IMPROVED MRI TEMPERATURE IMAGING USING A SUBJECT-SPECIFIC BIOPHYSICAL MODEL

1
IMPROVED MRI TEMPERATURE IMAGING USING A
SUBJECT-SPECIFIC BIOPHYSICAL MODEL

• Nick Todd, Allison Payne,
• Douglas A. Christensen, Henrik Odeen, Dennis L.
  Parker
• Utah Center for Advanced Imaging Research,
  University of Utah

2
Background

• Utah Projects in MRI guided HIFU
• Large animal MRgHIFU system (Siemens/IGT)
• Small animal MRgHIFU system (IGT)
• Breast MRgHIFU system
• (UofU/IGT/Siemens)
• See poster 4.8 by
• Allison Payne

3
Background Utah Projects

• MR guided HIFU
• Breast
• Develop the Utah Breast MRgHIFU system
• Brain
• Develop 3D MRI Temperature measurements for MRI
  guided Brain HIFU
• Temperature measurement requirements
• Breast
• glandular tissues AND fat
• Near-field protection
• Brain
• cover entire skull volume
• high temporal and spatial resolution

4
MR Temperature Basics
Proton Resonance Frequency Shift (PRF).
MR signal frequency depends on local chemical
environment of water Hydrogen. Temperature
changes affect this environment.
Current Time Frame
Reference
Difference
Temperature Map
Frequency changes measured as image phase changes.
-

5
Breast Temperature Measurements
3-point Dixon Images

• Requirements
• Control treatment in glandular tissue
• Avoid fat necrosis
• Coverage, speed, and resolution
• Temperature in water and fat?
• Hybrid PRF/T1 method
• 2D GRE
• 2D/3D Segmented EPI

Fat
Water
fat
agar
6
MRI Thermometry - Breast
Hybrid PRF/T1
Signal from Spoiled GRE sequence
Image sequence 2 alternating flip angles PRF
from phase of each image T1 from two images
Deoni, Rutt, Peters. Magn Reson Med 2003
49515-526.
7
Breast Temperature Measurements
A) 3-Pt Dixon Water Image
B) 3-Pt Dixon Fat Image
C) PRF/T1 Magnitude Image
D) PRF Temperature Map
Pork Muscle
Breast Fat
Targeted Area
Transducer
8
Breast Temperature Measurements
PRF Temperatures in Pork
T1 Percent Change in Breast Fat
B)
C) PRF/T1 Magnitude Image
D) PRF Temperature Map
Pork Muscle
Breast Fat
Targeted Area
Transducer
9

• Transcranial
• MRI guided HIFU

• Funding
• Focused Ultrasound Surgery Foundation
• NIH R01 EB013433

10
Transcranial MRI guided HIFU

• Cover all heated regions Skull within
• Resolution Speed Coverage (FOV)
• 1mm isotropic 1s Full head/breast
• 205x160x100 TR35, ETL7 80s
• 205x160x33 (1x1x3mm), TR35, ETL7 27s

11
Required Values
1 x 1 x 3 mm
Spatial Resolution Temporal Resolution Volume
Coverage Signal - to - Noise
2 seconds per image
256 x 162 x 72 mm
Brain
Image Volume 256 x 162 x 72 mm
Image Volume
12
Transcranial MRI guided HIFU

• How to go faster
• 2D Spatially selective RF excitation
• Prefer full FOV
• Parallel imaging UNFOLD1
• Temporally Constrained Reconstruction (TCR)2
• Model Predictive Filtering (MPF)3
• 1 Chang-Sheng Mei, et al. Magnetic Resonance in
  Medicine 66112122 (2011)
• 2 N. Todd et al. Magn Reson Med 62(2)406-419
  (2009).
• 3 N. Todd, A. Payne, D. L. Parker, Magn Reson
  Med 6312691279 (2010)

13
Data Acquisition Reconstruction
F Fourier Transform m Image Estimate d
Undersampled Data

Data Space (k-space)
Image Space
Inverse Fourier Transform
256 x 162 x 24 pixels
256 x 162 x 24 pixels
14
Constrained Reconstruction
F Fourier Transform m Image Estimate d
Undersampled Data Gradient in time

is iteratively updated subject to
constraints
Image must match acquired data
Image must change smoothly in time
iteration 5
iteration 25
iteration 50
iteration 100
15
TCR Constrained Reconstruction

• Sequence Parameters
• 1.5 x 2 x 3 mm
• 288 x 216 x 108 mm
• 192 x 108 x 36 matrix
• EPI Factor 7 lines per excitation
• TR/TE 35 / 9 ms

Data Undersampling
ky
kz
Constrained Reconstruction
Scan Time 1.8 s / time frame 25 s / full data set
Not real time
16
Constrained Reconstruction Results
Validation Tests Truth Full Data used 1.5
x 1.5 x 3.0 mm 2.8 seconds per image 288 x 162 x
24 mm Test Cases 288 x 162 x 48 mm 288 x 162 x
90 mm 288 x 162 x 144 mm
2.8 s
Full Data
Truth
5.4 s
10.1 s
16.2 s
Constrained Reconstruction 6X data reduction
2.8 s
Truth
0.9 s
1.7 s
2.7 s
17
Model Predictive Filtering
Thermal Model
Artifact-free Temperature maps
Undersampled k-space
Goal real time
N. Todd, A. Payne, D. L. Parker, MRM 6312691279
(2010)
18
Model-Predictive Filtering

• Segment tissues
• Determine tissue-specific thermal and acoustic
  properties
• TCR Modeling
• Use tissue-specific properties in dynamic MPF
  temperature measurements
• Realtime, 3D, large FOV
• From highly undersampled 3D segmented EPI PRF

19
Tissue Segmentation

• Breast tissue segmentation
• Hierarchal Support Vector Machine algorithm

3pt Dixon H2O only
FS PD-w
3pt Dixon Fat only
Non-FS T1
FS T2-w
h-SVM w/ Zero-Filled-Interpolation
20
Tissue property estimationAcoustic parameters
Segment treatment volume into a small number of
tissue types
4-8 low power pulses cover targeted volume
TCR reconstruct temperature images
In-vivo estimates of the change in the
attenuation coefficient with log10 of thermal
dose using the iterative parameter estimation
technique . Urvi Vyas et al. ISTU 2011
MR temps to get SAR patterns
Use ultrasound model (HAS) to determine
absorption and speed of sound to match measured
pattern
Tissue acoustic values for Model Predictive
Filtering.
21
Tissue property estimationThermal parameters
Segment treatment volume into a small number of
tissue types
4-8 low power pulses cover targeted volume
TCR reconstruct temperature images
MRI temps during cooling
Determine thermal diffusivity using cooling
temperature curves
Cheng et al., JMRI 16(5), 2002
22
Hybrid Angular Spectrum (HAS) Pressure Modeling
23
HAS SAR prediction
24
HAS Head Model
Courtesy Guido Gerig, University of Utah
25
(No Transcript)
26
(No Transcript)
27
Model Predictive Filtering
Multi-step, recursive algorithm
3
2
Phase (n1)
1
3
Temp (n1, model)
Temp (n)
K-space (n1)
5
Magnitude (n)
Step 1 Use model to predict temperature at time
n1. Step 2 Convert temperature map to phase map
for time n1. Step 3 Use this phase and the
magnitude from time n to create k-space for time
n1. Step 4 Insert any actually acquired k-space
lines. Step 5 Recalculate the temperature for
time n1 using the data updated k-space.
Temp (n1, model and data)
28
Model Predictive Filtering
Use the Pennes Bioheat Equation, tissue
properties, and a pre-treatment heating to
determine the thermal model.
Full Data
Model Only
T temperature r density C tissue and blood
heat capacity k thermal conductivity Wb blood
perfusion Q heat applied
29
2-D MPF Results
Fully sampled k-space data sets 288x288x20mm
FOV, 2.3x2.3x4mm res, 8.3 sec/scan.
25 of k-space used in reconstruction.
Power 36W (Model Id data set)
Mean and STD of error over an ROI
MPF
Power 42W
Mean and STD of error over an ROI
MPF
Power 48W
Mean and STD of error over an ROI
MPF
30
3D (R12) vs 2D (R1) MPF Temperatures
Common Ultrasound pulse 36 W/58.1 sec 3-D
GRE FOV 256x256x32 mm3, Matrix 128x128x16
Resolution 2.0x2.0x2.0 mm3 TR/TE 25/8
ms Tacq 76.8 s/image volume (R1) 6.4
s/image volume (R12.1) 2D GRE FOV 256x256x20
mm (sl 3mm) Matrix 128x128 Resolution
2.0x2.0x3.0 mm3 TR/TE 65/8 ms 8.3 sec per
scan (R1) Scans repeated 8x for variability
N. Todd, A. Payne, D. L. Parker, MRM 6312691279
(2010)
31
Model Predictive Filtering Results
Phantom Heating 2.0 x 2.0 x 2.0 mm 0.5 seconds
per image 256 x 162 x 48 mm sT lt 1C
Transverse
Sagital
Coronal
32
Summary Work in Progress

• Brain requires
• Large FOV Cover insonified volume
• High speed 1s/volume
• High resolution lt 1 x 1 x 3 mm3
• Our solutions
• PRF Highly undersampled (gt8) 3D segmented EPI
• TCR
• Does not require tissue thermal and acoustic
  properties
• Achieves high spatial and temporal resolution,
  large FOV, LOW NOISE!
• Cannot (yet) be performed in real time
• Model-predictive Filtering (MPF)
• Requires
• tissue segmentation
• estimate of tissue acoustic and thermal
  properties
• Property estimates
• SAR Hybrid Angular Spectrum (HAS)
• Diffusivity/Perfusion MRI during cooling
• Also achieves high spatial and temporal
  resolution, large FOV, LOW NOISE!
• Potential real time application

33
Acknowledgments
Thank You
People Dennis Parker Bob Roemer Doug
Christensen Leigh Neumayer Allison Payne Nick
Todd Rock Hadley Nelly Volland Mahamadou
Diakite Yi Wang Urvi Vyas Emilee Minalga Joshua
de Bever Chris Dillon Joshua Coon Justin
Tidwell Lexi Farrer Robb Merrill Henrik Odeen

• Funding
• Focused Ultrasound Surgery Foundation
• Siemens Medical Solutions
• NIH grants
• F31 EB007892-01A1,
• R01 EB013433, and R01 CA134599.