Radiotherapy Process
Deformable Registration and
Dose Accumulation
Prepare
Patient
Model
Planning CT
DELIVERY
Treatment
Planning
Contour
Contour
Changing
Changing
Organs
Organs
Beam
Selection
Deformable
Deformable Image
Image
Registration
Registration
to
to Planning
Planning CT
CT
Record
Record
Dose
Dose
Kristy K Brock, Ph.D.
Physicist, Radiation Medicine Program, Princess Margaret Hospital
Assistant Professor, Depts of Radiation Oncology & Medical Biophysics,
University of Toronto
Scientist, Ontario Cancer Institute
MRI, PET
Image
Registration
Contouring
Uses of Deformable Registration
for Adaptive H&N RT
Dose-Volume
Objectives
IMRT
Optimization
On-line
On-line
IGRT
IGRT
Off-line
Off-line
IGRT
IGRT
(accelerator)
(accelerator)
(e.g.,
(e.g., PET,
PET, MRI)
MRI)
Evaluate,
Export &
Verify Plan
Deliver
Treatment
Target
Localization
• Deformable Registration
to
– Accuracy
– Validation
– Efficiency
• Volumetric propagation
• Dose Accumulation
– Full deformation
– Tissue Tracking
• Dose accumulation
Compute
Compute Dose
Dose In
In
Deforming
Deforming Organs
Organs
Requirements
• Contour propagation
– ‘Surface’ deformation
– Mapping the target for image guidance
– Re-planning
M Sharpe
PLANNING
t1
– Volume reduction
– QA
– Reference state
– Accounting
1
Validation Techniques
• Pretty Pictures
• Volume Overlap
– DICE, etc
• Intensity Correlation
DSI =
– Difference Fusions
– CC, MI, etc
2⋅ A∩ B
A+B
NCC ( A, B) =
(A − A )(B − B )
(A − A ) (B − B )
2
• Landmark Based
– TRE, avg error, etc
2
Demons Implementation
Rigid
Demons
S. Nithiananthan, et. al.
10 H&N Radiotherapy
Patients
• CBCT images acquired on Fx 1
and Fx “N” (weeks apart)
Eight Target Points
Left & right temporal bone
Left & right coronoid process
Cervical vertebra inferior aspect
Left & right auditory canal
Soft tissue point in oropharynx
Voxel size: 1x1x2 mm3
Accuracy: 0.8-1.6 mm (TRE)
Time: ~ 50 s
Demons Implementation
• P Castadot, et. al, Rad
Onc 89, 2008
• 5 Patients
• Repeat CT images
• Edge-preserving
denoising filter
followed by level-set
• DSI median 0.85
• CC median 0.97
Demons Implementation
• Wang, et al, PMB 2005
• Difference in images (ext)
and gradient of image
(int) act as forces
• Addition of active force
(gradient of moving
image)
• Accuracy: 96% voxels <
2 mm for mathematical
phantom
2
Feature Matching + Min of Elastic E
a CT as reference and b CBCT as target
image; c/d overlays before/after registration
• M Söhn et. al. , Med Phys
35(3) 2008
• NMI, constrained by min
elastic E, combined with
B-spline
• CBCT to Planning CT
• 5200 featurelets (15x15x5
voxels)
• CT to MR
• 1800 featurelets (10x10x8
voxels)
• Qualitative results
• 14 min on 8 CPUs
Contour Matching and Elastic Body
• M Sharpe, ASTRO 2005
• Initial CT and MR for
planning
• Eight Patients
• Seven repeat MR (weekly)
• Contour for each CT and
MR dataset
• Over 100 structures per
patient
• Qualitative visual
validation
• Time: contouring +++
Optical Flow w/GPU
• K Noe, et.al., Acta Onc
47(7) 2008
• Planning CT + 6 CBCTs
• 6 boney anatomy points
• Errors reduced from 2.2
± 0.6 mm (rigid) to 1.8 ±
0.6 mm
• Time: 64 s
Red/blue visualization of the difference between the rigid registration (left) and the
deformable registration (right) of CBCT image 3 to CBCT image 1
Biomechanical Models
• Boundary Conditions:
– Rigid registration on
each VB
– Rigid Registration on
mandible
• Parotid glands and
tumor
– Linear elastic FEA
• Multi-organ encased
in body contour
• Time: ~10 min
3
Biomechanical Models
Success and Limitations
• Surface-based Accuracy
• Initial (Rigid Reg at Tx)
• 5.4±0.3 (GTV) mm
4.1±0.4 (LPG) mm
3.1±0.7 (RPG) mm
• 0.7±0.2 (GTV) mm
0.7±0.2 (LPG) mm
1.7±0.4 (RPG) mm
Retrospective Study of Volumetric Changes in Major
Salivary Glands (MSGs) with curative RT in H & N
Cancer Patients
Success!
Limitations
• Fast
• Multi-modality
• ‘Accuracy’
• Fast… enough?
– Need real-time?
• Accurate enough?
– Pretty pictures?
– Just surfaces?
– Only a few points?
Prospective Monitoring of Changes in Parotid
Gland (PG) Size vs Dose Accumulated
10 patients: weekly MRIs during RT
PGR 1
PG Volume
during RT
Bilateral RT > 47 Gy
PGL 1
PGR 2
40
4 months
pre
PGL 2
35
29 cc
cc
post
PGR 3
48 %
PGL 3
30
PGR 4
25
PGR 5
PGL 5
20
PGR 6
15
PGL 6
PGL 7
10
Unilateral RT
Gy, d/F < 0.5
Dmean = 1.9
6 weeks
14.8 cc
PGR 8
5
PGL 8
PGR9
0
pre
0
w1
w2
w3
w4
10
20
30
40
w5
50
w6
60
w7
70 Gy
PGL 9
PGR 10
PGL 10
Average Reduction: 48 %
4
Parotid Gland Response
Model the volume change
MORFEUS
Right
Create contours in
TPS
Contours
Surface Mesh
Analysis and
visualization
Geometric changes using MVCT
• C Lee, et. al., Rad Onc, 89
2008
• 10 patients, 330 daily
MVCTs
• Deformable registration
– Fast freeform deformable
registration via calculus of
variations (Lu PMB 2004)
• Visual validation
• Median loss: 21.3% (6.7 –
31.5%)
• Migrated to patient center:
-5.26 mm (0 – 16.35 mm)
Geometric changes using CT
• O’Daniel et. al. IJROBP
2007
• 11 patients, 2 CTs/week
• Demons Deformable
Registration
• Increase in parotid dose:
median 1 Gy
5
Shrinking Volume
• How do we model the reduction?
• Does it have dosimetric consequences?
• What volume to we use for the DVH?
Modeling Volume Reduction
• Tumor with ‘core’
• Heterogeneous plan
• Variation in volume
reduction
– Homogeneous
– Dissolving rim
– Necrotic Core
Modeling Volume Reduction
• Tumor with ‘core’
• Heterogeneous plan
• Variation in volume
reduction
Modeling Volume Reduction
• Tumor with ‘core’
• Heterogeneous plan
• Variation in volume
reduction
– Homogeneous
– Homogeneous
– Dissolving rim
– Necrotic Core
– Dissolving rim
– Necrotic Core
6
Modeling Volume Reduction
Dosimetric Effect
• Tumor with ‘core’
• Heterogeneous plan
• Variation in volume
reduction
– Homogeneous
– Dissolving rim
Homogeneous
Necrotic Core
Dissolving Rim
Plan
% Volume
Modeling Volume Reduction
– Necrotic Core
% Dose
Modeling Volume Reduction
Dosimetric Effect
Scenario
V33 [%]
D75
∆ From
plan
Plan
V33
[cc]
142
57
2411
Homogeneous
170
68
3058
647
Necrotic Core
161
64
2885
474
Dissolving Rim
177
70
3167
756
Accuracy of Dose Accumulation
• How do we QA dose accumulation?
• What is the ‘gold’ standard?
– Ion chambers/TLDs/Film can’t deform
– Put them in a deforming phantom?
• How accurate does it need to be?
– Every voxel exactly right?
– Isodose line comparison (2%/2mm)?
7
– ~ 1% ± 5%
• 95% of isodose
surfaces are within 1.5
mm
Accumulated Dose
• Polymer based gel
• MR read out
• Mean difference (4 Gy
max):
Use of Deformable Gel Dosimeter
Gel Dose
Gel Dose
Planned Dose
Use of Deformable Gel Dosimeter
– Mean: 1% ± 13%
• 95% Isodose within 2.5 mm
• 92% of voxels within SD of
reference
Acknowledgements
Summary
• Deformable registration is a promising tool for
understanding changes in H&N
• Validation for contour propagation is very positive
• Validation of dose accumulation is challenging but
possible
• Use for dose accumulation is very promising…
• However, we must work to understand how the
tumor and normal tissues are responding over the
course of radiation
• Deform gel cyclically by 1
cm
• Deliver 4 Gy in 8 beam plan
• Defm Acc: < 2 mm
• Gel readout in MR
• Calibration using control gel
• Difference:
•
•
•
•
Carolyn Niu – Gel dosimetry for dose accumulation
Adil Al-Mayah – Biomechanical modeling of neck flexion
Mike Velec – Effects of tumor volume change
G Studer – Parotid glad change
Funding
• NIH R01
• NCI Canada – Terry Fox Foundation
• Ontario Institute for Cancer Research,
• Cancer Care Ontario Research Chair
• Elekta Oncology Systems, Philips Medical Systems,
RaySearch Laboratories
8