISCOPOT • View sharing of k-space between a sequence of angles • Fully sampled center of k-space, under sampled outer • Outer k-space pattern is pseudo-random but complementary with shared angles • Mixing scheme: – AB 1. *fa_{i} + B 2. *fa_{i-1} + B 3. *fa_{i+1} – Edge cases are slightly different • Tested on raw SPGR P-file data with fa 1 -13 • Many angles collected with the goal of mc. DESPOT in mind

DISCOPOT Sampling

Scanner Recon, Fully Sampled

offline. recon, Fully Sampled

offline. recon, Fully Sampled

offline. recon, DISCOPOT

offline. recon, DISCOPOT

Scanner Recon, Fully Sampled

offline. recon, DISCOPOT

offline. recon, DISCOPOT

DESPOT: T 1

DISCOPOT: T 1

DISCOPOT: T 1

DISCOPOT-Alternate: T 1

DISCOPOT-Alternate: T 1

Conclusions • B-S 1 ms seems like the best choice of the ones seen here • ARC 3 x 2 is promising • DISCOPOT view sharing is a viable way to accelerate the acquisition for DESPOT 1 – Further testing required for mc. DESPOT

Further Work • Collect fully sampled mc. DESPOT P-files – Create g-factor maps – Apply parallel imaging on the P-files to do a more direct comparison of each • Phantom creation to examine B 1 correction in fast volume fraction maps • DESPOT and compressed sensing