E Kanoulas 1 , JA Aslam 1 , GC Sharp 2 , RI Berbeco 3 , S Nishioka 4 , H Shirato 5 , SB Jiang 2
description
Transcript of E Kanoulas 1 , JA Aslam 1 , GC Sharp 2 , RI Berbeco 3 , S Nishioka 4 , H Shirato 5 , SB Jiang 2
Derivation of the tumor position from external respiratory surrogates with periodical updating of external/internal correlation
E Kanoulas1, JA Aslam1, GC Sharp2, RI Berbeco3, S Nishioka4, H Shirato5, SB Jiang2
(1)Northeastern University, Boston, MA, (2)Mass General Hospital and Harvard Medical School, Boston, MA, (3)Brigham and Women's Hospital and Harvard Medical School, Boston, MA, (4)Department of Radiology, NTT Hospital, Sapporo, Japan, (5)Hokkaido University, School of Medicine, Sapporo, Japan
Motivation Beam tracking or gating of moving tumors requires
precise real-time tumor localization
Fluoroscopic marker tracking is accurate however requires large imaging dose
Deriving tumor position from external surrogates is dose free however inaccurate due to the uncertainties in internal/external correlation
This work Derive tumor position using external surrogate Periodically image to update internal/external
surrogate correlation Study the minimum updating rate and optimal
updating approach
Data used for the study
X-ray imagers
Laser housing
NTT Hospital (Dr. Seiko Nishioka)
Mitsubishi RTRT system(track 3D tumor position)
AZ-733 “Resp-gate” systemMonitors abdominal surface
3-D internal tumor motion + 1-D abdominal motion
Berbeco et al., “Residual motion of lung tumours in gated radiotherapy with external respiratory surrogates”, PMB, (2005 Aug 21), 50(16), p. 3655-67
Updating the correlation function
During patient setup both external and
internal surrogate position at 30Hz
During treatment external surrogate
position at 30Hz periodically (at low
frequency) internal marker position
0 50 100 150 200 250-10
-5
0
5
10
15
External surrogate position (mm)
Mar
ker p
ositi
on o
n S
I (m
m)
Training data during patient setupActual data during treatmentUpdate PointLine fit to training data
y = -0.081 * e + 13.254
Update Methods
1. Aggressive update (through the update point)
a. Shift line through update point
b. Re-fit line and force it through update point
2. Conservative update (balance between update and training points)
a. Re-fit line with extra weight to update point
b. Minimize the distances to update point and previous line
Method 1a. Shift line through point
0 50 100 150 200 250-10
-5
0
5
10
15
External surrogate position (mm)
Mar
ker po
sitio
n on
SI (
mm
)Training data during patient setupActual data during treatmentUpdate PointLine fit to training dataLine after update
y = -0.081 * e + 9.784
y = -0.081 * e + 13.254
Method 1b. Re-fit line through point
0 50 100 150 200 250-10
-5
0
5
10
15
External surrogate position (mm)
Mar
ker po
sitio
n on
SI (
mm
)Training data during patient setupActual data during treatmentUpdate PointLine fit to training dataLine after update
y = -0.081 * e + 13.254
y = -0.101 * e + 11.884
0 50 100 150 200 250-10
-5
0
5
10
15
External surrogate position (mm)
Mar
ker po
sitio
n on
SI (
mm
)
Training data during patient setupActual data during treatmentUpdate PointLine fit to training dataLine after update
y = -0.081 * e + 13.254
y = -0.092 * e + 12.483
Method 2a. Re-fit line with extra weight to point
0 50 100 150 200 250-10
-5
0
5
10
15
External surrogate position (mm)
Mar
ker po
sitio
n on
SI (
mm
)
Training data during patient setupActual data during treatmentUpdate PointLine fit to training dataLine after update
y = -0.081 * e + 13.254
y = -0.0757 * e + 11.434
Method 2b. Minimize the distances to update point and previous line
Results
0 1 2 5 10 150
0.5
1
1.5
2
2.5
3
3.5
Image Acquiring Frequency (hz)
95%
Con
fiden
ce In
terv
al (m
m)
No update
1a.Shift Line through update point
1b.Re-fit line & force it through update point
2a.Re-fit line with extra weight to update point
2b.Min. dist. to update point & previous line
5 patients 25 data sets Large SI
motion only (>20 mm)
Conclusions
Tumor localization using external surrogates requires periodical update of the internal/external correlation
Update frequency down to 2Hz gives 2 mm motion error (95% confidence level)
The aggressive update methods outperform the conservative ones at high update frequencies while the opposite is true for low update frequencies