Evaluation A Automatic Deform

Register Methods on Head-and-Neck

CBCT Images for IGART

Yixin SONG†§,Chi Y, Ph.D.†, Liang J, Ph.D.† and Yan D, D.Sc.†

†
Radiation Oncology, William Beaumont Hospital, Royal Oak, MI
§
Radiation Oncology, Cancer Hospital Chinese Academy of Medical Sciences,
Beijing, China
 The Problems
• In recent years, image-guided RT(IGRT) has become an
important technique in tightening the planning target volume
margin.
• Currently, in IGRT sessions, daily on-line images are
registered into planning CT images by rigid-body image
registration with six degrees of freedom.
• The shapes or relative positions of target and region-of-interest
(ROI) change, simple rigid-body correction techniques may
not be sufficient for high-precision radiotherapy.
 The Problems
• The major obstacle that hinders the clinical implementation of
on-line adaptive RT is on-line region-of-interest (ROI)
delineation.
• It is impractical to manually contour all ROIs with the patient
on the table.
 Purpose of Our Work
• We implemented a fast variational-based deformable image
registration algorithm. The algorithm was used in the atlas-
based segmentation of daily head-and-neck (HN) CBCT
images.
• Develop a clinically practical offline and eventually online
adaptive therapy procedure and evaluate the potential impact of
the CBCT-based adaptive head-and-neck IMRT treatment.
• Evaluate the accuracy of image registration and segmentation
on H&N CBCT images.
 Methods and Materials

Deformable image registration maximizes the similarity

between the reference image R and the floating image T by
warping the floating image and, also, keeps the displacement
vector fields smooth.
 Treatment Protocol

Head-and-Neck IMRT Radiotherapy

– CT Scan: 120 kVp, 200 mA, 480mm FOV
– 2mm helical scan, reconstructed at 2mm slice thickness.
– 35 fractions @200cGy/fraction
– 7 Gantry angles
Online Image Guidance
– Head and Neck CBCT Scan
– F0,S0 Filter
– Gantry rotation Angle:179.9~181.1o
The Manually Contoured ROIs
Red-GTV1
Seashell-CTV1
Orange-CTV2_nodes
Forest-Brain stem
Brown-Left parotid
Maroon-Right parotid
Yellow-Mandible
Green-Spinal Cord
 Methods and Materials
• A general form of the objective function of grayscale-based
deformable image registration is given by the following
equation:

F (u f )  D ( R, T (u f ))   S (u f ),   0
2
D 
x
R( x)  T ( x  u f ) dx

2
u f
S  
x
x
dx

Tiezhi Zhang, Yuwei Chi, et al. Automatic delineation of on-line Head-and-neck computed
tomography images: toward on-line adaptive radiotherapy Int J Radiat Oncol Biol Phys
2007; 68:522-530
 Methods and Materials
• The planning CT images as floating images and the CBCT
images as reference images in deformable image registration,
so that the displacement map uf could be directly used to
transform ROI masks on planning images without reversion.

M d ( x)  M p ( x  u f )
 Daily Procedure
1. Patient alignment to marks

2. CBCT Initial Acquisition

3. CBCT registration &couch correction

4. CBCT 2th Acquisition

5. Treatment delivery
 Methods and Materials
• Three difference indexes, namely Volumei , Cenij , Modi
were calculated for each ROI

Volumei=Volumei (Reg ROI)-Volumei (ROI on Planning CT).

Cenij=Center of ROIij (Reg) - Center of ROIij (XVI)

  x 
3
1 2
Mod i  ij (Re g )  xij ( XVI )
N xROI j
Methods and Materials
 Verify the accuracy of the rigid registration
1.8
1.6
Translation Error(mm)

1.4
1.2
1.0 Pinnacle
0.8 Deformable
0.6
0.4
0.2
0.0
LR PA SI Magnitude
 Results

Overlap Planning Helical CT and Deformable CT

Reference ROIs VS Deformed ROIs
 Volume Change
6.0
4.0
2.0
Difference Volume(cc)

0.0
-2.0
-4.0
-6.0
-8.0
-10.0
-12.0
-14.0
Center of ROIs Mean followed by standard deviation of
discrepancies between automatically delineation and XVI
registration
Center Coordinates ( Mean±SD, mm)
ROI Name RL PA SI Mag
Brainstem 0.2±0.2 -0.5±0.3 1.9±1.3 2.0±1.2
Cord 0.5±0.1 -0.1±0.3 0.4±0.1 0.7±0.1
CTV_1 0.4±0.2 0.3±0.3 0.8±0.6 1.0±0.5
CTV2_nodes 0.5±0.2 0.1±0.4 0.5±0.2 0.8±0.1
GTV_1 0.4±0.2 0.3±0.3 0.8±0.7 1.0±0.6
Left Parotid 0.3±0.1 0.8±0.3 1.7±0.8 1.9±0.9
Mandible 0.4±0.1 -0.7±0.3 1.5±0.2 1.8±0.2
Right
Parotid 0.9±0.2 1.1±0.4 1.7±0.8 2.2±0.7
 Voxel of ROIs Mean followed by standard deviation of
discrepancies between automatically delineation and XVI
registration
Voxel Coordinates ( Mean±SD, mm)
ROI Name RL PA SI Mag
Brainstem 0.2±0.2 -0.5±0.3 1.9±1.3 2.0±1.2
Cord 0.5±0.1 -0.1±0.3 0.4±0.1 0.7±0.1
CTV_1 0.4±0.2 0.3±0.3 0.8±0.6 1.0±0.5
CTV2_nodes 0.5±0.2 0.1±0.4 0.5±0.2 0.8±0.1
GTV_1 0.4±0.2 0.3±0.3 0.8±0.7 1.0±0.6
Left Parotid 0.3±0.1 0.8±0.3 1.7±0.8 1.9±0.9
Mandible 0.4±0.1 -0.7±0.3 1.5±0.2 1.8±0.2
Right Parotid 0.9±0.2 1.1±0.4 1.7±0.8 2.2±0.7
 Conclusions
•Atlas-based image segmentation incorporates a priori
knowledge about the shape and image characteristics of the
ROIs.
•Atlas-based image segmentation is able to contour the ROI
with a boundary that is not located on the edge.
•Atlas-based image segmentation is able to contour all ROIs
at once.
•Deformable image registration relies on image quality.
•Patient’s daily anatomic position and shape can be
feasibility determined by image auto-registration between
daily CBCT and reference HCT.
 Acknowledgements
• Yuwei CHI Ph.D.
• Liang JIANG Ph.D.
• Di YAN D.Sc.
• Support in Part by NCI Grant CA091020