Sweep Imaging with Fourier Transform (SWIFT) in Breast Cancer

1

• 457
• Sweep Imaging with Fourier Transform (SWIFT) in
  Breast Cancer
• Curtis A. Corum, Andrew Babcock, Djaudat
  Idiyatullin,
• Angela L. Styczynski-Snyder, Diane Hutter,
• Lenore Everson, Michael Nelson, and Michael
  Garwood
• University of Minnesota, Minneapolis, MN, United
  States

2
Declaration of Relevant Financial Interests or
Relationships
Speaker Name Curtis A. Corum I have the
following relevant financial interest or
relationship to disclose with regard to the
subject matter of this presentation Dr. Corum
is entitled to sales royalties under an agreement
between the University of Minnesota and GE
Healthcare, which is developing products related
to the research described in this paper. The
University of Minnesota also has a royalty
interest in GE Healthcare. These relationships
have been reviewed and managed by the University
of Minnesota in accordance with its Conflict of
Interest policies.
3
Breast MRI

• While many MRI sequence types are sometimes
  indicated in Breast MRI the two main image sets
  usually desired are
• High spatial resolution pre and post-contrast T1
  weighted images (and subtractions) for
  morphological assessment (circumscribed vs
  spiculated, homogeneous vs heterogeneus
  enhancing, etc.)
• High temporal resolution dynamic contrast
  enhanced (DCE) T1 weighted image series with at
  least 1 min temporal resolution for contrast
  kinetics (uptake vs washout)
• Emerging standard of care utilizes semi and
  fully-quantitative pharmacokinetic modelling,
  with active research in improving models

4
SWIFT

• SWeep Imgaing with Fourier Transform
• Simultaneous interleaved excitation and
  acquisition
• 3D Radial Sampling (Halton sequence)
• PD or T1 weighted
• Smooth Gradient Update (Quiet) robust against
  motion, eddy currents, and system timing

5
SWIFT

• SWeep Imgaing with Fourier Transform
• Simultaneous interleaved excitation and
  acquisition
• 3D Radial Sampling (Halton sequence)
• PD or T1 weighted
• Smooth Gradient Update (Quiet) robust against
  motion, eddy currents, and system timing

6
SWIFT Timing
SWIFT has extremely short dead time On the order
of 2-6 µs Sensitive to fast relaxing
spins Preserves signal from off resonant spins
7
4 T SWIFT Breast Coils
SWIFT compatible Dual Breast Coil4 ch
Transmit/Receive, 4 TUMN Physics Machine Shop,
Peter NessCMRR Gregor Adriany, Carl SnyderNow
in imaging testing
Modified Single Breast Coils2 ch
Transmit/Receive, 4 TCMRR Carl SnyderHelmut
Merkle (now at NIH)Currently in use
8
Halton View Order
Pseudo Random 3d radial view-ordering Sorted for
smooth gradient transition Full sphere coverage
every 512 views Designed for View Sharing and CS
reconstruction
9
Goals

• Implement SWIFT based protocol for Breast MRI
• SWIFT compatible (no short T2 background from
  polymers, fast switching and/or ring-down times)
  transcieve coil(s)
• Demonstrate high temporal resolution SWIFT DCE
  imaging
• Demonstrate high spatial resolution morphological
  pre and post contrast imaging from same scan data
• Scan an initial cohort of patient volunteers

10
SWIFT Protocol

• 2 min shimming, pre-scan, scout
• 20 sec SWIFT pre-scans, phase reference and gain
• 1-2 min SWIFT FOV check, FS
• (2-4 min) (optional) Double Angle Method GRE B1
  map
• (2-4 min) (optional) SWIFT Variable Flip Angle T1
  map
• 2-6 min SWIFT DCE FS, pre-contrast (MagnavistTM
  0.1 mM/kg at 2 cc/s)
• 6 min SWIFT DCE FS post-contrast,
• (optional) further SWIFT test scans
• 11.33 min Minimum total time

11
4 T SWIFT Parameters

• TR 4.4 ms, 62 kHz, 4.1 ms HS1, Flip 8-16 deg, 256
  points
• Fat Suppression (FS)1/8 views, 4 ms Gauss, Flip
  120 deg, offset -625 Hz
• 3d Radial Isotropic Vieworder
• Sorted Halton sequence, 512 views per k-space
  sphere
• 128 full spheres per 4.5 min acquisition (6 min
  with FS)
• 65,536 views total before restarting
• Gridding based reconstruction
• Sliding window reconstruction for DCE, 6 sec
  frames
• 10 ms HS4 R20 pulse for dual fat and silicone
  suppression
• Wong TT, Sampling with Hammersley and Halton
  Points,J Graph Tools archive, Volume 2 , Issue
  2, 1997., Chan RW et al., MRM 2010.

12
Case FA
13
Case mass like DCIS
14
Case IDC
15
Ongoing Study...
We have now recruited 12 patients and have 8
successful sessions 3 of the incompletes were due
to last minute exclusions one due to scanner
failure
16
Conclusions

• SWIFT can produce high temporal resolution DCE
  and high resolution morphological data from the
  same scan data

Work in progress....

• Model based evaluation of DCE data
• Compressed Sensing reconstruction
• Case reviews and search for novel contrast (short
  T2)
• Continue recruiting patients....

17
Acknowlegdements
We gratefully acknowledge NIH R21 CA139688, P41
RR008079, S10 RR023730, S10 RR027290,and the
Minnesota Medical Foundation 3932-9227-09for
grant support. Thanks to physicians and
residents at the Fairview University Breast
Center and Jinjin Zhang for assistance with
patient studiesThanks to S. Suddarth and A.
Rath of Agilent, B. Hannah,J. Strupp, and P.
Anderson of CMRR for software and hardware
support. Thanks especially to Djaudat
Idiyatullin, Mike Garwood, Mike Tesch, and Ryan
Chamberlain (The rest of the SWIFT team) and
colleagues at the UMN CMRR!

18
NMR and Convolution
h(t)
spin impulse response
r(t)

system response

NMR and Convolution The fundamental basis of
SWIFT signal processing is that a frequency
modulated pulse alters the system response away
from the familiar hard pulse impulse response. In
the small flip angle limit the relationship is
convolution. Practically it works well up to 90.
19
SWIFT and Correlation
r(t)
system response
h(t)
spin impulse response

Recovering a standard FID by correlation SWIFT
produces an FID if the raw data (system
reposnse) is correlatied with the complex RF
pulse shape as a post processing step. In
practice this is performed in the frequency
domain by multiplication with the complex
conjugate of the complex pulse profile.