Development of Anatomically Realistic Numerical Breast Phantoms with Accurate Dielectric Properties

1
Development of Anatomically Realistic Numerical
Breast Phantoms with Accurate Dielectric
Properties for Modeling Microwave Interactions
with the Human Breast

• E. Zastrow(1), S. K. Davis(2), M. Lazebnik(1), F.
  Kelcz(3), B. D. Van Veen(1), and S. C.
  Hagness(1)
• (1) Department of Electrical and Computer
  Engineering, University of Wisconsin Madison
• (2) MIT Lincoln Laboratory, Lexington, MA
• (3) Department of Radiology, University of
  Wisconsin Madison

2
Motivation

• Emerging diagnostic and therapeutic microwave
  applications in breast cancer detection and
  treatment
• Invaluable pre-clinical tool FDTD computational
  electromagnetics models of the breast
• Relevance of numerical breast phantoms highly
  realistic, flexible complements to physical
  phantoms
• Anatomical and physical-properties realism is
  critical to conducting meaningful studies because
  of the heterogeneous nature of breast tissue
• We have developed a database of 3D anatomically
  realistic numerical breast phantoms with accurate
  dielectric properties
• Anatomical realism achieved through the use of
  breast MRI
• Dielectric-properties realism achieved through
  the use of data from the recently completed
  large-scale Wisconsin-Calgary study

M. Lazebnik et al, Phys. Med. Biol., vol. 52,
pp. 2637-2656, 2007. M. Lazebnik et al, Phys.
Med. Biol., submitted April 2007.
3
Schematic

• Goal To transform a raw MRI into a numerical
  breast phantom for use in FDTD simulations

4
Raw MRI

• 84 sagittal slices
• In-plane resolution
• 0.78125 x 0.78125 mm
• Between-plane resolution
• 1.5 mm

5
Raw MRI Examples from 4 ACR classes
American College of Radiology (BI-RADS, 2003)
I. Almost entirely fat
II. Scattered fibroglandular
III. Heterogeneously dense
IV. Extremely dense
6
MR image processing Homomorphic filtering
Coronal image before filtering
Coronal image after filtering
7
MR image processing Resolution adjustment
Interpolate the 3-D MR image
New resolution 0.5 x 0.5 x 0.5 mm
Coronal slice
(cm)
(cm)
(cm)
8
MR image processing Image segmentation
Perform edge-finding
New resolution 0.5 x 0.5 x 0.5 mm
Binary mask traversing
left to right
Coronal slice
(cm)
(cm)
(cm)
9
MR image processing Image segmentation
Combine masks
Perform edge-finding
Binary mask traversing
left to right
Other masks for same coronal slice
Composite coronal mask
left to right
right to left
top to bottom
bottom to top
10
MR image processing Image segmentation
Find best-fit ellipse
Composite coronal mask with best-fit ellipse
left to right
right to left
bottom to top
top to bottom
11
MR image processing Image segmentation
Form smooth breast surface from set of elliptical
masks
Smooth breast surface
Elliptical mask
12
MR image processing Image segmentation
Segment the breast interior from the background
region
Segmented breast volume
Segmented coronal image
13
Additional structural development
Skin and chestwall
Smooth breast surface
Elliptical mask
14
Additional structural development
Skin and chestwall
Create skin layer
Skin contour on 3D model
Skin layer on coronal slice
15
Additional structural development
Skin and chestwall
Create skin layer
3-D anatomical model with chestwall
Coronal image with skin layer
16
Overview of normal breast tissue properties
(Wisconsin-Calgary study1)
354 normal tissue samples from reduction surgeries

• Normal tissue samples divided into three
  adipose-defined groups 0-30, 31-84, and
  85-100 adipose content
• Cole-Cole1 and Debye2 model parameters reported
  for each group

1M. Lazebnik et al, Phys. Med. Biol., vol. 52,
pp. 2637-2656, 2007. 2M. Lazebnik et al, IEEE
MWCL, in press, 2007.
17
Dielectric properties mapping Overview

• MRI breast interior shows a mixture of fatty
  tissue (high) and fibrous, connective, and
  glandular (FCG) tissue (low).
• We fit the histogram of MRI voxel intensity with
  a two-component Gaussian Mixture Model (GMM).
• Each component covers the MRI intensity region
  corresponding to either fatty or FCG tissue.
• GMM parameters determine the 7 mapping intervals.
• Each interval is linearly mapped to a range of
  dielectric properties.

18
Dielectric properties mapping
Gaussian mixture model (GMM)

• Probability density function (pdf) for a
  two-component GMM

f1
19
Dielectric properties mapping
GMM parameters

• Probability density function (pdf) for a
  two-component GMM
• Solve for
• ?1,?2,?1,?2,?12,?22

f1
20
Dielectric properties mapping
Piecewise-linear (PL) approach

• Probability density function (pdf) for a
  two-component GMM
• Define piecewise-linear mapping parameters
• mfib,?fib,m?fib,Mfib
• mfat,?fat,m?fat,Mfat

21
Dielectric properties mapping
Fitting of the GMM
extremely dense
heterogeneously dense
CASE 1
scattered fibroglandular
almost entirely fat
CASE 2
22
Dielectric properties mapping Case 1(two
tissue classes are well separated)
23
Dielectric properties mapping Case 2(two
tissue classes are not well separated)
24
Dielectric properties mapping Normal tissue
properties from Wisconsin-Calgary study
Source Lazebnik et al., Phys. Med. Biol.,
vol.52 (2007)
25
Dielectric properties mapping Normal tissue
properties from Wisconsin-Calgary study
Source Lazebnik et al., Phys. Med. Biol.,
vol.52 (2007)
26
Result 3D grid-based numerical breast phantoms
355 x 253 x 310 grid cells
352 x 307 x 316 grid cells
?r _at_ 6 GHz
III. Heterogeneously dense
IV. Extremely dense
(cm)
(cm)
332 x 202 x 269 grid cells
328 x 212 x 215 grid cells
(cm)
(cm)
(cm)
(cm)
27
Phantoms available via online repository

• The repository is publicly accessible through
  http//uwcem.ece.wisc.edu/.
• The repository currently includes
• 2 type-I phantoms (almost entirely fat)
• 2 type-II phantoms (scattered fibroglandular)
• 2 type-III phantoms (heterogeneously dense)
• 2 type-IV phantoms (extremely dense)
• detailed instructions
• how to interpret data files (key for successfully
  importing phantom data into FDTD codes)
• how to customize dielectric properties models for
  frequency range of interest (within constraints
  of 0.5-mm grid resolution)

28
Applications

• The phantoms are currently being used in a
  variety of research projects at the UW-Madison.
• Example UWB microwave hyperthermia treatment of
  breast cancer

Space-time transmit beamforming results
(antenna array not shown)
Desired focus
absorbed energy (dB)
29
Acknowledgements

• We thank the following individuals for their
  assistance and support
• Mr. Henri Tandradinata, ZS Associates
• UW Hospital and Clinics radiology staff
• This work was supported by
• National Institutes of Health
• National Science Foundation