8/2/2012
Hybrid linac-MR for Real-time
Tumor tracking and RT
B.Gino Fallone
Dept. of Medical Physics, Cross Cancer Institute &
University of Alberta, Edmonton Canada
Linac-MR.ca
OUTLINE: 1/2
Need for Real-time MR-guided RT
Design issues: Advantage of bi-planar MR
Avoid irradiation of MRI
Avoid eddy currents when Linac rotates wrt MRI
Allows perpendicular and parallel configuration
OUTLINE: 2/2
Proof of Concept:Cross Cancer Institute
Tumour Tracking (4/s, auto-contouring, prediction for
MLC action, etc)
Auto-contouring
Prediction for MLC deadtime
Advantage of Parallel Configuration
Simplification of Magnetic Shielding
Improved Internal Patient Dosimetry
Avoid electron return artifacts
Improved penumbra
INSIGNIFICANT Increase in surface skin dose
Site: 0.5 T bi-planar MR (60 cm gap) - 6 MV linac
1
8/2/2012
Real-time tracking & Adaptive RT
• Minimize PTV margin in dynamic target
10~20 mm
CTV
CTV
Critical
Structure
Critical
Structure
PTV
State of the art
Real-time MR guided RT
Bi-Planar linac –MR
CCI
6 MV medical linear accelerator (linac).
Low-field biplanar MR imager.
CCI Concept
Design Advantages
4. Source of Linac RF is
away from the MRI
avoiding RF interference
5. RF & magnetic shielding
avoids
mutual interferences
MRI pole
Linac
beam stop
1. Unobstructed Beam
2. Linac is stationary wrt
MRI avoiding MR field
distortions
3. Field strength avoids
distortion of dose
distribution,
susceptibility artifacts
MRI pole
2
8/2/2012
Rotating Biplanar Options
PERPENDICULAR
TRANSVERSE
PARALLEL
LONGITUDINAL
IN-LINE
C. Kirkby, B. Murray, S. Rathee, B.G. Fallone. Lung dosimetry in a linac-MRI radiotherapy unit with
longitudinal magneticfield, Med. Phys. 37(9), 4722-4732 (2010).
Rotating Biplanar Options
longitudinal, parallel, inline
transverse, perpendicular
C. Kirkby, B. Murray, S. Rathee, B.G. Fallone. Lung dosimetry in a linac-MRI radiotherapy unit with longitudinal
magneticfield, Med. Phys. 37(9), 4722-4732 (2010).
DEC, 8, 2008
CT- IMAGE
acrylic rectangular cube, 15.95
x 15.95 x 25.4 mm with holes of
diameters 2.52 mm, 3.45 mm
and 4.78 mm.
immersed in a 10 mM solution
of CuSO4 within a plastic
container 22.5 mm inner
diameter.
MR- IMAGE
Taken without Linac Irradiation
Inside an inductively tuned
solenoid RF coil with an
integrated pin-diode
transmit/receive switch
SNR: 80
gradient echo sequence, 90.0o, TE=14 ms,
TR =300 ms,
BandWDTH=10 kHz,
slice = 7 mm, 128 x128,
FOV 100 mm x 100 mm,
1 averages
Each image was obtained in about 38 secs.
Max Grad strength: 30 mT/m per channel
Slew rate: 300 mT/m/ms
MR-IMAGE
Taken with Linac Irradiation
SNR: 61
Fallone et al. First MR images
obtained during megavoltage photon
irradiation from a prototype
integrated linac-MR system,
Med Phys, 36(6), 2084- 2088 (2009).
3
8/2/2012
SAME SNR WITH/WITHOUT
LINAC IRRADIATION
May, 2009
Lamey, Burke, Rathee, Fallone, SU-FF-J-84, AAPM 2009.
RF Shielding
k-space data of the first (left) and second (right) phantoms taken during linac
pulsing while the RF shielding was insufficient. The data illustrate clear ‘lines’ in
the k-space from the RF interference picked up by the MRI coil.
Lamey, Burke, Blosser, Rathee, DeZanche, Fallone, Radiofrequency shielding for a
linac-MRI system, Phys. Med. Biol., 55(4), pp. 995-1006(2010).
RF in Magnetic Fields
Millenium MLC Motor
Lamey, Yun, Burke, Rathee, Fallone, Radiofrequency noise from an MLC: a feasibility study of
the use of MLCs for linac-MR systems, Phys. Med. Biol., 55(4), pp..981-994 (2010).
One leaf moving
4
8/2/2012
MR Images with MLC at 70 cm
from Isocentre
UNSHIELDED
STATIONARY
MOVING (13 leafs)
Lamey et al, COMP, Vancouver, BC, July 21-24, 2009
MR Images with MLC at 70 cm
from Isocentre
STATIONARY
MOVING (13 leafs)
M. Lamey*, J. Yun*, B. Burke*, S. Rathee, B.G. Fallone, Radiofrequency noise from an MLC:
a feasibility study of the use of MLCs for linac-MR systems, Phys. Med. Biol., 55(4), pp.981994 (2010).
Tumour Tracking Concept:
Cross Cancer Institute
Yun, Mackenzie, Rathee, Fallone, COMP, Vancouver, BC, July 21-24, 2009
5
8/2/2012
Summary of lung tumour motion
studies
Upto 15 mm AP
Upto 50 mm SI
Upto 10 mm LR
-Speed of lung tumour motion : 4~94 mm/s
-Up to 20% volume change & 50 degrees rotation
Intra-fractional tumour tracking scheme
On-line MLC
conformation
Intra-fractional
MR imaging
Motion
prediction
Autocontouring
Current
tumour
position
(i.e. centroid)
AUTO-CONTOURING
6
8/2/2012
Auto-contouring algorithm
ROI
Pre-treatment image
Motion Range
Performed by an Expert user before treatment
- ROI: single image containing least motion artifacts (ex. end of exhale period)
- Motion Range: superimpose pre-treatment images for a few breathing cycles
Auto-contouring algorithm
Approx
tumour
position
:
ROI
Crosscorrelation
Motion Range Max correlation:
Tumor position
Contour
tumor
:
Tumor Position
Contour
Auto- contouring:
Evaluation Experiment
• Lung Phantom is driven in by programmable 1-D motor.
•Optical Encoder independently records position of tumour for comparison
•Realistic tumour movement (inf-sup) as determined through Cyberknife patient data
7
8/2/2012
MR Lung Phantom
Lung Motion Phantom
Imaging at 3 T
Relaxations times and
proton densities
corresponding to 0.2
and 0.5 T
Noise added to obtain
SNR/CNR for 0.2 and
0.5 T
Evaluation of auto-contouring
Phantom study
Scan lung tumour motion phantom in 3T MRI w/ dynamic
bSSFP (~ 4 fps)
Process images to approximate lung tumour images at 0.2/0.5T
Compare auto-contoured shape and its centroid to reference ones
In-vivo study
Scan NSCLC patient in 3T MRI for 3 min (~ 4 fps)
Degrade images to simulate 0.2 ~ 1.5 T images
Compare auto-contoured shape and its centroid to reference
ones.
8
8/2/2012
Auto-contouring – in-vivo study
Manual contouring
3T bSSFP, ~4 fps
Auto-contouring
0.5T equiv.bSSFP
Auto-contouring results
Phantom study
– 2 tumour shapes, 4 motion patterns (600 images each)
0.2T
0.5T
RMSE (mm)
< 0.92
< 0.55
Dice’s coeff.
> 0.93
> 0.96
In-vivo study
– 1 patient, 650 images
0.2T
± 1.15
0.84 ± 0.06
∆ d (mm) in 2D
1.71
Dice’s coeff.
Auto-contouring
0.5T
1.38
0.86
± 0.69
± 0.03
takes < 5 ms/image (Intel i7, 32 bit Windows7, Labview)
Real Time Dynamic Scans (~4 fps)
3T
CNR ~ 48
Estimated 0.5T
CNR ~ 8
(Reduce by Factor of 6,
Conservative Estimate)
9
8/2/2012
Auto-contouring (3 T images)
Original image
Auto-contoured image
Auto-contouring (Est. 0.5 T
images)
Original image
Auto-contoured image
Auto-contouring (Est. 0.2 T
images)
Original image
Auto-contoured image
10
8/2/2012
Centroid Positions
RMSE = 0.5mm
Summary of Tracking Measurements
CNR 13.2 at 0.5 T for bSSFP (~ 4 fps)
CNR 5.6 at 0.2 T for bSSFP ( ~ 4 fps)
our auto-contouring algorithm obtained good quality
contours of spherical tumours with CNR ≥ 2.5
Conventional dynamic lung imaging sequences
provide sufficient tumour-tissue CNR and temporal
resolution for real time MR lung tumour tracking at
0.2T and for 0.5 T
J. Yun, E. Yip, K. Wachowicz, S. Rathee, M. Mackenzie, D. Robinson, B.G. Fallone,
Evaluation of a lung tumor auto-contouring algorithm for intra-fractional tumor tracking
using low-field MRI A phantom study,, Medical Physics, 39(3), pp. 1481-1494(2012).
PREDICTION
11
8/2/2012
Motion Prediction
Sources of system delay
Detection of
tumour position
Signal
acquisition
Beam
delivery
Image
Image
MLC
reconstruction processing motion
Total system delay
Motion prediction is required to compensate for the
tumour motion during the system delay
Artificial Neural Networks (ANN)
A mathematical model inspired by biological NN
structure
Number of studies used ANN in breathing motion
prediction in Medical Physics
Comparative studies stated superior performance of
ANN in breathing motion prediction over other
methods (linear filter, Kalman filter, etc)
Human neuron vs. Neuron model
Human neuron
Neuron model
Single neuron
Single neuron
Neural networks
Neural networks
12
8/2/2012
Linear neuron
i1
i2
w11
…
w31
Output
w21
in
i1
w1
i2
w2
…
w3
Output
∑ inwn
in
Nonlinear neuron
i1
i2
w1
…
w3
Output
w2
in
i1
w1
i2
w2
…
w3
Output
t=∑ inwn
in
Feedforward neural networks
Forward phase
i1
i2
…
in
Input layer
(Previous positions)
1w
11
1w
21
1w
31
1w
12
1w
22
1w
32
1w
13
1w
23
1w
33
3w
11
2w
21
2w
31
2w
12
2w
22
3w
11
3w
21
Output
2w
32
Hidden layer
Output layer
(Future position)
13
8/2/2012
Learning (i.e. weight updating)
i1
i2
w11
…
w31
output
w21
Error = Desired output – Actual output
in
Cost function:
C = ∑ Error 2
1. Find ∆wij that could minimize C
(least mean square, steepest gradient, etc)
2. Update each w : wnew = wold+∆w
ANN structure in our study
i1
i2
Output
…
w1
w2
w3
in
Output layer
Input layer
(Previous positions)
•
(Future position)
Hidden layer
non-linear neurons, sigmoid func.f(x) =
• freedom to choose 1 or 2 hidden layers
•
1
1 + e-x
no limitation selecting # neurons in each layer
Challenges in motion prediction in MRI
based tumour tracking
1.Determining the best ANN structure for each patient
.
- ANN performance depends on its structure
- Large variations in lung tumor motions in each patient
2. Imaging rate of MRI (typically 3 - 4 fps)
- MLC is triggered at same rate: 3 – 4 Hz
14
8/2/2012
Optimization of ANN structure
• Particle Swarm Optimization (PSO)
Stochastic optimization technique inspired by social behavior of
bird flocking or fish schooling
<source: http://www.pbs.org/wgbh/nova/sciencenow/3410, http://old.kookje.co.kr/news2006/asp/center.asp>
Optimization of ANN structure
• Particle Swarm Optimization (PSO)
: Min [ RMSE (measured position – predicted position) ]
of a training pattern
1st iteration
(updated
locations)
Start with
7 particles
(7 ANN structures)
End
(best particle)
Solution to tracking failures
• Multiple ANNs & predictions
More frequent MLC motion triggering
(ex. 1 ANN vs. 7 ANN, imaging at 3.6 fps, predict 280 ms future)
1 ANN predicts a single
tumour position at 560 ms
280 ms
MLC motion
based on previous 2 positions
280ms
1 ANN:
P(0 ms)
P(280)
triggered
every 280 ms
P(560)
P’(560)
P(t) : tumour position
at t (trained
(ms) detected
7 ANNs
separately)
predict
seven tumour positions
from each MR
image
at
560
800
ms
(3.6 fps = every 280 ms)
7 ANN:
P(0 ms)
P(280)
P’(560)
P(560)
40 ms
MLC motion
triggered
P’(800)
every 40 ms
15
8/2/2012
Evaluation of ANN performance using patient
data
• Cyberknife system used to produce 3D estimated tumour positions at 25Hz
• 29 lung cancer patients data (≥ 3 fractions of treatment per patient) was used.
• RMSE is calculated between original & predicted tumor positions in SI
directions of lung tumour motion
Motion prediction
Our Algorithm’s Prediction-- 1D sup-inf tumour position in 280 ms advance
Tested on 3D tumour motion by Cyberknife: (Suh et al. PMB, 53,36-23-40(2008)
: Original data
: Prediction
: Error (mm)
12
Position
(mm)
12
10
Position
(mm)
10
8
8
6
6
4
4
2
0
-2 0
5
10
15
20
25
30
35
40
45
50
-4
60
Time (s)
2
0
-2 0
5
10
15
20
25
30
35
40
45
50
-4
55
60
Time (s)
12
Position
(mm)
10
Position
(mm)
16
14
12
10
8
6
4
2
0
-2 0
-4
55
8
6
4
2
5
10
15
20
25
30
35
40
45
50
55
60
Time (s)
0
-2 0
5
10
15
20
25
30
35
40
45
-4
50
55
Time (s)
J.Yun, M. Mackenzie, S. Rathee, D. Robinson, B.G. Fallone, An artificial neural network (ANN) based lung
tumor motion predictor for intra-fractional MR tumor tracking, Medical Physics, 39(7), pp. 4423-4433 (2012).
On-line MLC PREDICTION
CCI –linac-MR Prototype
16
8/2/2012
On-line MLC conformation
Control MLC on-line to conform to the tumour shape
at predicted position
Varian 52-leaf MLC (5 mm width, ~ 1 cm/s speed)
Lead screws
26 servo
motors
(motor +
encoder)
26 Leaves
On-line MLC conformation
Tracking software
Desired leaf
positions
While loop
1. Motor on/off
FPGA card
2. Direction
Electronics
MLC
Encoder output (current leaf positions)
FPGA : Field Programmable Gate Array
Demonstration of intra-fractional tracking
17
8/2/2012
Experimental Set-Up
RF shielding
MRI (0.2 T)
Phantom
Linac
10 MLC leaf
(5 leaf each
side, each
motor has
encoder)
Shaft
Encoder
Motor
Intra-fractional tracking scheme
Beam on while tracking
MR imaging at 4 fps (250 ms)
MLC Conformation
Target shape +
its future position
MR image of
moving target
Target auto-contouring
Motion prediction
Target shape
Determine centroid
Centroid (current
target position)
-10 MLC leaves (5 in each side) are used for tracking
- Encoder readings from phantom & MLC motors every 50 ms
- 2 minutes Beam-On (200 MUs delivered)
MR compatible motion phantom
Gafchromic film
Assembled
phantom
Polystyrene
case
Target (porcelain gel
+ 12 mM CuSO4)
MR image
during
tracknig
Phantom moving
direction
-
Phantom motion follows pre-defined 1-D motion pattern during tracking
-
To register films to each other, the edges of the phantom (indicated by
red dots) are manually marked.
18
8/2/2012
Phantom motion patterns used for tracking
2 cm
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Time(s)
- 2 cm
Sine pattern (period: 6.7 s, amp: 4 cm, max. speed : 1.8 cm/s)
Modified cosine pattern (simulating diaphragm motion during breathing)
(period: 5.1 s, amp: 4 cm, max: speed : 3.1 cm/s)
Demonstration of intra-fractional tracking
- For each motion pattern,
moving target was irradiated for 2 minutes as follow:
1. Without tracking
(MLC conformed to initial target position during irradiation)
2. With tracking + without motion prediction
3. With tracking + with motion prediction
Sine pattern - without tracking
Moving target
without tracking
Film profiles (while lines)
Dose (cGy)
Non-moving target
300
250
200
150
Non-moving
Moving w/o
tracking
100
50
0
Well defined edges
1
Blurred edges
51
101
151
201
251
301
pixel number
Encoder
reading
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
1
0.5
0
0
: Phantom
2
encoder reading
4
6
: MLC
8
Time(s)
motor encoder readings (averaged)
19
8/2/2012
Sine pattern - tracking without prediction
Moving target
tracking w/o prediction
Film profiles (while lines)
Dose (cGy)
Non-moving target
300
250
200
150
Non-moving
Moving w/o
prediction
100
50
0
Well defined edges
1
Blurred edges
51
101
151
201
251
pixel number
Encoder
reading
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
1
275 ms
system delay
0.5
0
0
2
: Phantom
4
encoder reading
6
: MLC
Time(s) 10
8
motor encoder readings (averaged)
Sine pattern - tracking with prediction
Non-moving target
Moving target
tracking with prediction
Film profiles (while lines)
Dose (cGy)
300
250
200
150
Non-moving
Moving with
prediction
100
50
0
Well defined edges
1
Well defined edges
51
101
151
201
251
pixel number
Encoder
reading
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
Time(s)
: Phantom
encoder reading
: MLC
motor encoder readings (averaged)
Sine pattern - penumbra comparisons
Dose (cGy)
300
250
200
Non-moving
Moving w/o tracking
Tracking w/o prediction
Tracking w/ prediction
150
100
50
0
1
51
101
151
201
251
pixel 301
number
- Physical target dimension : 60 mm
50 % profile
Non-moving
62.9 mm
20/80 % profile
(avg. both side)
6.5 mm
Moving without tracking
63.5 mm
33.0 mm
Tracking without prediction
62.4 mm
11.5 mm
Tracking with prediction
62.0 mm
7.3 mm
20
8/2/2012
Modified cosine pattern - without tracking
Moving target
without tracking
Non-moving target
Film profiles (while lines)
Dose (cGy)
300
250
200
150
Non-moving
Moving w/o
tracking
100
50
0
Well defined edges
1
Blurred edges
51
101
151
201
251
301
pixel number
Encoder
reading
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
1
0.5
0
0
: Phantom
2
4
encoder reading
6
: MLC
Time(s)
8
motor encoder readings (averaged)
Modified cosine pattern - tracking without prediction
Moving target
Non-moving target tracking w/o prediction
Film profiles (while lines)
Dose (cGy)
300
250
200
150
Non-moving
Moving w/o
prediction
100
50
0
Well defined edges
1
Blurred edges
51
101
151
201
251
pixel number
Encoder
reading
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
1
340 ms
system delay
0.5
0
0
: Phantom
2
4
encoder reading
6
: MLC
8
10
Time(s)
motor encoder readings (averaged)
Modified cosine pattern - tracking with prediction
Moving target
Non-moving target tracking with prediction
Film profiles (while lines)
Dose (cGy)
300
250
200
150
Non-moving
Moving with
prediction
100
50
0
Well defined edges
1
Well defined edges
51
101
151
201
251
pixel number
Phantom motion pattern vs. MLC motion pattern (recorded every 50 ms)
Encoder
reading
1
0.5
0
0
: Phantom
2
4
encoder reading
6
: MLC
8
10Time(s)
motor encoder readings (averaged)
21
8/2/2012
Dose (cGy)
Modified cosine pattern - penumbra comparisons
300
250
200
150
Non-moving
Moving w/o tracking
Tracking w/o prediction
Tracking w/ prediction
100
50
0
1
51
101
151
201
251
301number
pixel
- Physical target dimension : 60 mm
50 % profile
Non-moving
62.6 mm
20/80 % profile
(avg. both side)
6.6 mm
Moving without tracking
63.6 mm
34.1 mm
Tracking without prediction
61.9 mm
15.8 mm
Tracking with prediction
62.2 mm
8.6 mm
Advantages:
Rotating Bi-planar Design
PERPENDICULAR
TRANSVERSE
PARALLEL
LONGITUDINAL
IN-LINE
C. Kirkby, B. Murray, S. Rathee, B.G. Fallone. Lung dosimetry in a linac-MRI radiotherapy unit with
longitudinal magneticfield, Med. Phys. 37(9), 4722-4732 (2010).
MAJOR ADVANTAGES
Parallel, longitudinal, inline
1) SIMPLIFIED MAGNETIC SHIELDING
2) IMPROVED INTERNAL PATIENT DOSIMETRY
22
8/2/2012
Dosimetric Validation
Achieved
40x40
cm2
20x20
cm2
10x10
cm2
5x5 cm2
Beam Loss
Transverse
Magnetic
Field Effects:
Beam Loss
Longitudinal
Magnetic shielding
• Transverse magnetic fields:
– 20% beam loss at ~ 4 G (homogeneous field)
• Longitudinal magnetic fields:
– 20% beam loss at ~ 120 G (homogeneous field)
• 0.5 T open MRI
– No magnetic shielding required
• Given a 20% beam loss constraint
J. St.Aubin*, S. Steciw, B.G. Fallone, Effect of transverse magnetic fields on a simulated in-line 6
MV linac, Physics in Medicine and Biology, 55(16), pp. 4861-4869(2010).
J. St-Aubin*, S. Steciw, D-M Santos, B.G. Fallone, Effect of longitudinal magnetic fields on a
simulated 6 MV linac, Medical Physics, 37(9), pp. 4916-4923 (2010).
D-M Santos, J. St-Aubin, B.G. Fallone, S. Steciw, Magnetic shielding investigation for a 6 MV in-line
linac within the parallel configuration of a linac-MR system, Medical Physics 39(2): 788-797 (2012)
23
8/2/2012
MAJOR ADVANTAGES
Parallel, longitudinal, inline
1) SIMPLIFIED MAGNETIC SHIELDING
2) IMPROVED INTERNAL PATIENT DOSIMETRY
Perpendicular vs. Parallel
In the parallel configuration, the effect of the magnetic field is
to somewhat collimate the shower of secondary electrons,
thereby reducing the beam penumbra and focusing the dose
depositing by a given beam.
Dosimetry at AP
Axial and Sagittal (0.5 )
24
8/2/2012
5 field Lung Plan
Eclipse TPS
Relative Difference (0.2 T)
Relative Difference (0.5 T)
25
8/2/2012
1.5 T
FC geometry
Cylindrical
MRs
Parallel Configuration:
Lower penumbra
C. Kirkby, B. Murray, S. Rathee, B.G. Fallone. Lung dosimetry in a linac-MRI radiotherapy
unit with longitudinal magneticfield, Med. Phys. 37(9), 4722-4732 (2010).
Parallel Configuration:
PTV-Dose increase per Bo
POTENTIALLY USED TO DECREASE DOSE TO
NORMAL TISSUE WITH SAME DOSE TO PTV
26
8/2/2012
Surface Dose:
Longitudinal
Schematic diagram of the
longitudinal linac-MR
system with the isocenter
at 126 cm.
The CAX magnetic field
maps of our realistic 3-D
models versus
a 1-D model
ENTRANCE
SURFACE SKIN DOSE
10 x10 cm2 photon beam ;
Phantom surface at 21.5 cm air gap
ENTRY SURFACE DOSE
AIR GAP
between surface and pole piece
27
8/2/2012
EXIT SURFACE DOSE
CAX
10 x10 cm2 photon beam;
phantom surface at 136 cm from source
Parallel- Perpendicular
0.5 T MR – 6 MV linac
longitudinal, parallel, inline
transverse, perpendicular
Areas of Active Research
Dosimetry in a Magnetic Field
Earth’s Magnetic Field -Homogeneity of the Rotating MR Field
Magnetic Susceptibility Effects - Subject Rotation wrt magnet
Optimization: MR imagimg affected by Linac RF Noise
Optimization: Magnetic Shielding of the Linac from the MR
Optimization of Magnet Shape, Size and Configuration
Shielding the MR from the Linac
MR Simulation for Radiation Treatment Planning
Real-Time MR-based Tumour Tracking
Fast Pulse Sequence Development at Low Field
Image Characteristics & hyperpolarized gas for lungs at low field
Pre-commercialization Studies
28
8/2/2012
The TEAM
B.G. Fallone,
B. Murray,
S. Rathee
E.Blosser, S. Steciw, K. Wiechowicz, N. deZanche, C. Field
C. Kirkby, M. Mackenzie, D. Robinson, B. Warkentin
B. Burke, J. St-Aubin, A. Ghila, D. Santos, E. Yip, D. Baillie,
J. Yun, M. Reynolds, T. Tadic
R. Pearcey, R. Urtasan (Radiation Oncologists Leads)
Acknowledgement
FUNDED by
Alberta Cancer Foundation
Alberta Cancer Board – Alberta Health Services
Western Economic Diversification, Canada
Alberta Advanced Education & Technology
Alberta Innovates
Natural Science & Engineering , Canada (NSERC)
Canadian Institutes Health Research(CIHR)
(ART)2
linac-MR.ca
29