Received: 20 June 2019 | Revised: 5 September 2019 | Accepted: 14 October 2019

DOI: 10.1002/acm2.12767

RADIATION ONCOLOGY PHYSICS

Automated planning of whole breast irradiation using hybrid

IMRT improves efficiency and quality

Bingqi Guo | Chirag Shah | Ping Xia

Department of Radiation Oncology, Taussig

Cancer Institute, Cleveland Clinic, Abstract
Cleveland, OH, USA Purpose: To develop an automated workflow for whole breast irradiation treatment
Author to whom correspondence should be planning using hybrid intensity modulated radiation therapy (IMRT) approach and to
addressed. Bingqi Guo demonstrate that this workflow can improve planning quality and efficiency when
E‐mail: guob@[Link]; Telephone: 216‐445‐
1831. compared to manual planning.
Methods: The auto planning framework was built based on scripting with MIM and
Pinnacle systems. MIM workflows were developed to automatically segment normal
structures and targets, identify landmarks for beam placement, select beam energies,
and set beam configurations. Pinnacle scripts were generated from the MIM work-
flow to create hybrid IMRT plans automatically. Each hybrid IMRT plan included two
prescriptions: a three‐dimensional (3D) prescription consisted of two open tangent
beams, and an IMRT prescription consisted of two step‐and‐shoot IMRT beams. The
3D prescription delivered a full prescription dose to the maximum dose point, and
the IMRT prescription was optimized to deliver a uniform dose to the entire breast
while sparing dose to the normal structures. For 30 patients, the auto plans were
compared with clinically accepted manual plans using the paired sample t‐test.
Results: The auto planning process took approximately 8 min to complete. The
mean dice coefficients between auto‐segmentation and manual contours were 0.98,
0.94 and 0.88 for the lungs, heart, and PTVeval_Breast, respectively. The MUs of
the auto plans was on average 13% higher than that of the manual plans. Auto plan-
ning improved plan quality significantly: percentage volume receiving 95% of the
prescription dose (V95%) of the PTVeval_Breast increased from 91.5% to 93.2%
(P = 0.001), V105% of the PTVeval_Breast decreased from 7.2% to 1.2%
(P = 0.013), V20Gy of the ipsilateral lung decreased from 13.1% to 10.4%
(P = 0.001) and mean heart dose for left‐sided breast patients decreased from
1.2 Gy to 0.9 Gy (P < 0.001).
Conclusion: An automated treatment planning process can make the planning pro-
cess efficient with improved plan quality.

KEY WORDS
auto‐segmentation, auto planning, breast IMRT, whole breast irradiation

----------------------------------------------------------------------------------------------------------------------------------------------------------------------
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2019 The Authors. Journal of Applied Clinical Medical Physics published by Wiley Periodicals, Inc. on behalf of American Association of Physicists in Medicine.

J Appl Clin Med Phys 2019; [Link] 87–96 [Link]/journal/acm2 | 87

 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
88 | GUO ET AL.

1 | INTRODUCTION
2 | MATERIALS AND METHODS
Whole breast irradiation (WBI) using tangential fields is an estab-
2.A | Patients
lished technique for adjuvant radiation therapy as part of breast‐
conserving treatment of early‐stage breast cancer.1 The planning This institutional review board (IRB) approved retrospective study
goals of WBI include delivering a uniform dose to the breast tis- included 30 patients randomly selected from patients treated at our
sue while minimizing dose to the lungs and the heart. Based on institution with tangential field‐in‐field WBI between November
patient‐specific anatomy, planners manually select tangent beam 2016 and November 2018. In this cohort, 10 patients were left
configuration (gantry angles and collimator angles) and beam ener- breast cancer treated with deep inspiration breath hold (DIBH), 10
gies, design the beam apertures (including segments), and set patients were left breast treated without DIBH, and ten patients
beam weights through a time‐consuming trial and error approach. were right breast cancer treated. The clinical plans for these patients
Planning time was on average 39 min (range: 15–70 min) in a 20 used 6/10 MV photon with four standard fraction and 26 hypofrac-
patient study.2 Several auto and semi‐auto planning techniques tionated schemes. All patients were planned using Pinnacle treat-
have been developed to improve the efficiency of each step of ment planning system, version 9.10 (Philips Healthcare, Fitchburg,
WBI treatment planning. These steps included automatically seg- WI) and treated on Truebeam Linear accelerator (Varian Medical Sys-
3–6
menting the treatment targets and normal structures, automati- tems, Palo Alto, CA).
cally selecting beam angles,7 optimizing wedge angles for tangent
beams,8 optimizing segment shapes and weights,3,9,10 and automat-
2.B | Auto planning workflow
ically creating an inverse planned intensity modulated radiation
therapy (IMRT) plan.4,11,12 For this study, all patients were auto planned using the same data-
Automation can reduce planning time with comparable or better set, treatment machine, isocenter, and prescription dose as the cor-
plan quality. Zhao et al. proved that automatic beam angle selection responding clinical plans. Figure 1 shows the auto planning
reduced the volume of heart receiving 5 Gy and volume of ipsilateral workflow. It consists of the following steps:
lung receiving 10 Gy.7 Mitchell et al. used auto‐segmentation soft-
a Auto‐segmentation of volumes of interests and beam boundary
ware to contour critical structures and scripting in the treatment
points
planning system to generate beam segments from isodose lines and
optimize segment weights.3 They concluded that auto contours
Computed tomography images were sent to the MIM system
agreed closely with clinician delineation and scripted treatment plans
(MIM Software Inc., Cleveland, OH, USA). An assistant rule had been
demonstrated equivalence with their clinical counterparts with mod-
set up in MIM to automatically perform atlas‐based segmentation for
est deduction in planning time. Purdie et al. developed a fully auto-
normal structures including bilateral lungs, heart, and spinal cord
matic planning technique for tangent step‐and‐shoot IMRT.4 This
using an MIM workflow and a breast atlas, which was developed
technique used radio‐opaque markers placed at CT simulation to
based on 20 patients outside of this study cohort.
determine the beam geometry and generate whole breast volume
Physicians reviewed the CT images to determine the extent of
for inverse IMRT optimization. For the 158 patients studied, the
breast tissue to be treated based on wires placed at simulation and
mean planning time was 6.8 min. Ninety‐nine percent of the auto
clinical judgment. They had the option to redefine the medial and
plans were deemed clinically acceptable, and 87% were deemed clin-
lateral boundaries (using a contour named “box”) and the superior
ically improved or equal to manual plans. Purdie’s method was
and inferior boundaries (using a contour named “borders”). Another
applied clinically to over 1600 patients and was shown to reduce
MIM workflow was developed to detect four boundary points: med-
plan rejection rates.8
ial, lateral, superior and inferior based on the “box” and “borders”
We adapted Purdie's method to our clinical practice and
contours if they existed. Otherwise, the wires were automatically
improved the workflow to overcome some limitations of the origi-
detected and use for boundary placement. In addition to the bound-
nal technique: (a) we allow flexibility in determining treatment
ary points, two chest wall points, one at the chest wall & superior
boundaries and do not require wiring the patient in a specific way;
boundary and one at the chest wall & inferior boundary of the treat-
(b) we allow the use of mixed energies and perform beam weight
ment region were placed by the MIM workflow. The boundary
optimization automatically; (c) we automate and standardize the
points and chest wall points were used to set up beams. The middle
use of heart and lung blocks; and (d) we use a new hybrid IMRT
panel in Fig. 1(a) illustrates how to define the boundary and chest
technique which can maximize the weight of open beam to
wall points from contours and CT.
improve delivery efficiency and robustness. In this study, we
Once beam boundaries were defined, the MIM workflow contin-
described our auto planning workflow and compared plans created
ued to segment the target volumes following the definitions of
with this workflow with clinical plans for volume delineation, beam
RTOG 1005 (Table 1).13
arrangement, planning parameters, plan quality and delivery robust-
ness. b Beam placement and beam parameters optimization
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
GUO ET AL. | 89

(a)

(b)

(c)

F I G . 1 . Auto planning workflow. (a) Auto‐segmentation: normal structures including bilateral lungs, heart, and spinal cord were contoured
using atlas‐based segmentation; the superior, inferior, lateral and medical boundary landmarks were identified from the “box” and “border”
contours or the wires placed on the patient skin. Two chest wall contours were identified from the computed tomography; targets were
segmented following RTOG 1005 recommendations. (b) Beam placement: two tangent beams were placed based on the boundary points and
chest wall points. The gantry angle, collimator angle, and jaw positions were then optimized to maximize target coverage and minimize normal
lung and heart volume in the beam. Beam energy was selected based on the maximal separation; (c) Hybrid IMRT: the automatic breast plan
includes two prescriptions: a three‐dimensional (3D) prescription with two static tangent beams and an intensity modulated radiation therapy
(IMRT) prescription with two step‐and‐shoot tangent beams. The beam weightings of the 3D prescription were optimized using dose points
selected uniformly inside PTVeval_breast. The 3D prescription delivers full prescription dose to the maximum dose. The IMRT prescription was
optimized to deliver uniform dose to breast and reduce dose to lungs and heart.

Two tangent beams were created automatically using a Matlab points. For each gantry/collimator angle combination, the beam’s
(the MathWorks, Inc., Natick, Massachusetts, USA) script embedded posterior jaw (X1) position was set to cover 95% of the PTVeval_-
in the MIM workflow. The gantry angle of the medial beam was set breast and 100% of the tumor bed with 5 mm margin. Both lateral
so that the medial and lateral boundary points overlapped each other and medial beams were restricted not to cross the midline of the
in the beam's eye view. The collimator angle was set so that X jaws patient. The beam's superior (Y2) and inferior (Y1) jaw positions were
of the beam were parallel with the two chest wall points. The posi- set to cover the entire PTVeval_Breast plus 5mm and the superior/
tion of the posterior jaw (X1) was placed at the medial and lateral inferior boundary points, whichever was larger. The volume of the
boundary points, the anterior jaw (X2) covered the entire breast vol- heart and lungs inside the beam was calculated and an objective
ume with 2 cm skin flash, and the superior (Y2) and inferior (Y1) jaw function defined in Eq. (1) was minimized.
positions were defined at superior and inferior boundary points. The
f ¼ 50  % heart volume in the beam þ % lung volume in the beam (1)
lateral beam was a mirror of the medial beam adjusted for nondiver-
gence posteriorly. The left panel in Fig. 1(b) illustrates how to set up A ratio 50 was arbitrarily chosen to reflect the importance of
beams based on boundary and chest wall points. reducing heart dose. The combination of gantry angle, collimator
The beam angles and jaw positions were further optimized to angle, and jaw positions which gave the smallest f value was chosen.
cover the targets and reduce the lung and heart volumes in the After beam geometry optimization, if more than 10 cc heart vol-
beam. Because the lateral beam matched to the medial beam, only ume were exposed in the beam, a heart block was added to block all
the medial beam was optimized. The gantry angle of the medial or part of the heart without blocking the lumpectomy cavity with a
beam was limited to ±10° from its starting angle defined using the 5 mm margin. For ipsilateral lung and the normal tissue inferior to
boundary points. The collimator angle of the medial beam was also the ipsilateral lung, a lung block was added if it did not block any
limited to ±10° from the starting angle defined by the chest wall part of PTVeval_Breast with 1 cm margin.
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
90 | GUO ET AL.

A patient‐specific pinnacle script was created at the end of the

2.C | Comparing auto and clinical plans
MIM workflow for automatic hybrid IMRT planning.
The auto‐segmented contours were compared with the clinical con-
c Hybrid IMRT planning
tours using the Dice coefficient and mean Hausdorff distance. For a
fair plan comparison, the auto contours were discarded, and the clin-
The hybrid IMRT plan included a 3D prescription and an IMRT
ical contours were used for subsequent beam placement and hybrid
prescription. The 3D prescription was associated with a pair of open
IMRT planning.
medial and lateral beams (with heart and lung blocks if used), and
To compare beam parameters between auto plans and clinical
the IMRT prescription was associated with two step‐and‐shoot
plans, we calculated the absolute differences of gantry and colli-
beams using the same beam geometry as the 3D tangent beams.
mator angles, the difference in the posterior jaw (X1), the inferior
The energy of the beams was defined based on the maximum tan-
jaw (Y1) and the superior jaw (Y2). The anterior jaw X2 covered
gent separation, same as our clinical practice. If the maximum sepa-
the entire breast with >=2 cm margin in both clinical and auto
ration of the tangent beams was less than 20 cm, 6 MV was used
plans. The auto placed beams were used for hybrid IMRT
for both 3D and IMRT prescriptions; for separation great than
planning.
23.5 cm, 10 MV was used for both 3D and IMRT prescriptions; for
To compare the plan quality between auto plans (with clinician
separation between 20 and 23.5 cm, 6 MV was used for the 3D pre-
derived volumes) and clinical plans, target coverage PTVeval_Breast
scription, and 10 MV was used for the IMRT prescription. If the
V95 and PTVeval_Lumpectomy V95, high dose volume PTVeval_-
tumor bed volume was within 5 mm from the skin, 6X was used for
Breast V105, ipsilateral lung receiving 20 and 5 Gy and heart mean
the 3D prescription regardless of the tangent separation. The 3D
dose were compared. Statistical analysis was performed using two‐
prescription delivered a full prescription dose to the maximum dose
sided, paired sample t‐test in Excel with significance defined as
point. The weights of the medial and lateral 3D beams were initially
P < 0.05.
set to be equal and then optimized (beam weight optimization) based
on dose points uniformly placed within the breast tissue. IMRT pre-
scription was optimized to increase homogeneity and reduce dose to
3 | RESULTS
the heart and lungs. Table 2 lists the optimization criteria. The IMRT
plan allowed a maximum of 20 segments (10 segments for breath
The entire auto planning workflow took approximately 8 min, of
hold plans to reduce treatment time). Minimum segment area was
which auto‐segmentation took about 3 min (without manual editing),
set to 10 cm2 to force the use of large segments. The “use current
beam placement took about 2 min, and hybrid IMRT planning took
jaw as max” feature was checked so that the IMRT segments did not
about 3 min.
exceed the edges of open beams.

T A B L E 1 Target volume generation following Section 2.A and 2.B 3.A | Auto‐segmentation
of RTOG 1005.
The auto‐segmentation achieved good agreement with the clinical
Target volume Definition
contours. Figure 2 shows an example. Table 3 lists the dice coeffi-
GTV_Lumpectomy Drawn by physicians
cients and mean Hausdorff distances between auto contours and
CTV_Lumpectomy Expand GTV_Lumpectomy by 1 cm in all
clinical contours. The Dice coefficients were more than 0.9, and the
directions, limited posteriorly to the pectoralis
major, anterolaterally to 5 mm from the skin
mean Hausdorff distances were within 1.5 mm for all normal struc-
and medially to the midline tures. No manual editing was necessary for the contours of the lungs
PTV_Lumpectomy Expand CTV_Lumpectomy by 7 mm in all or spinal cord while only small edits were needed for the heart con-
directions, excluding the heart tours. For targets, dice coefficients of 0.84 and 0.88 were achieved
PTVeval_Lumpectomy PTV_Lumpectomy within ipsilateral breast with CTV_breast and PTVeval_breast, respectively.
tissue and 5 mm from the skin
CTV_Breast Drawn by physicians, OR, Breast tissue,
segmented based on the HU threshold, within T A B L E 2 Intensity modulated ratiation therapy (IMRT) optimization
the volume defined by boundary points criteria for hybrid IMRT planning.
PTV_Breast Expand CTV_breast by 7 mm in all directions, Volume Optimization criteria Weight
excluding the heart and not crossing the PTVeval_Breastopt Receive a uniform dose of 100% 50
midline prescription
PTVeval_Breast PTV_breast limited posteriorly to ribs and Maximum dose < 105% of 100
anterolaterally to 5 mm from the skin prescription
PTVeval_Breastopt The intersection of PTVeval_Breast and the PTVeval_Lumpectomy Minimum dose> 100% prescription 100
volume under beam contracted 5mm
uniformly. This volume was used for Lungs V40% < 15% 5
optimization only Heart Max EUD (α = 1) < 1 Gy 1
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
GUO ET AL. | 91

(a)

(b)

F I G . 2 . Comparing auto contours

(bottom figure) with clinical contours (top
figure) for case #1.

T A B L E 3 Dice coefficients and mean Hausdorff distances between auto‐segmentation and clinical contours for all patients. (mean ± standard
deviation).
Left lung Right lung Heart Spinal cord CTV_Breast PTVeval_Breast
Dice coefficient 0.97 ± 0.01 0.98 ± 0.01 0.94 ± 0.03 0.91 ± 0.03 0.84 ± 0.05 0.88 ± 0.05
Mean Hausdorff 0.7 ± 0.2 0.7 ± 0.2 1.5 ± 0.8 0.5 ± 0.2 4.4 ± 1.6 3.5 ± 1.7
distance/mm

changes in gantry (3.6° on average) and collimator (4.5° on average)

3.B | Auto plan VS clinical plan
angles. Jaw positions were the same (average difference <2 mm).
Figure 3 and Table 4 compares plan parameters (beam geometry, MU of auto plans increased by 13% compared with clinical plans.
energy, and MUs) and plan quality dose volume histogram [(DVH The increased MU was a result of increased intensity modulation.
and isodose)] for case #1. The auto plan reproduced the clinical While open beam MUs were almost the same between auto plans
beam set up with slight changes of the gantry (2°) and collimator and the clinical plans, percentage of MU delivered by open beam
(4°) angles. Plan monitor unit (MU) increased by 6% as a result of decreased from 88% of clinical plans to 76% of auto plans.
increased intensity modulation. The angle optimization and inverse
planned intensity modulation improved the uniformity of dose in the
breast and reduced the dose to ipsilateral lung and heart without 4 | DISCUSSION
compromising the target coverage.
For all patients, Fig. 4 compares PTVeval_Breast V95, Forward planned, field‐in‐field IMRT is a widely used technique for
PTVeval_Lumpectomy V95, PTVeval_Breast V105, ipsilateral Lung whole breast irradiation and has been shown to improve cosmetic
V20, V5 and Heart Dmean between auto plans and clinical plans. results and reduce acute side effects.14,15 The treatment planning
On average, auto plan increased the coverage of PTVeval_Breast process of field‐in‐field technique is time‐consuming and user depen-
V95 from 91.5% to 93.2% (P = 0.001), PTVeval_lumpectomy V95 dent. Auto and semi‐auto techniques have been developed to emu-
stayed the same (97.1% vs 97.6%, P = 0.110), hotspot in breast late all or parts of the manual planning process, improving planning
PTVeval_Breast V105 decreased from 7.2% to 1.2% (P = 0.013), efficiency and reproducibility.
ipsilateral lung V20 and V5 decreased from 13.1% to 10.4% We adapted and improved previously published auto planning
(P = 0.001) and from 22.4% to 18.7% (P < 0.001) respectively, and techniques to our clinical practice and developed a fully automatic
heart mean dose for left‐sided breast patients decreased from 1.2 to workflow for whole breast irradiation using tangent hybrid IMRT.
0.9 Gy (P < 0.001). Compared with clinical plans, auto plans The auto planning technique kept all the desirable features of man-
improved target coverage, dose uniformity and reduced dose to lung ual planning including tangent beam setup, field‐in‐field techniques
and heart. with the majority of dose delivery by open beams, skin flash, ability
Table 5 compares the plan parameters between clinical plans and to use mixed energies, beam weight optimization and heart and lung
auto plans. Auto plans reproduced the clinical beam setup with slight blocks. In hybrid IMRT planning, prescribing to max point dose in
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
92 | GUO ET AL.

the 3D prescription allowed the maximum percentage of dose to be planning process may increase the costs of treatment (slightly
delivered by open beams while improving robustness and preventing increasing treatment time) and must be weighed against improve-
IMRT segments from being placed at the skin surface. Replacing the ments in treatment planning quality and efficiency. Whether or not
manually created segments with inversely planned IMRT, although insurance companies consider this hybrid plan as an IMRT plan or
increased plan MU slightly, allowed auto plans to significantly 3D plan is beyond the scope of this study.
improve the homogeneity of dose in the breast and reduce dose to In breast IMRT planning, the choice of the maximum number of
lungs and heart. However, the use of IMRT as part of the auto segments is crucial as it affects the plan quality, delivery efficiency,

(a)

(b)

(c)

F I G . 3 . Dose volume histogram (DVH)

and isodose comparisons between the
auto plan and the clinical plan for case #1.
Top figure: DVH comparison; middle
figure: isodose distribution of the clinical
plan; bottom figure: isodose distribution of
the auto plan.
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
GUO ET AL. | 93

T A B L E 4 Comparing plan parameters and plan quality between T A B L E 5 Comparing plan parameters between auto plans and
auto plan and clinical plan for case #1. clinical plans for all patients.
Clinical plan Auto plan Mean ± SD
Prescription 2.67 Gy × 15 Gantry angle/° |auto‐clinical| 3.6 ± 2.6
Energy 10X 10X Collimator angle/°|auto‐clinical| 4.5 ± 2.5
Gantry (medial beam)/° 309 311 X1/cm auto‐clinical −0.2 ± 0.5
Collimator (medial beam)/° 12 8 Y1/cm auto‐clinical −0.1 ± 0.6
X1 (posterior jaw)/cm 4.5 4.5 Y2/cm auto‐clinical 0.1 ± 0.9
Y1 (inferior jaw)/cm 9 9.5 Total MU auto/clinical 1.13 ± 0.08
Y2 (superior jaw)/cm 12 11 Open beam MU auto/clinical 0.97 ± 0.02
Plan MU 312 332 Fraction of MU delivered by open beam Clinical plans: 0.88 ± 0.03
PTVeval_Breast V95/% 93.5 93.3 Auto plans: 0.76 ± 0.05

PTVeval_Lumpectomy V95/% 95.7 96.4

PTVeval_Breast V105/% 17.7 4.7 for free breathing patients, and 10 for breath hold patients. To study
Lt Lung V20/% 14.7 12.1 the impact of segment number, case #1 was planned with 5, 10, 20,
Heart Dmean/Gy 1.0 0.8 and 50 maximum number of segments respectively. Table 6 com-
pares the influence of the maximum number of segments to the plan
and plan robustness to motion. Purdie et al. chose the maximum quality, plan MUs, and beam delivery time. For as few as five seg-
number of segments between 6 and 12 based on the target vol- ments, the auto plan showed better plan quality compared with the
ume.4 We arbitrarily chose 20 as the maximum number of segments manual plan. Increasing the number of segments does not reduce

(a) (b)

(c) (d)

(e)

F I G . 4 . Comparing plan quality between

auto plans and clinical plans for all
patients.
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
94 | GUO ET AL.

lung or heart dose because beam geometry was the determining fac- used for breath hold patients to improve efficiency with a slight
tor for lung and heart sparing. For segment number <20, increasing compromise in dose homogeneity.
the maximum segment number reduced hot spot without increasing Another common question for breast IMRT is whether the use
plan MU. However, although plan MU is almost independent of the of intensity modulated beams will compromise plan robustness and
segment number, more segments increased delivery time. For breath increase sensitivity to the patient motion. To explore this question,
hold patients, shorter delivery time is desirable. Therefore, a maxi- for an example patient, we simulated the delivered dose in the pres-
mum of 20 segments were used for free breathing patients to maxi- ence of motion for the auto plan (with maximal 20 segments) and
mize the dose homogeneity, and a maximum of 10 segments were the clinical plan. Case #1 was treated with active breath hold using

T A B L E 6 The influence of the maximum number of segments to plan quality, MU and delivery efficiency of automatic breast intensity
modulated radiation therapy (IMRT) planning.
PTVeval_ PTVeval_ Ipsi Lung Heart Total Delivery
Breast V95/% Breast V105/% V20/% Dmean/Gy MU time/s
Clinical plan (3 segments) 93.5 17.7 14.7 1.0 312 43
Auto plan (5 segments) 93.1 7.7 12.2 0.8 320 41
Auto plan (10 segments) 93.3 4.7 12.1 0.8 332 57
Auto plan (20 segments) 93.2 2.3 12.1 0.8 328 82
Auto plan (50 segments) 93.2 3.1 12.1 0.8 339 112

(a) (b)

(c) (d)

F I G . 5 . Comparison of beam fluence pattern (top figures) and delivered dose volume histogram (DVH) (bottom figures) with the measured
patient motion for case #1 between the clinical (left figures) and auto (right figures) plans. In the DVH figures, the dashed lines represent the
planned DVH, the shaded regions represent delivered DVHs with the measured patient motion for each treatment fraction, and the solid lines
represent the total delivered DVH of all fractions.
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
GUO ET AL. | 95

T A B L E 7 Comparison of the auto plan using auto contours with for CTV_Breast, lungs, and heart, respectively. This result was com-
the auto plan using clinical contours. parable with previous publications: Eldesoky et al.5 reported mean
Auto plan w dice coefficients of 0.86, 0.97 and 0.92 and Velker et al.6 reported
clinical mean dice coefficients of 0.88, 0.97 and 0.90 for breast, lungs, and
Auto plan w auto contours contours
heart, respectively.
Energy 10X 10X To further evaluate the impact of the auto‐contouring errors on
Gantry/° 313 311 automatic breast treatment planning, for an example patient (case
Collimator/° 10 8 #1), Table 7 compares the plan parameters, dose distribution, and
X1 (posterior jaw)/cm 4.5 4.5 plan quality between the auto plan using auto contours and the auto
Y1 (inferior jaw)/cm 9.7 9.5 plan using clinical contours. For this particular patient, the Dice coef-
Y2 (superior jaw)/cm 11.6 11 ficients between auto contours and clinical contours were 0.80,
Plan MU 326 332 0.98, and 0.92 for CTV_breast, lungs, and heart, respectively.

Dice coefficient of 0.94 Although target and normal structure volumes varied, the plan
90% IDL parameters (beam gantry and collimator angles, jaw sizes, and plan
Dice coefficient of 0.95 MU) were similar between the two plans. The 90% and 50% isodose
50% IDL volumes of the two plans overlapped by 94% and 95% respectively.
Plan quality DVH to DVH to DVH to clinical DVH metrics to targets and normal structures (evaluated using clini-
auto clinical Contours cal contours) were comparable between the two plans. The auto
contours contours
planning technique introduced in this study is robust to the errors of
PTVeval_Breast V95/ 85.8 92.7 93.3
auto‐contouring. Two main reasons may explain the insensitivity of
%
the auto plan to contouring errors: (a) the auto‐contoured normal
PTVeval_Lumpectomy 97.1 97.4 96.4
structures closely matched the clinically approved manual contours
V95/%
(>0.9 dice coefficients); (b) we applied restrictions in beam geometry
PTVeval_Breast V105/ 6.1 5.6 4.7
% optimization: beam gantry and collimator angles cannot deviate from
initial angles by more than 10° and beams were not allowed to cross
Lt Lung V20/% 12.3 12.2 12.1
midline.
Heart Dmean/Gy 0.7 0.8 0.8
It is worth noting that although the plan parameters and dose
distribution were robust to contour variations, DVH metrics were
sensitive to how the targets and organs at risk were contoured. As
ABC (active breathing coordinator, Elekta, Stockholm, Sweden) and shown in Table 7, PTVeval_Breast V95 from the auto plan was lower
monitored with the AlignRT system (VisionRT Ltd., London, UK). The when evaluated with the auto contour than with the clinical contour.
delta of the real‐time patient surface compared with the reference As shown in Fig. 2, the auto‐segmented CTV_Breast and PTVeval_-
skin surface extracted from the CT images was recorded during the Breast was larger than the manual contours. The heart mean dose
treatment. We extracted the delta at the time of the delivery for was slightly lower when evaluated with the auto contour than with
each beam in every fraction and calculated the delivered dose for the clinical contour, again due to smaller auto‐contoured heart than
each fraction. Throughout the treatment course, the delta in vertical, the clinical contour. Therefore, it is still recommended to review and
longitudinal, lateral, roll, yaw and pitch ranged from 0.1–5.9 mm, edit the auto contours for plan evaluation.
−1.6–3.4 mm, −2.4–2.8 mm, −1–1.4°, −1.6–1.3° and −1.0–2.6°
respectively. Figure 5 compares the simulated delivered dose of each
fraction, the total delivered dose of all fractions and the planned 5 | CONCLUSIONS
dose for auto and clinical plans. DVHs to the lumpectomy cavity,
CTV_breast, CTV_Lumpectomy, ipsilateral lung, and heart were com- An automated treatment planning technique was developed for
pared. Both clinical and auto plans were robust to the patient whole breast irradiation using hybrid IMRT. Compared with manual
motion. The auto plan showed slightly better agreement between planning, auto planning improved planning efficiency and plan qual-
planned and delivered total dose to CTV_Lumpectomy and ipsilateral ity. A future study will focus on the assessment of the robustness of
lung. A possible reason may be that auto plan created fluence pat- auto plans with more patient data.
tern smoother than the clinical plan, as shown in Fig. 5.
Auto‐contouring targets and normal structures is a challenging
CONFLICT OF INTEREST
task in radiation therapy. For a fully automatic treatment planning
workflow, the accuracy of auto contours is crucial as it affects beam Dr. Chirag Shah received a research grant from VisionRT, Inc. out-
setup, plan optimization, and plan evaluation. In this study, we com- side of this study. Dr. Ping Xia received a grant from Philips Health-
pared auto contours with clinically approved manual contours for all care outside of this study. Dr. Chirag Shah also receives consultation
patients and reported mean dice coefficients of 0.84, 0.98, and 0.94 fees from Impedimed and Varian Medical Systems.
 15269914, 2019, 12, Downloaded from [Link] Wiley Online Library on [14/02/2025]. See the Terms and Conditions ([Link] on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
96 | GUO ET AL.

REFERENCES 8. Chen GP, Ahunbay E, Li XA. Automated computer optimization for

3D treatment planning of breast irradiation. Med Phys.
1. Fisher B, Anderson S, Bryant J, et al. Twenty‐year follow‐up of a ran- 2008;35:2253–2258.
domized trial comparing total mastectomy, lumpectomy, and lumpec- 9. Zhang G, Jiang Z, Shepard D, Zhang B, Yu C. Direct aperture opti-
tomy plus irradiation for the treatment of invasive breast cancer. N mization of breast IMRT and the dosimetric impact of respiration
Engl J Med. 2002;347:1233–1241. motion. Phys Med Biol. 2006;51:N357–N369.
2. Nguyen SK, Cao F, Ramaseshan R, et al. Template‐based breast IMRT 10. Farace P, Zucca S, Solla I, et al. Planning hybrid intensity modulated
planning for increased workload efficiency. Radiat Oncol. 2013;8:67. radiation therapy for whole‐breast irradiation. Int J Radiat Oncol Biol
3. Mitchell RA, Wai P, Colgan R, Kirby AM, Donovan EM. Improving Phys. 2012;84:e115–e122.
the efficiency of breast radiotherapy treatment planning using a 11. Purdie TG, Dinniwell RE, Fyles A, Sharpe MB. Automation and inten-
semi‐automated approach. J Appl Clin Med Phys. 2017;18:18–24. sity modulated radiation therapy for individualized high‐quality tan-
4. Purdie TG, Dinniwell RE, Letourneau D, Hill C, Sharpe MB. Auto- gent breast treatment plans. Int J Radiat Oncol Biol Phys.
mated planning of tangential breast intensity‐modulated radiotherapy 2014;90:688–695.
using heuristic optimization. Int J Radiat Oncol Biol Phys. 12. Kim H, Kwak J, Jung J, et al. Automated field‐in‐field (FIF) plan
2011;81:575–583. framework combining scripting application programming interface
5. Eldesoky AR, Yates ES, Nyeng TB, et al. Internal and external valida- and user‐executed program for breast forward IMRT. Technol Cancer
tion of an ESTRO delineation guideline ‐ dependent automated seg- Res Treat. 2018;17:1533033818810391.
mentation tool for loco‐regional radiation therapy of early breast 13. RTOG 1005. [Link]
cancer. Radiother Oncol. 2016;121:424–430. [Link]?study=1005&mode=broadcasts&ptxml:id=387
6. Velker VM, Rodrigues GB, Dinniwell R, Hwee J, Louie AV. Creation 14. Pignol JP, Olivotto I, Rakovitch E, et al. A multicenter randomized
of RTOG compliant patient CT‐atlases for automated atlas based trial of breast intensity‐modulated radiation therapy to reduce acute
contouring of local regional breast and high‐risk prostate cancers. radiation dermatitis. J Clin Oncol. 2008;26:2085–2092.
Radiat Oncol. 2013;8:188. 15. Barnett GC, Wilkinson JS, Moody AM, et al. Randomized controlled
7. Zhao X, Kong D, Jozsef G, et al. Automated beam placement for trial of forward‐planned intensity modulated radiotherapy for early
breast radiotherapy using a support vector machine based algorithm. breast cancer: interim results at 2 years. Int J Radiat Oncol Biol Phys.
Med Phys. 2012;39:2536–2543. 2012;82:715–723.