Supplementary material
for the paper
Combined proton-photon treatments - a new approach to proton therapy without a gantry
A. Treatment plan optimization details
In this section, we detail the treatment plan optimization for the three clinical cases with
malignancies of the nasopharynx, oropharynx, and larynx. We first consider the optimization of
the single-modality IMRT and IMPT plans. Combined proton-photon treatments are obtained by
simultaneously optimizing IMRT and IMPT plans based on their cumulative physical dose. Solutions
for all the optimization problems are found by using our own implementation of the L-BFGS quasi-
Newton method [1], together with an augmented Lagrangian method for handling constraints [2].
Dose influences matrices for photons are calculated with the open-source radiotherapy planning
research platform CERR [3], assuming a beamlet resolution of 5x5 mm2. Dose calculations for
protons are performed with the open-source radiation treatment planning toolkit matRad [4],
assuming a lateral spot spacing of 5 mm. The pencil beam sizes that are used represent a generic
proton machine. Assuming a Gaussian lateral profile, the initial sigma value at the patient surface
ranges from 5 to 2.3 mm for energies of 31.7-236.1 MeV.
A.1. Single-modality treatment plans optimization
A1.1. Nasopharyngeal cancer
For the NPC case, we solve optimization problems for the following choice of the objective
function:
f (d ) = 1N T 1
∑iϵT 1
❑
¿¿ (A.1
)
+ 1N T 2
∑iϵT 2
❑
¿¿ (A.2
)
+ 1N T 3
∑iϵT 3
❑
¿¿ (A.3
1
1
2
3
4
5
67
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
)
+ 1N R
∑iϵR
❑
(di−d imax)+¿2¿
(A.4
)
+ 1N H
∑iϵH
d i(A.5
)
+ ( 1NP∑iϵP
d ip)1p (A.6
)
+ 1N C
∑iϵC
d i(A.7
)
+ 1N B
∑iϵB
d i(A.8
)
+ 1N PG
∑iϵPG
d i(A.9
)
and constraint function:
d imax≤26 ∀ i∈PG (A.10
)
where T1,2 and 3 denote the set of voxels in the three PTVs, R is the set of voxels in a 1cm
rim of normal tissues surrounding the PTV3; P, C, B and PG denote the set of voxels in the
PCM, cochlea, brainstem and parotid glands, and H is the set of voxels in the remaining
normal tissues. The exponent in the gEUD objective (Eq. A.6) for the PCM is set to p=5.
The maximum dose d imax in the conformity objective (Eq. A.4) is given by:
d imax≤54−zi(54−27) (A.11
)
where z i is the euclidian distance of a normal tissue voxel i from the PTV3 contour.
A1.2. Oropharyngeal cancer
For the OPC case, a complete list of the objective and constraint functions is as follows.
A dose of 70 Gy is prescribed to the PTV1. A dose exceeding 73.5 Gy is penalized
quadratically.
2
26
27
28
29
30
31
32
33
34
35
36
37
A dose of 60 Gy is prescribed to the PTV2. A dose exceeding 70 Gy is penalized
quadratically.
A dose of 54 Gy is prescribed to the PTV3. A dose exceeding 60 Gy is penalized
quadratically.
The plan is to be conformal. A dose falloff to half the PTV3 prescription at 1 cm from
the PTV3 contour is aimed for.
The gEUD in the PCM is minimized (p=5).
The mean dose to the oral cavity and parotid glands is minimized.
The mean dose in the remaining healthy tissues is minimized.
The maximum mean dose to the union of the parotid glands is constrained to 26 Gy.
For the exact mathematical definition of the planning objectives see equations A.1-A.11.
A1.3. Laryngeal cancer
For the laryngeal cancer case we solve optimization problems for the following choice of
the objective function:
f (d ) = 1N T
∑iϵT
❑
¿¿ (A.12
)
+ 1N R
∑iϵR
❑
(di−d imax)+¿2 ¿
(A.13
)
+ 1N H
∑iϵH
d i(A.14
)
+ ( 1NP∑iϵP
d ip)1p (A.15
)
+ ( 1NL∑iϵL
d ip)1p (A.16
)
+ 1N C
∑iϵC
d i(A.17
)
where T, P, L, C and H denotes the set of voxels in the target volume, PCM, larynx,
arytenoid cartilage, and remaining normal tissues; and R is the set of voxels in a 1 cm rim of
3
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
normal tissues surrounding the PTV. The exponents in the gEUD objectives (Eq. A.15 and
A.16) for the PCM and larynx are set to p=5 and p=0.8, respectively. The maximum dose
d imax in the conformity objective (Eq. A.13) is given by:
d imax≤68−zi(68−34) (A.18
)
A.2. Multi-modality treatment plans optimization
For each patient, the optimized combination is obtained by considering the same clinical goals
as for the single-modality IMRT and IMPT plans. A subset of objective functions is minimized
while the remaining ones are constrained to be no worse than in the IMRT plan. Thus, if f i
represents a single objective (e.g. target coverage or dose conformity) and f IMRT denote its
value in the IMRT plan, we add constraints of the form:
f i≤ f IMRT i(A.19
)
This is illustrated for the laryngeal cancer case where the objective is to minimize the weighted
sum of Eq. A.14-A.17 while all the remaining objectives (Eq. A.12 and A.13) are constrained to
their values in the IMRT plan.
B. NTCP models
The NTCP models used in this paper are selected from literature and have in parts been proposed
for model-based patient selection in the Netherlands [5]. These are based on three different
formulas:
1. Lyman-Kutcher-Burman (LKB) model with parameters m, n, and TD50:
NTCP=Φ( gEUD (n )−TD50
m ∙TD50) (B.1)
where Φ is the cumulative distribution function of the standard normal distribution.
2. Logistic regression model (LM) with parameters TD50 and k :
NTCP=(1+(TD50
D )k
)−1 (B.2)
3. Multivariable logistic regression model (MV) with parameters a, b, and c:
4
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
NTCP=(1+ea−b ∙ X 1−c∙ X 2 )−1 (B.3)
Details regarding the different NTCP models are provided in Table B.1.
C. Proton gantry treatments
In this section, the IMPT plans that could be delivered with a gantry will be discussed. When
designing the gantry-based IMPT plans, several beam arrangements were tested and the best one
was chosen. For all the patients, the gantry treatment uses 5 equispaced beams on a transversal
plane (gantry angles of 36°, 108°, 180°, 252°, and 324°). For optimizing the gantry-based IMPT
plans, Table B.1: Details of the NTCP models used in the present study. Considered toxicities, impacted organs, and used NTCP models with corresponding coefficients are listed.
the same objective functions were used as for generating the IMRT and fixed beamline IMPT plans
(Appendix A).
C.1. Nasopharyngeal cancer
5
Toxicity Model type Impacted organ(s) Model coefficients Mean
Xerostomia [6, 7] MV Parotid gland abX1
1.5070.052Dmean controlat parotid
Dysphagia [6, 7] MV Oral mucosa and PCM
abX1cX2
3.3030.024Dmean oral cavity
0.024Dmean sup-PCM
Oral mucositis [8] LM Oral mucosa kTD50 (Gy)
151
Aspiration [9] LKB Larynx TD50 (Gy)nm
46.510.5
LKB PCM TD50 (Gy)nm
64.50.20.09
Laryngeal edema [10] LKB Larynx TD50 (Gy)nm
47.31.170.23
78
79
80
81
82
83
84
858687
89
90
91
92
Figure 2b of the main manuscript compares the DVHs for the IMPT plans with and without a
gantry. The DVHs for the IMRT plan and the optimal combination are also shown. It can be
seen that all plans are similar in terms of target coverage with the main differences observed
in the OARs. The use of a proton gantry improves the sparing of the parotid glands compared
to single-modality fixed beamline IMPT. In fact, if a gantry is used the mean dose can be
reduced from 23 to 14 Gy in the parotid glands and from 32 to 27 Gy in the PCM. On the other
hand, the oral cavity can be spared to a lesser extent due to the use of the anterior beams
(mean dose of 17 Gy). The gantry treatment has a 11% lower risk for xerostomia compared to
the IMRT plan, while no benefit is expected from the IMPT plan with the fixed beam line
(NTCP=0). Since the superior PCM is part of the target volume, both the IMPT plans have no
benefit for the risk of aspiration versus the IMRT plan. Small differences are observed in the
risk assessment of dysphagia for both the IMPT treatments compared to the IMRT plan
(NTCP=4% for the FBL treatment and NTCP=2% for the gantry treatment). The benefit in
oral mucositis versus the IMRT plan is smaller for the gantry treatment than for the FBL
treatment but a moderate NTCP-value reduction is still observed (NTCP=5% for the gantry
treatment and NTCP=12% for the FBL treatment).
C.2. Oropharyngeal cancer
Figure C.1a shows the DVHs for the IMPT plans with and without a gantry as well as for the
optimal combination and the IMRT plan. Note that if a FBL is used, the IMPT plan is suboptimal
for the parotids and PCM whereas the oral cavity can be better spared. The mean dose of 6 Gy
that can be obtained with the gantry treatment in the contralateral parotid gland translates
into a 6% lower risk for xerostomia compared to the IMRT plan. The IMPT plan with a FBL is
instead associated with a higher risk for this toxicity (NTCP=-13%). The benefits for the risk of
dysphagia and oral mucositis versus the IMRT plan are the same for both the IMPT plans
(NTCP=3% for dysphagia and NTCP=4% for oral mucositis). Finally, a NTCP=2% is seen with
regard to aspiration when using a gantry whereas no favourable result is observed with a FBL.
Interestingly, the optimal combination improves on the single-modality plans for the parotids
and oral cavity and achieves most of the integral dose reduction in the PCM that is possible
with the gantry IMPT plan. The target coverage remains similar for all the plans.
C.3. Laryngeal cancer
6
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
Figures C.1b shows the DVHs for the IMPT plans, the IMRT plan, and the optimal combination.
In this case, the IMPT plans deliver similar doses to the larynx. Therefore, both plans are
associated with the same risk for larynx-based aspiration (NTCP=4%) and laryngeal edema
(NTCP=1%) with respect to the IMRT plan. The cartilage and PCM cannot be better spared by
using a gantry. Interestingly, the optimal combination reduces the dose to the larynx and
arytenoid cartilage with respect to the single-modality plans. The PCM can be spared to a
lesser extent, however, the optimal combination maintains most of the integral dose reduction
that is possible with the FBL protons-only plan.
Figure C.1: DVHs compared for the IMPT plans with and without a gantry, the IMRT plan and the optimal combination for: (a) the OPC case; and (b) the laryngeal cancer case.
D. Robustness of combined treatment plans against proton range uncertainties
One of the main concerns in cancer treatment with protons is the uncertainty in the range of a
proton beam within the patient. The dose distribution delivered to the patient may highly degrade
compared to the planned one if the true range of protons is uncertain. This uncertainty may
originate from several sources including the conversion of CT Hounsfield units to stopping power
values [11], artefacts in the CT image [12], delivery uncertainties and changes in the patient’s
geometry. Therefore, in this section, we address the issue of the robustness of combined
treatment plans against proton range uncertainties. Range errors are modelled via three error
scenarios (nominal scenario, range overshoot and undershoot) by uniformly scaling the CT
Hounsfield units by 3.5%, such that all proton pencil beams penetrate further or less within the
patient. The stochastic programming approach [13] is used to incorporate the set of error
scenarios into the planning of combined treatments while safety margins are applied to account
for other potential sources of uncertainty including setup errors. The method minimizes a
7
125
126
127
128
129
130
131
132
133134135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
weighted sum of objective functions for cumulative dose evaluated for all error scenarios.
Formally, the optimization problem is:
minimizex γ , xp
❑∑s
ps f [ (dγ+d p )s ] (D.1)
where the parameter ps represents an importance weight for error scenario s and is chosen as 0.5
for the nominal scenario and 0.25 for the overshoot and undershoot scenarios [14, 15].
Figure D.1a shows the results of a sensitivity analysis of the non-robust combined treatment plan
for the NPC case. The undershoot of protons yields inhomogeneous dose distributions within the
target volumes whereas if the range of protons is larger than expected, higher doses are delivered
to the main OARs. Consequently, the benefit of the combined treatment over the single-modality
IMRT plan is reduced (Table D.1). Figure D.1b shows that proton range uncertainties can be
efficiently accounted for through stochastic optimization, with the robustly optimized combined
plan being less affected by uncertainties. Robustness is achieved by increasing the photon
component as shown in Figure D.2 for the nominal scenario. The price to pay for robustness is a
lower benefit over the single-modality IMRT plan (Table D.1). However, a substantial clinical
benefit remains.
Table D.1: NTCP-value reductions for the nasopharyngeal cancer case using non-robust and robust combined treatment plans versus IMRT plan.
NTCP in %Non-robust plan Robust plan
Toxicity nominal undershoot overshoot nominal undershoot overshoot
Xerostomia 10 13 7 9 11 7Dysphagia 3 3 2 2 2 2Mucositis 9 11 8 6 7 5
Figure D.1: DVHs from the non-robust (a) and robust (b) combined proton-photon treatment plans for the nasopharyngeal cancer case for the nominal condition (no range errors) and for scenarios in which the proton range is underestimated (overshoot) or overestimated (undershoot).
8
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164165166
167
168
Figure D.2: Proton-photon combinations for the nasopharyngeal cancer case. Top row: non-robust plan. Bottom row: robust plan. (a) and (d) photon dose contributions; (b) and (e) proton dose contributions; (c) and (f) cumulative doses. The contours show the high-risk PTV (red), the elective PTV (blue), the parotid glands (green), the mandible (magenta), and the oral cavity (orange).
9
170171172173
174
References
1. Wright SJ and Nocedal J, Numerical Optimization. Vol. 2. 1999, New York: Springer.2. Bersekas DP, Nonlinear programming. 1999: Athena Scientific.3. Deasy JO, Blanco AI, and Clark VH, CERR: a computational environment for radiotherapy
research. Med Phys, 2003. 30: p. 979-985.4. Wieser HP, et al., Development of the open-source dose calculation and optimization toolkit matRad. Med Phys, 2017. 44(6): p. 2556-2568.5. Jakobi A, et al., Identification of patient benefit from proton therapy for advanced head and
neck cancer patients based on individual and subgroup normal tissue complication probability analysis. Int J Radiat Oncol Biol Phys, 2015. 92(5): p. 1165-1174.
6. Langendijk JA, et al., Selection of patients for radiotherapy with protons aiming at reduction of side effects: the model-based approach. Radiother Oncol, 2013. 107: p. 267-273.7. Widder J, et al., The Quest for Evidence for Proton Therapy: Model-Based Approach and
Precision Medicine. Int J Radiat Oncol Biol Phys, 2016. 95(1): p. 30-36.8. Bhide SA, et al., Dose-response analysis of acute oral mucositis and pharyngeal dysphagia in patients receiving induction chemotherapy followed by concomitant chemo-IMRT for head
and neck. Radiother Oncol, 2012. 103: p. 88-91.9. Eisbruch A, et al., Chemo-IMRT of oropharyngeal cancer aiming to reduce dysphagia:
swallowing organs late complication probabilities and dosimetric correlates. Int J Radiat Oncol Biol Phys, 2011. 81(3): p. 93-99.
10. Rancati T, Fiorino C, and Sanguineti G, NTCP modeling of subacute/late laryngeal edema scored by fiberoptic examination. Int J Radiat Oncol Biol Phys, 2009. 75(3): p. 915-923.
11. Schaffner B and Pedroni E, The precision of proton range calculations in proton radiotherapy treatment planning: experimental verification of the relation between CT-HU and proton stopping power. Phys Med Biol, 1998. 43(6): p. 1579-1592.12. Jäkel O and Reiss P, The influence of metal artefacts on the range of ion beams. Phys Med
Biol, 2007. 52(3): p. 635-644.13. Unkelbach J, Chan TCY, and Bortfeld T, Accounting for range uncertainties in the optimization of intensity modulatec proton therapy. Phys Med Biol, 2007. 52: p. 2755-2773.14. Unkelbach J and Paganetti H, Robust proton treatment planning: physical and biological
optimization. Semin Radiat Oncol, 2018. 28: p. 86-96.15. Unkelbach J, et al., Robust radiotherapy planning. Phys Med Biol, 2018. 63: p. 28pp.
10
175
176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207
208
209
210
211
212
213
214
Top Related