Targeting based on biological
imaging
Robert Jeraj, Stephen Bowen, Michael Deveau,
Paulina Galavis, Ngoneh Jallow
Department of Medical Physics
University of Wisconsin – Madison, WI
rjeraj@wisc.edu
Back to the DARK AGES
Biological/molecular imaging
 “Visualization, characterization and measurement of
biological processes at the molecular and cellular levels
in humans and other living systems”
 It can probe molecular abnormalities that are the basis of
disease rather than to image the end effects
 Enabled by developments in cell and molecular biology, with
better understanding of the cellular function and disease
changes
 Various other “definitions” of molecular imaging exist
(e.g., small animal imaging, PET imaging)
Current state of affairs…
Different levels of trust into imaging
are the main reason for variability
Hong and Harari, 2005
What is the real tumor extent?
Daisne et al 2004, Radiology 233, 93.
Different modalities – different answer
Daisne et al 2004, Radiology 233, 93.
Same modality - different answers
256, 2 iterations
6 mm filter width
128, 2 iterations
3mm filter width
128, 2 iterations
6 mm filter width
128,4 iterations
6 mm filter width
3D ITER
256, 2 iterations
3 mm filter width
SUV
10
2D OSEM
5
0
256, 2 iterations
3 mm filter width
256, 2 iterations
6 mm filter width
SUV measure
SUVmax
SUVmean
128, 2 iterations
3mm filter width
CV (%)
9
5
128, 2 iterations
6 mm filter width
128,4 iterations
6 mm filter width
min - max (%)
4 - 15
1-8
This is a BIG, BIG problem!!!
 From QUALITATIVE DIAGNOSTIC IMAGING
(Diagnosis and Staging)…
 …To QUANTITATIVE THERAPEUTIC IMAGING
(Target definition, Treatment assessment)
 Limited experience with imaging in treatment
context, compared to diagnostic (except CT)!
 Dangerous to use diagnostic quality imaging in a
therapeutic context (Qualitative ≠ Quantitative)
Problem: PET imaging uncertainties
 Technical factors
– Relative calibration between PET scanner and dose calibrator
– Residual activity in syringe
– Incorrect synchronization of clocks
– Injection vs calibration time
– Quality of administration
 Physical factors
– Scan acquisition parameters
– Image reconstruction parameters
– Use of contrast agents
 Analytical factors
– Region of interest (ROI) definition
– Image processing
 Biological factors
– Patient positioning
– Patient breathing
– Uptake period
– Blood glucose levels
Jeraj 2010
Boellaard et al 2009, J Nucl Med 50: 11S
PET imaging uncertainties
 Technical factors
– Relative calibration between PET scanner and dose calibrator (10%)
– Residual activity in syringe (5%)
– Incorrect synchronization of clocks (10%)
– Injection vs calibration time (10%)
– Quality of administration (50%)
 Physical factors
– Scan acquisition parameters (15%)
– Image reconstruction parameters (30%)
– Use of contrast agents (15%)
 Analytical factors
– Region of interest (ROI) definition (50%)
– Image processing (25%)
 Biological factors
– Patient positioning (15%)
– Patient breathing (30%)
– Uptake period (15%)
Jeraj 2010
– Blood glucose levels (15%)
Unacceptable imaging uncertainties
for therapy applications!!!
Boellaard et al 2009, J Nucl Med 50: 11S
Target definition using PET
 Manual segmentation by an expert physician
 Auto segmentation
• Thresholding (Erdi 1997, Paulino 2004)
• Gradient-based (Geets 2007)
• Region-growing (Drever 2006)
• Feature-based (Yu 2009)
•…
Manual segmentation
Manual segmentation is
NOT PRECISE and NOT ACCURATE
Cadwel 2000
Auto segmentation:
Number of iterations
Volume Variations (%)
150
Impact of Iteration on target volumes
100
Threshold-based
Gradient-based
Region-growing
50
0
-50
-100
128x128 grid size
3D Acquisition, 6mm PF
2D Acquisition, 5mm PF
B
C
D
E
F
Segmentation Techniques
G
Auto segmentation:
Post-iteration filter width
Volume Variations (%)
Impact of post-reconstruction filter width on target volumes
150
Threshold-based
Gradient-based
Region-growing
100
50
0
-100
128x128 Grid-Size
256x256 Grid-Size
-50
3D Acquisition
2D Acquisition
B
C
D
E
F
G
I
J
K
L
Segmentation Techniques
M
N
Auto segmentation:
Grid size
Impact of grid size on target volumes for 3mm PF
200
Threshold-based
Gradient-based
Region-growing
Volume Variations (%)
150
100
50
Auto segmentation is
PRECISE but NOT ACCURATE
0
-50
-100
3D Acquisition
2D Acquisition
B
C
D
E
F
Segmentation Techniques
G
Margins
30
Max. Margin Axial Plane (mm)
Avg. Margin Axial Plane (mm)
5
4
3
2
1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Patients
25
20
15
10
5
0
1
2
3
4
5
6
7
8
9 10 11 12 13 14
Patients
Margin
plane
Mean ±
SD (mm)
Maximum
(mm)
Avg. Max ±
SD (mm)
Maximum
(mm)
Axial
1.0 ± 0.4
1.8
8.4 ± 6.0
26.0
Coronal
0.5 ± 0.3
1.2
10.6 ± 6.1
26.7
Sagittal
0.5 ± 0.3
1.2
10.2 ± 5.3
22.1
15 16 17 18 19 20
Summary
 Biological imaging is not QUANTITATIVE
– Many national/international efforts focused on
improving image quantification
• IRAT, CQIE, QIBA
– Physicists should carry the main burden/effort to
improve quantification/educate physicians
• Are YOU ready?
 Target segmentation is not ACCURATE/PRECISE
– Manual segmentation is not an option
– Auto-segmentation carries many uncertainties
– Much more work to be done
• AAPM TG211
Traditional approach
 Anatomical imaging - GTV
 Uncertainties:
– Not seen on imaging - CTV
– Motion, positioning - PTV
GTV
CTV
PTV
But, is it optimal?
X
Uniform dose → Uniform response
Uniform dose
Uniform
response
Heterogeneous
response
How can we get uniform response?
Targeted dose ← Heterogeneous response
Targeted dose
Cure
Heterogeneous response
DOSE PAINTING
(Biologically targeted radiotherapy)
Ling et al 2000, Int J Rad Oncol Biol Phys 47: 551
Dose painting
 Anatomical → Biological imaging – BTV
anatomical
functional image
biological
targets
Dose painting
 Anatomical → Biological imaging – BTV
 Uncertainties:
– All uncertainties - PTV
biological
targets
Dose painting workflow
Tumor
biology
1
Biological
imaging
2
Bio-based
prescription
3
Planning &
delivery
4
Clinical
outcome
What are extra uncertainties?
Biological imaging
Tumor
biology
1
Biological
imaging
Microscopy → Macroscopy
Microscopy
1 mm
Courtesy of A. van der Kogel, Nijmegen, NL
Proliferation
Hypoxia
PET/CT imaging
5 cm
Spatial resolution
40 μm
0.5 mm
1 mm
2 mm
We do not see hot heterogeneities
0.5 mm
0.4
0.2
0.0
0
20
40
60
Normalized Frequency
Normalized Frequency
0.6
2 mm
0.8
0.8
0.8
Normalized Frequency
1 mm
0.6
0.4
0.2
0.8
Normalized Frequency
40 μm
0.6
0.4
0.2
0.6
0.4
0.2
Very small localized high activities
can not be visualized
80
100
120
0.0
0
140
20
40
60
80
100
120
0.0
0
140
20
0.8
60
80
100
120
140
0.0
0
20
Pixel Intensity (8 bit)
Pixel Intensity (8 bit)
Pixel Intensity (8 bit)
40
60
80
100
120
140
0.8
0.8
0.8
40
Pixel Intensity (8 bit)
0.5
0.4
0.3
0.2
0.1
0.0
0
20
40
60
80
100
120
Pixel Intensity (8 bit)
140
0.6
0.4
0.2
0.0
0
20
40
60
80
100
120
Pixel Intensity (8 bit)
140
Normalized Frequency
0.6
Normalized Frequency
Normalized Frequency
Normalized Frequency
0.7
0.6
0.4
0.2
0.0
0
20
40
60
80
100
120
Pixel Intensity (8 bit)
140
0.6
0.4
0.2
0.0
0
20
40
60
80
100
120
Pixel Intensity (8 bit)
140
We do not see small heterogeneities
Partial volume effects
Recovery coefficients
Recovery Coefficient
1.2
1.0
Even relatively large
localized
0.8
activities need to be
0.6 corrected for
0.4
0.2
0.0
0
5
10
15
20
25
Sphere Diameter (mm)
30
Spatial distribution of tumor phenotypes
FDG PET/CT
(metabolism)
FLT PET/CT
(proliferation)
Cu-ATSM PET/CT
(hypoxia)
Spatial distribution of tumor phenotypes
FDG PET/CT
(metabolism)
FLT PET/CT
(proliferation)
Cu-ATSM PET/CT
(hypoxia)
How to combine this information?
Proliferation
[18F]FLT
Proliferative and
hypoxic
Proliferative
Hypoxia
[64Cu]Cu-ATSM
Other
Hypoxic
Biologically-based prescription
Tumor
biology
1
Biological
imaging
2
Bio-based
prescription
Tracer retention mechanisms
Cu IIATSM 
Cu IIATSM
CR
FMISO 
NR
O2
H3O+
H2O
REDOX
compartment
BOUND
compartment
RSH
H 2 ATSM  H 2 ATSM
Cu I RS
FRNO 2
DISSOCIATION
compartment
O2
FRNO 2
NR
FRNO
MM
FRNH 2
Bowen et al 2011, Nucl Med Biol
pO2 transformation functions
60
60
Cu-ATSM model
1 SD CI
Cu-ATSM meas.
Lewis et al. 1999
PO2 (mmHg)
50
40
50
40
30
30
20
20
10
10
0
0
1
2
3
4 5
SUV
6
7
FMISO model
1 SD CI
FMISO meas.
Lewis et al. 1999
FMISO meas.
Piert et al. 2000
8
0
1
2
3
4
SUV
5
6
7
8
Prescription function
Pmid = 3 mmHg, OERmax = 1.4
Prescribed Dose (Gy)
Cu-ATSM
92
Dose
pH
= 7.2
pH = 7.1
pH = 7.2
pH = 7.3
88
84
80
76
72
68
64
pH = 7.1
60
pH = 7.3
0
1
2
3
4
5
Cu-ATSM SUV
6
Overall uncertainty in a patient
Prescribed Dose (Gy)
120
Upper Limit
Average
Lower Limit
110
100
90
80
70
60
0
1
2
3
4
5
6
Cu-ATSM SUV
 Mean uncertainty of 20 % (max 60 %) in prescribed
dose to individual patient
Uncertainties in population
Symbol
Parameter
pH
Intracellular Acidity
HP2.5 Fit
Dose Boost vs. Hypoxic
Proportion Function
Pmid
OERmax
Range
Dose Uncertainty
Mean (Max)
7.1 – 7.3
4 % (10 %)
95 % CI
5 % (14 %)
(Gerweck 1998)
Translating biological imaging
uncertainty
to dose2 –is
uncertain
Half-max Sensitization pO
5 mmHg
1 % (2 %)
2
Max Oxygen Enhancement
Ratio
(Nilsson 2002)
1.4 – 3.0
1 % (2 %)
Overall
10 % (17 %)
Patient
20 % (60 %)
(Chan 2008)
Extraction of biological information
1 min
15 min
Tracer pharmacokinetics can lead to
large uncertainties
60 min
FLT PET/CT
Planning and delivery
Tumor
biology
1
Biological
imaging
2
Bio-based
prescription
3
Planning &
delivery
Setup uncertainties
Axial shift
Radial shift
Setup uncertainties
120
0.9 [0.7-1.0]
80
40
0
0.6
0.7
0.8
0.9
1
120
0.8 [0.6-1.0]
Even small
setup
errors
will
lead
to
Both
significantly different appearance of
tumor heterogeneity
80
40
0
0.6
120
0.7
0.8
0.9
1
0.9 [0.8-1.0]
80
40
0
0.6
0.7
0.8
0.9
1
Impact of motion uncertainties
 Quantification errors due to respiration motion
– Blurring of recovered activity – loss of spatial and
functional info
– Delineation errors – what/where are the targets ?
“Motion-free”
Motion-blurred
ROI50%
ROI50% delineated in
motion-blurred image
ROI50% delineated in
motion-free image
Motion uncertainties
Threshold
Dose (IPET)
-30
(Dose) > 3%
-25
(V50%) [%]
Continuous
Dose (IGated)
-20
Motion can have a detrimental
impact on accuracy
95 Gy
≥ +3%
≤ -3%
60
-15
-10
-5
0
0
5
10
15
20
Amplitude of motion [mm]
Clinical outcome correlation
Tumor
biology
1
Biological
imaging
2
Bio-based
prescription
3
Planning &
delivery
4
Clinical
outcome
Dose vs. PET for recurrence
Recurrence at Time Point
No Recurrence at Time Point
Upper 95 % C.I.
Empirical fit
Lower 95 % C.I.
Real1 clinical
Dose
world is not perfect
Dose 2
Dose 3
PET Uptake
Summary
Tumor
biology
1
Biological
imaging
2
Bio-based
prescription
3
Planning &
delivery
Micro→Macro
Tracer
Set-up
Which biology
Extraction
of biology
Motion
4
Clinical
outcome
Outcome
uncertainties
Thanks to:


Image-guided therapy group
– Vikram Adhikarla
– Steve Bowen
– Tyler Bradshaw
– Michael Deveau
– Joseph Grudzinski
– Ngoneh Jallow
– Matt La Fontaine
– Paulina Galavis
– Keisha McCall
– Courtney Morrison
– Matt Nyflott
– Surendra Prajapati
– Urban Simoncic
– Peter Scully
– Chihwa Song
– Benny Titz
– Matt Vanderhoek
– Stephen Yip
– Former students…
Funding
– NIH, NSF, UWCCC, Pfizer, Amgen,
AstraZeneca, EntreMed

Human Oncology
– Søren Bentzen
– Paul Harari
– Mark Ritter

Medical Oncology/Hematology
– Glenn Liu
– George Wilding
– Mark Juckett

Radiology
– Scott Perlman
– Chris Jaskowiak

Medical Physics
– Jerry Nickles
– Onofre DeJesus
– Todd Barnhard
– Dhanabalan Murali
– Rock Mackie

Veterinary School
– Lisa Forrest
– David Vail

Phase I Office

External collaborators
– Albert van der Kogel (Nijmegen)