Poster HCPBM

1
Simulation of 4D-CT Images from Deformable
Registration between Inhale and Exhale
Breath-hold CT Scans
David Sarrut, Vlad Boldea, Serge Miguet and
Chantal Ginestet Léon Bérard anti-cancer center
radiation oncology departement
1 LIRIS, Université Lumière Lyon 2, 5, avenue
Pierre Mendès-France, 69676 Bron, France 2
Radiotherapy Department, Centre Léon Bérard, 28,
rue Laënnec, 69353 Lyon, France
darrut_at_gmail.com vboldea_at_gmail.com

• account for organ motion in lung cancer
  radiation treatment is an important challenge
• need patient-specific information about
  movements induced by breathing
• 4D images are not sufficient alone to provide
  useful information patient-specific breathing
  thorax model is needed

GOAL To build an artificial volumetric and
temporal breathing thorax model from two CT scans
acquired at inhale and exhale breath-hold.
II. Material and Methods

• Clinical Data
• 4 patients (Léon Bérard Center)
• 3 different breath-holds with ABC device (-0.2l,
  0.2l and 80)

III. Deformable registration
II. APriori Lung Densities Modification APLDM
Global energy minimization trade-off between 2
energies similarity (E1) et regularization(E2)
modifying density I1(x) of voxel x belonging to
slice z
I. Pre-processing

• Threshold, morphological
• tools
• Initial rigid registration

Two implementation schemes
ABC (air flow) signal - normal breathing followed
by a breath-hold of about 20 sec duration.
Slice z
?1, ?2 - Mean of lung intensities for slices z,z
1. Linear elastic (LE) regularization
Slice z
Image I1
Image I2
2. Gaussian regularization
Parameters i iteration ß?0,1 energies
trade-off ? 0.1- gradient descent step
??0.5, 0.65 ?max voxel displacement ?0.77,
1.0 ? trade-off laplacian divergence s
variance.
III. Intermediate vector field interpolation
linear pathway along its displacement vector

• 1 patient 4D-CT (MGH Boston)
• 10 3D-CT over the free breathing cycle

V. Jacobian based density generation (JBDG) x
original voxel x position of voxel x after
deformation jacobian in x HU2,HU1 original
and new Hounsfield values
0
20
40
60
80
100
Pan et al. 2004
III. Validation
IV. Results
Landmarks manually defined by experts
(reference) compared to Landmarks automatically
warped by deformable field
Vector Fields
Landmarks distances reference deformed images
Inspiration CT (reference)
Expiration CT
Landmarks distance for artificially generated
intermediate image 2.1 mm
Displacement field super-imposed on slices
Green selected by physician
Blue automatically found

• APLDM leads to statistically significantly
  better results
• Gaussian (G) and Linear Elastic (LE) similar
  results

• Simulated intermediate image close to
  intermediate image in 4D CT
• Preliminarily model does not take into account
  (yet) motion hysteresis
• 4D model is a basis for 4D radiation therapy
  planning
• Artifial 4D CT ?
• same information with low dose
• to correct artifact in acquired 4D CT

Acknowledgement The authors wish to thank Steve
B. Jiang, Gregory C. Sharp and Noah C. Choi from
Massachusetts General Hospital, Boston, MA, USA
(Radiation Oncology Department) for their
collaboration and for supplying 4D-CT
acquisitions.