8/12/2011
Outline
► Past
Developing 4D-MRI
For 4D Radiation Therapy
- MRI for motion imaging
- 4D-MRI strategies
► Current
- Sequences
- Surrogates
Jing Cai, PhD
Department of Radiation Oncology
Duke University Medical Center, Durham NC
► Future
- Clinical implementation
- Future directions
Disclosure: No conflict of interest
Pros and cons of MRI
►
Pros
- Superior soft-tissue contrast to CT
- No risk of radiation exposure (long time imaging)
- Flexible in image plane selection
- Functional/molecular imaging
- Variety of image contrasts
►
Cons
- Poor spatial accuracy (image distortion)
- Various image artifacts (ghost, susceptibility)
- Signal not correlated to electron density
MRI for Motion Imaging
► Sites:
lung, esophagus, liver, spinal cord, H&N,
pancreas, etc.
► Tumor motion ~ location/size/type of cancer, etc.
► Correlation: external motion ~ tumor motion
► Statistical tumor motion (PDF)
► 4D-CT pitfalls
► Lung deformation
► 4D tumor motion in patients (hemi-diaphragmatic
paralysis)
1
8/12/2011
Tumor Motion Probability
obj  (
V tum or
V lung
2
) w

i , j , k  lung
d
2
i , j ,k

1
2
pm (

Tumor Motion PDF Evolution
( d i , j , k  Pi , j , k ) ) m (1  sgn( d i , j , k  Pi , j , k ))
2
i , j , k C T V m
► Large
► Tends
Simulate 4D-CT using MRI
1st couch
2nd couch
variation during the scan
to stabilize after certain time
MRI
‘4DCT’
3rd couch
……
Image Acquisition
Couch Movement
Respiratory signals from internal surrogate
2
8/12/2011
4DCT MIP
underestimates
tumor ITV
70
60
y = 91.703x
R2 = 0.7747
3cm
5cm
ITV Error (%)
70
Linear (1cm)
60
Linear (5cm)
y = 55.104x
R2 = 0.8469
40
30
y = 36.334x
R2 = 0.8523
20
50
40
y = 45.917x - 0.4786
R2 = 0.7558
30
20
10
10
0
0.00
4D-CT AIP: inaccurate probability?
80
1cm
Linear (3cm)
50
‘4DCT’
ITV Error (%)
80
MRI
0.20
0.40
Breathing Varibility
0.60
0.80
0
0.00
0.50
1.00
vR/S
1.50
2.00
‘4DCT’ MRI
4D-CT AIP: Patients
‘4DCT’ MRI
‘4DCT’
MRI
Spinal Cord Motion
►4
frames/sec, 20 sec continuous
motion generally < 0.5 mm
► Cord
3
8/12/2011
Brain Pulsatile Motion
CC
►
►
►
Lung Deformation using HP Gas MRI
AP
DENSE sequence
Pulsatile motion originated from brain stem and radiates
toward peripheral brain regions.
CSF can be easily identified due to its opposite movement
2D Dynamic Tagged Lung: Sagittal
Strategies of 4D-MRI
► Real
0.0s
0.7s
1.4s
2.1s
2.8s
time 3D
- ultra-fast 3D MR sequence
- fast gradient, multi-channel coils
- parallel processing
- current: voxel 3-4mm, 1.5 sec/frame
► Retrospective-sorted
2D
- fast 2D MR sequence
- respiratory signals (external, internal, etc.)
4
8/12/2011
Retrospective 4D-MRI
Potential MR Sequences
►
TrueFISP/FIESTA (balanced steady state gradient echo)
Aim: simple, robust, quickly implementable
- T2*/T1, sensitive to fluid, band artifacts from long TR
►
Image Acquisition
►
►
►
►
►
►
Fast 2D cine MR
Multiple slices
Cine duration > 1 cycle
Frame rate: ~3 frames/sec
Slice thickness: 3-5 mm
In-plane pixel size: 1-2 mm
Respiratory Signal
HASTE/SSFSE (single shot fast spin echo)
- T2, good CNR, signal decay from lung echo train, blurring
►
Surrogates
- External
- Internal
- Image-based
► Signal processing
► Phase determination
►
FLASH/Fast SPGR (fast spoiled gradient echo)
- T1 (poor), tumor hypo-intensity
►
EPI (echo-planner imaging)
- GE-EPI (T2*), SE-EPI (T2), IR-EPI (T1)
- susceptibility, ghosting, chemical shift, fat suppression
Examples: 2D cine-MRI
HASTE v.s. TrueFISP
►
►
HASTE: visualize parenchyma better, tumor blurred
TrueFISP: visualize vascular better, motion artifact
5
8/12/2011
Dependence on Cancer Type
CNR reduction in TrueFISP relative to HASTE
Percentage (%) .
100
80
Summary
► HASTE
and TrueFISP both can monitor lung
tumor motion during free breathing.
► Tumor
conspicuity and image artifacts depend
upon tumor characterizations.
60
40
► HASTE
images show better tumor conspicuity
than TrueFISP images.
20
0
Adenocarcinoma
Squamous Cell
► HASTE
►
Heterogeneous structure of squamous cell carcinoma may
reduce its conspicuity in TrueFISP images.
Surrogates: external
Surface
Imaging
Surrogates: internal/image-based
Belt
RPM
Spirometer
has local blurring artifact; TrueFISP has
motion artifacts in the phase encoding direction.
•
•
•
•
•
•
•
•
•
Implanted markers
Diaphragm
Air content
Lung density
Implanted Makers
Lung area
Body area (axial, sagittal)
Normalized cross correlation
Deformable image registration
Fourier transform (magnitude, phase)
6
8/12/2011
External surrogate: PMU
4D-MRI with PMU
Slice 3/10
Slice 5/10
Slice 7/10
Amplitude (A.U.)
•TrueFISP in lung volunteer
(dark blood pulse on)
•Average
frames/bin/slice = 17
Acquire external respiratory signal
(Physiological Monitoring Unit -Time (seconds)
PMU)
PMU logging:
• synchronized with the image acquisition computer, auto started/stopped
within sequence run sampled at 50 Hz
Erik Tryggestad, et al. 2011 Joint AAPM/COMP Meeting
Internal surrogate: diaphragm
Slice 2/9
Slice 4/9
Slice 6/9
•HASTE in abdo volunteer
•Average
frames/bin/slice = 26
Erik Tryggestad, et al. 2011 Joint
AAPM/COMP Meeting
Image-based surrogate: Body Area
von Siebenthal, et al., "4D MR imaging of respiratory organ motion and its
variability," Phys Med Biol 52, 1547-1564 (2007).
7
8/12/2011
Validation of BA as Surrogate
Axial
BA
Pre-sorted 4DCT Images (n=31)
Breathing signal
from BA
Breathing signal
from RPM
4DCTBA
Sagittal
BA
Marker position (P)
Breathing period (T)
Breathing amplitude (A)
Period variability (VT)
Amplitude variability (VA)
Space-dependent phase shift (F)
4DCTRPM
Phase Comparison
( R, D, DA )
Correlations
Image Quality Comparison
( SBA, SRPM )
4DCT: BA v.s. RPM
Signal and phase: BA v.s. RPM
4DCTBA
4DCTRPM
8
8/12/2011
Summary of measurements
Patient
P
T (s)
VT
A (mm)
VA
R
D
DA
F (s)
SBA
SRPM
6.5
0.20
0.90
-5.1
13.8
0.47
3.1
2.6
6.8
0.21
0.94
-1.3
8.5
0.32
2.8
2.6
Image quality evaluation
Lung Cancer Patients
Mean
L2
3.4
0.18
Abdominal Cancer Patients
Mean
L2
3.7
0.19
All Patients
Mean
L2
3.6
0.19
6.6
0.20
0.92
-3.3
11.4
0.40
2.9
2.6
p*
0.61
0.34
0.50
0.78
0.52
0.04
0.28
0.001
0.03
0.23
0.92
Significant differences in R, DA, and F between the two
groups of patients.
Image-based surrogate: FFT
FFT surrogate: patient example
FFT
9
8/12/2011
Validation of FFT surrogate
Phantom Study
Bolus
Fixed
Spring
Motor
Gel
►
►
►
►
►
10 Subjects, 2 min MRI scan, sagittal / coronal
Respiratory signal: ROI tracking v.s. FFT phase angle
Small phase difference (-3.13±4.85%), high correlation
(r2=0.97±0.02 )
Phantom: axial BA
Axial
Coronal
4D-MRI
Sagittal
►
►
►
MRI-compatible phantom driven by a sinusoidal pattern
Mimics tumor and body motion
1.5T GE clinical scanner
FIESTA: ~ 3 frames/sec, 6 sec/slice
BA is the area under bolus
Phantom: sagittal FFT
Sagittal
Single slice cine-MRI
Axial
Coronal
Saggittal
10
8/12/2011
XCAT:
axial BA
4DCT
XCAT:
sagittal BA
Clinical Implementation
4DMRI
End points:
Single slice cine MRI
• Tumor-to-tissue CNR (4DMRI v.s. 4DCT)
• Tumor ITV accuracy (4DMRI v.s. cine MRI)
• Dosimetric impact (4DMRI v.s. 4DCT)
11
8/12/2011
Future Directions
►
Study tumor CNR v.s. cancer characteristics (sequences)
►
Optimize imaging parameters (CNR, irregularity)
►
Improve respiratory surrogate accuracy and robustness
►
Use contrast: SPIO (super-paramagnetic iron oxide, CNR)
►
Real 4D (sequence programming, hardware)
►
Functional 4DMRI (ventilation, perfusion)
►
Clinical implementations
MIP imaging without sorting
►
►
Goal: to develop a simple MRI technique to generate MIP
for treatment planning in radiation therapy
Why: many cases only need MIP, no need of individual
phases (respiratory-gated RT)
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
2D MIP
- Immediate needs
S. A. Schmitz, et al. Iron-Oxide-Enhanced MRI of
the Liver, ACTA RADIOLOGICA, 2006; 634-642
Phantom Study
3D MIP
Preliminary Phantom Results
phantom
motion platform
string
SS-MRI
string
Cine-MRI
4DCT
AP = 7.6 cm
RL = 2.1 cm
CT sagittal slice
Sagittal cine-MRI
MIP
weight
V = 66cm3
SI = 8.0 cm
12
8/12/2011
Acknowledgements
Summary of Phantom Results
Duke University
Volume of Phantom in 3D MIP (cm 3)
Area of Phantom in Sagittal 2D MIP (cm 2)
Fang-Fang Yin, PhD
Zhiheng Wang, PhD
Brain Czito, MD
Irina Vergalasova, BS
Jim Chang, PhD
Paul Segars, PhD
Chris Kelsey, MD
Raj Panta, BS
diff from
4DCT (%)
SS-MRI
4DCT
Cine-MRI
diff from
4DCT (%)
diff from
cine-MRI (%)
66.0
2.3%
45.6
47.4
44.2
-3.7%
3.3%
98.6
-4.2%
46.1
48.4
45.3
-4.8%
1.9%
University of Virginia
94.1
95.0
-1.0%
43.9
45.6
43.7
-3.7%
0.5%
patient 1
87.2
88.9
-2.0%
39.7
40.7
43.8
-2.5%
-9.4%
patient 2
82.3
78.3
5.0%
41.5
41.1
40.8
0.8%
1.7%
patient 3
82.3
81.1
1.5%
29.6
31.6
29.2
-6.3%
1.5%
Stanley Benedict, PhD James Larner, MD
Paul Read, MD, PhD
David Schlesinger, PhD
Talissa Altes, MD
James Brookeman, PhD
G. Wilson Miller, PhD
John Mugler III, PhD
patient 4
82.1
88.1
-6.8%
46.5
45.3
47.2
2.6%
-1.5%
UCLA
patient 5
80.3
85.7
-6.4%
47.5
44.8
46.2
6.0%
2.8%
Ke Sheng, PhD
patient 6
77.2
82.6
-6.6%
45.9
43.7
45.7
5.2%
0.5%
patient 7
80.1
86.1
-6.9%
45.7
42.9
44.0
6.5%
3.9%
patient 8
80.2
88.3
-9.2%
46.2
44.1
44.7
4.6%
3.3%
trajectory
SS-MRI
4DCT
static
67.5
sin
94.5
(1-cos)2
Siemens
Xiaodong Zhong, PhD
Johns Hopkins University
Erik Tryggestad, PhD
Research was supported by NIH grant R01-HL079077, grant IN2002-01,
Siemens Medical Solutions, and University of Virginia Cancer Center
13