Use of a Deformable Atlas to Identify Cryptic Critical
Structures in the Treatment of Glioblastoma Multiforme
David C. Weksberg1, Stephen D. Bilton1, Eric L. Chang2*
1MD Anderson Cancer Center, Department of Radiation Oncology, University of Texas, Houston, Texas, United States of America, 2Department of Radiation Oncology,
University of Southern California Keck School of Medicine, Norris Cancer Hospital, Los Angeles, California, United States of America
Dose constraints for traditional neural critical structures (e.g. optic chiasm, brain stem) are a standard component of
planning radiation therapy to the central nervous system. Increasingly, investigators are becoming interested in accounting
for the dose delivered to other non-target neural structures (e.g. hippocampi), which are not easily identified on axial
imaging. In this pilot study, a commercially available digital atlas was used to identify cryptic neural structures
(hippocampus, optic radiations, and visual cortices) in 6 patients who received intensity modulated radiation therapy (IMRT)
as part of multimodal management of glioblastoma multiforme (GBM). The patient’s original IMRT plans were re-optimized,
with avoidance parameters for the newly identified critical structures. Re-optimization was able to reduce both mean and
maximum dose to the volumes of interest, with a more pronounced effect for contralateral structures. Mean dose was
reduced by 11% and 3% to contralateral and ipsilateral structures, respectively, with comparable reduction in maximum
dose of 10% and 2%, respectively. Importantly, target coverage was not compromised, with an average change in coverage
of 0.2%. Overall, our results demonstrate the feasibility of incorporating tools for cryptic critical structure identification into
the treatment planning process for GBM.
Citation: Weksberg DC, Bilton SD, Chang EL (2012) Use of a Deformable Atlas to Identify Cryptic Critical Structures in the Treatment of Glioblastoma
Multiforme. PLoS ONE 7(3): e32098. doi:10.1371/journal.pone.0032098
Editor: Kevin Camphausen, NIH, United States of America
Received November 3, 2011; Accepted January 19, 2012; Published March 26, 2012
Copyright: ? 2012 Weksberg et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: No current external funding sources for this study.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: eric.L.firstname.lastname@example.org
Malignant gliomas represent the most common primary central
nervous system (CNS) malignancy in adults, with anaplastic
astrocytoma (AA) and glioblastoma multiforme (GBM) accounting
for 60% of malignant CNS gliomas . Radiation therapy is a
cornerstone of treatment for these neoplasms, and, owing to the
diffusely infiltrative nature of these tumors, large treatment volumes
are often required to deliver optimal treatment. While the prognosis
for high-grade glioma generally remains poor, with the advent of
concurrent temozolamide (TMZ) therapy [2,3,4], and the identifica-
tion of 1p19q loss of heterozygosity [5,6,7] and IDH1 mutations  as
favorable prognostic indicators, there are now subsets of patients for
ability to identify these patients with favorable responses improves,
minimizing iatrogenic neurotoxicity will be of growing importance.
Planning radiation treatment to the CNS involves the
delineation of critical organ at risk structures, such as the optic
chiasm, brainstem, and cochlea, in order to reduce the risk of
treatment toxicity. However, while structures such as the chiasm
are readily identifiable on axial imaging, other structures at
potential risk, such as the optic radiations, hippocampi, or visual
cortices, are not readily segmented using conventional planning
systems. Thus, we sought to demonstrate the feasibility of using a
deformable anatomic digital atlas as a tool to aid in identifying and
contouring these cryptic neural structures. Here we present the
results of a pilot studying applying this technique to six patients
who underwent radiation treatment for GBM.
Materials and Methods
After obtaining institutional IRB approval for retrospective
dosimetric analysis, the treatment plans of 6 patients treated for
high-grade glioma at the University of Texas MD Anderson
Cancer Center between 2008 and 2010 were selected for analysis.
5 patients were treated for primary disease, and 1 patient was
treated for recurrent disease. Patient characteristics are summa-
rized in Table 1. No external funding was received for this study.
Treatment Planning and Target Volumes
Pinnacle treatment planning software (Version 9) with Direct
Machine Parameter Optimization (DMPO) was used for treat-
ment planning. For patients with primary tumors, the patient’s
pre-operative contrast-enhanced MRI was registered with the
treatment planning CT scan obtained at the time of simulation, to
aid in delineation of the clinical target volume (CTV). A high-dose
CTV (CTV1) was identified, comprising the surgical cavity, as well
as the areas of contrast enhancement and areas of FLAIR signal
abnormality deriving from tumor infiltration. CTV1 was uni-
formly expanded by 5 mm to account for set-up uncertainty – this
expansion created a planning target volume (PTV), termed ‘‘Boost
PTV,’’ which was planned to receive 60 Gy in 30 fractions. To
create the low-dose CTV (CTV2), CTV1 was uniformly expanded
by 2 cm, and contours were then edited to exclude air, bone, and
brain parenchyma protected by anatomic barriers. A 5 mm
uniform expansion for set-up uncertainty was performed, with the
PLoS ONE | www.plosone.org1 March 2012 | Volume 7 | Issue 3 | e32098
resultant volume ‘‘PTV50’’ receiving 50 Gy in 30 fractions.
Intensity modulated radiation therapy (IMRT) treatment plans
were designed to deliver a simultaneous integrated boost (2 Gy per
fraction to ‘‘Boost PTV’’ and ,1.66 Gy per fraction to ‘‘PTV50’’).
A representative treatment plan is illustrated in Figure 1. For the 1
patient with recurrent disease (and a history of prior radiation for
his primary disease), gross recurrent disease was contoured, and
expanded 5 mm for set-up uncertainty – the resultant PTV was
planned for 40 Gy in 20 fractions.
Organ at risk (OAR) segmentation
Anatom-E (Houston, TX) is a commercially available deform-
able anatomic digital atlas  which provides axial MRI imaging
with functional annotation of CNS structures validated with
intraoperative neurosurgical data (J. Pagani, personal communi-
cation). We used the Anatom-E atlas to aid in delineation of the
optic radiations, visual cortices, and hippocampi. To accomplish
this, the Pinnacle treatment planning display console was mirrored
to the Anatom-E display. Software image controls allow blending
of MRI images (for ease of registration) and wireframe structural
anatomy (for display of functional areas and white matter tracts),
as shown in Figure 2. The deformable anatomic atlas was
registered to the planning CT and pre-operative MRI images. To
accomplish this, the atlas template was manually deformed on a
slice by slice basis using an affine (linear) transformation, which
allows compensation for variation in brain dimensions and patient
head position, with the surfaces of the cerebral hemispheres
serving as a landmark. The highly constant central sulcus was used
to verify alignment as previously described [10,11]. Selected
structures were contoured based on the atlas guideline (Figure 3).
In 3 patients, ipsilateral OAR structures were impacted by
resection defects or gross disease: patient 1 – gross recurrent
disease abutted ipsilateral OARs with minimal mass effect, patient
4 – the surgical cavity abutted a small section of the optic
radiations anteriorly, patient 6 - resected disease prevented the
contouring of the ipsilateral optic radiations. Contralateral
structures were unaffected in all patients.
IMRT treatments were then re-planned, incorporating the
newly identified volumes as avoidance structures using soft
constraints designed to reduce OAR dose without compromising
tumor coverage. In order to accomplish this, the original beam
arrangement and isocenter were preserved, as were the initial
optimization parameters. Additional optimization structures for
each critical structure were created and set approximately 10–20%
lower than the dose delivered by the original plan. Progressive
rounds of optimization were performed until the optimization
results began to yield plans with inadequate prescription dose
coverage (,95% target volume). At this point, OAR objectives
were relaxed to restore tumor coverage. Overall, the additional
optimization procedures added approximately 2 hours to the
planning time for a given patient.
Re-optimization of the IMRT treatment plans generated new
treatment plans that reduced dose to the identified OARs. A
Table 1. Patient characteristics.
Pt AgeGender Tumor typeLocationGTV PTV
1 63M GBM (recurrent) R temporal lobe123 cc 185 cc
2 50F GBM R frontal lobe17 cc 251 cc
3 56M GBM R frontal lobe80 cc 564 cc
4 60F GBMR basal ganglia 19 cc 297cc
5 62F GBM R temporal lobe101 cc482 cc
6 62F GBM L parietal lobe38 cc454 cc
Figure 1. Diagnostic imaging and treatment plan from a representative case. A) Pre-operative contrast enhanced MRI. B) Post-operative
contrast enhanced MRI showing resection cavity. C) Treatment planning CT with isodose lines depicting the dose distribution.
Cryptic Critical Structures in High-Grade Glioma
PLoS ONE | www.plosone.org2 March 2012 | Volume 7 | Issue 3 | e32098
representative case is illustrated in Figure 4, with dose-volume
histogram analysis (DVH) demonstrating shifts in the dose-volume
curves following re-optimization. A qualitative dose-reduction
effect was more pronounced for OARs contralateral to the tumor
In order to quantify the degree of dose-reduction, we analyzed
the change in mean dose and maximum dose to critical structures
after re-optimization. With regard to mean dose, an average dose
reduction of 3% was achieved for critical structures ipsilateral to
the tumor bed (range 0%–16%). The ability to reduce dose to
ipsilateral structures was limited by the need to preserve tumor
coverage – in many cases, these ipsilateral structures lay within or
in immediate proximity to the tumor bed. However, as
anticipated, the effect of re-optimization on contralateral struc-
tures was more pronounced – an average dose reduction of 11%
was achieved for critical structures contralateral to the tumor bed,
with 93% of these structures showing some dose reduction (range
0.4% increase to 22% reduction). Generally, the visual cortices
were located at the greatest distance from the tumor bed – these
structures correspondingly enjoyed the greatest magnitude of dose
reduction (Figure 5). Importantly, the goal of critical structure
sparing did not compromise tumor coverage – the average change
in dose to the PTV was a 0.2% increase (range 0.7% decrease to
0.9% increase). Analysis of reduction in maximum dose to OARs
revealed similar results, albeit with somewhat more modest
reductions (average reduction of 2.4% to ipsilateral structures
and 9.7% to contralateral structures).
Strategies to reduce iatrogenic neurotoxicity secondary to CNS
irradiation have been of longstanding interest, and in recent years,
growing attention has been given to efforts to assess and avert
adverse neurocognitive outcomes in a variety of clinical situations
(e.g. dose reduction for patients with primary CNS lymphoma
, and SRS alone for patients with 1–3 brain metastases
In addition to dose reduction where appropriate, an additional
strategy under active investigation is that of reducing unnecessary
dose to functional neural pathways. An ongoing trial administered
by the Radiation Therapy Oncology Group (RTOG) – RTOG
0933 – examines the potential benefit of hippocampal avoidance
for patients undergoing whole brain radiation therapy (WBRT) for
cancers metastatic to the brain. This trial seeks to test whether
IMRT techniques that spare the hippocampus will yield improved
neurocognitive outcomes. The patient population for this study
includes patients with a RTOG brain metastasis recursive
partitioning analysis (RPA) class of I–II – translating to a median
Figure 2. Anatom-E digital atlas. For structure delineation, the display of treatment planning system (Pinnacle) is mirrored to a second computer
monitor, and the Anatom-E images are superimposed (A) and adjusted and scaled to achieve registration with the planning CT images. Anatom-E
software controls allow blending (B) of the MRI and wireframe views for ease of registration and structure identification.
Figure 3. Identification of cryptic critical structures. In A) the
overlay of the Anatom-E atlas and planning CT is illustrated. In B), the
atlas image has been turned off, leaving the structures alone
superimposed on the planning CT. Panel C shows the overlay with
the pre-treatment MRI, and Panel D depicts the atlas images turned off,
with the resultant contours on the planning CT.
Cryptic Critical Structures in High-Grade Glioma
PLoS ONE | www.plosone.org3 March 2012 | Volume 7 | Issue 3 | e32098
overall survival of less than one year . The expected median
survival of GBM patients, by contrast, ranges from 12–17 months
[2,3] suggesting that efforts to reduce potential neurocognitive
effects of radiation are appropriate in this population as well.
Two recent reports also explore the feasibility of accounting for
dose to non-traditional OARs in treatment planning for high-
grade glioma. As part of an analysis of advanced treatment
delivery technologies, a Danish study  examined the dose
delivered to the hippocampi from IMRT plans compared to
treatment planning with inversely optimized rotational therapy
(RapidArc) and intensity modulated proton therapy (IMPT).
Details on the delineation of the hippocampal OAR were not
specified, but the authors found that IMPT plans gave the lowest
average dose to the contralateral hippocampus. Treatment
planning results in the absence of efforts to spare the hippocampus
were not described, precluding quantification of what dose
reduction was achieved. Another recent study from Marsh et al
 examined sparing of the limbic system, hippocampus and
neural stem-cell niches in high-grade glioma. This study used more
stringent optimization criteria, setting a maximum dose of 8 Gy to
the contralateral hippocampus, and reported a more dramatic
reduction of dose to the contralateral hippocampus (57%
reduction in mean dose). However, this analysis is not directly
comparable to the present study, as the inclusion of the posterior
optic radiations as avoidance structures in this analysis affects the
ability to minimize dose to multiple geographic regions of the
Additionally, while several guidelines for hippocampal sparing
have been published in the clinical literature [18,19], this complex
structure still requires expertise for accurate segmentation – as
Figure 4. Re-optimized treatment plans reduce dose to cryptic OARs. A) Comparison of isodose lines on an axial slice from a representative
patient. B) Dose-volume histogram analysis for the plan depicted in (A).
Figure 5. Re-optimization reduces mean and maximum dose parameters. The average percent reduction in mean (left panel) and maximum
(right panel) dose is shown for the ipsilateral and contralateral visual cortices, hippocampi (hip.) and optic radiations (OR). Error bars represent the
standard error of the mean.
Cryptic Critical Structures in High-Grade Glioma
PLoS ONE | www.plosone.org4 March 2012 | Volume 7 | Issue 3 | e32098
underscored by the fact that the RTOG 0933 trial requires central
review of hippocampal OAR contours. Our work demonstrates
the feasibility of using a deformable registration tool to accomplish
the delineation of this structure – previous dosimetric studies do
not provide detailed descriptions of the methods used to identify
the hippocampi. Additionally, to our knowledge, this study is the
first to include analysis of dose avoidance to the posterior optic
pathways, for which there are no consensus contouring guidelines.
Owing to the extensive annotation of the digital anatomic atlas,
this technique could be extended to include other functional areas
Ultimately, as data on neurocognitive effects of radiation
therapy emerge from ongoing clinical investigations, and knowl-
edge of functional neuroanatomy is further supplemented with
improvements in functional imaging, accounting for dose
delivered to these circuits will become increasingly important in
radiation oncology treatment planning. Our study reduced dose to
multiple OARs simultaneously while preserving target coverage.
As data regarding the clinical effects of cryptic OAR sparing
approaches emerges, treatment planning objectives can be altered
accordingly in pursuit of dose reductions of larger magnitude for
an individualized clinical scenario.
Future studies will help define the risk-benefit ratio of reduced
target coverage in favor of OAR preservation, as well as help
establish relative priorities for OAR preservation. Overall, our
work illustrates how a deformable anatomic atlas can be used to
reduce the dose delivered to cryptic critical structures – critical
structures that are not accounted for in conventional treatment
planning schema were identified and spared without compromis-
ing target coverage.
The authors thank John Pagani of Anatom-E for assistance with image
registration and software configuration.
Conceived and designed the experiments: DCW ELC. Performed the
experiments: DCW SDB. Analyzed the data: DCW SDB ELC. Wrote the
1.Report CS (2011) Primary Brain and Central Nervous System Tumors
Diagnosed in the United States in 2004–2007. Source: Central Brain Tumor
Registry of the United States, Hinsdale, IL. website: www.cbtrus.org.
Stupp R, Hegi ME, Mason WP, van den Bent MJ, Taphoorn MJ, et al. (2009)
Effects of radiotherapy with concomitant and adjuvant temozolomide versus
radiotherapy alone on survival in glioblastoma in a randomised phase III study:
5-year analysis of the EORTC-NCIC trial. Lancet Oncol 10: 459–466.
Stupp R, Mason WP, van den Bent MJ, Weller M, Fisher B, et al. (2005)
Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma.
N Engl J Med 352: 987–996.
Hegi ME, Diserens AC, Gorlia T, Hamou MF, de Tribolet N, et al. (2005)
MGMT gene silencing and benefit from temozolomide in glioblastoma.
N Engl J Med 352: 997–1003.
Jenkins RB, Blair H, Ballman KV, Giannini C, Arusell RM, et al. (2006) A
t(1;19)(q10;p10) mediates the combined deletions of 1p and 19q and predicts a
better prognosis of patients with oligodendroglioma. Cancer Res 66: 9852–9861.
van den Bent MJ, Looijenga LH, Langenberg K, Dinjens W, Graveland W,
et al. (2003) Chromosomal anomalies in oligodendroglial tumors are correlated
with clinical features. Cancer 97: 1276–1284.
Cairncross G, Berkey B, Shaw E, Jenkins R, Scheithauer B, et al. (2006) Phase
III trial of chemotherapy plus radiotherapy compared with radiotherapy alone
for pure and mixed anaplastic oligodendroglioma: Intergroup Radiation
Therapy Oncology Group Trial 9402. J Clin Oncol 24: 2707–2714.
Wick W, Hartmann C, Engel C, Stoffels M, Felsberg J, et al. (2009) NOA-04
randomized phase III trial of sequential radiochemotherapy of anaplastic glioma
with procarbazine, lomustine, and vincristine or temozolomide. J Clin Oncol 27:
Butler EB, Shukla V, Paulino AC, Lu HH, Butler R, et al. (2007) Computer
visualization techniques (CVTs) foster evidence-based target delineation. Cancer
Invest 25: 1–5.
10. HamiltonJD, K VA,Hayman LA, Prabhu SS,Hou P, JacksonEF,GronemanM,
Pagani J, Vabulas M, Kumar AJ. (in press). Deformable Anatomic Templates
Embed Knowledge into Brain Images: Part1. Construction and Display. Journal of
Computer Assisted Tomography.
11. Hamilton JD, K VA, Hayman LA, Prabhu SS, Hou P, et al. (in press)
Deformable Anatomic Templates Embed Knowledge into Brain Images: Part 2.
Validation Using Functional Magnetic Resonance Imaging of the Motor Hand.
Journal of Computer Assisted Tomography.
12. Shah GD, Yahalom J, Correa DD, Lai RK, Raizer JJ, et al. (2007) Combined
immunochemotherapy with reduced whole-brain radiotherapy for newly
diagnosed primary CNS lymphoma. J Clin Oncol 25: 4730–4735.
13. Chang EL, Wefel JS, Hess KR, Allen PK, Lang FF, et al. (2009) Neurocognition
in patients with brain metastases treated with radiosurgery or radiosurgery plus
whole-brain irradiation: a randomised controlled trial. Lancet Oncol 10:
14. Aoyama H, Shirato H, Tago M, Nakagawa K, Toyoda T, et al. (2006)
Stereotactic radiosurgery plus whole-brain radiation therapy vs stereotactic
radiosurgery alone for treatment of brain metastases: a randomized controlled
trial. JAMA 295: 2483–2491.
15. Gaspar L, Scott C, Rotman M, Asbell S, Phillips T, et al. (1997) Recursive
partitioning analysis (RPA) of prognostic factors in three Radiation Therapy
Oncology Group (RTOG) brain metastases trials. Int J Radiat Oncol Biol Phys
16. Rosenschold PM, Engelholm S, Ohlhues L, Law I, Vogelius I, et al. Photon and
proton therapy planning comparison for malignant glioma based on CT, FDG-
PET, DTI-MRI and fiber tracking. Acta Oncol 50: 777–783.
17. Marsh JC, Godbole R, Diaz AZ, Gielda BT, Turian JV (2011) Sparing of the
hippocampus, limbic circuit and neural stem cell compartment during partial
brain radiotherapy for glioma: A dosimetric feasibility study. J Med Imaging
Radiat Oncol 55: 442–449.
18. Chera BS, Amdur RJ, Patel P, Mendenhall WM (2009) A radiation oncologist’s
guide to contouring the hippocampus. Am J Clin Oncol 32: 20–22.
19. Gondi V, Tolakanahalli R, Mehta MP, Tewatia D, Rowley H, et al. (2010)
Hippocampal-sparing whole-brain radiotherapy: a ‘‘how-to’’ technique using
helical tomotherapy and linear accelerator-based intensity-modulated radio-
therapy. Int J Radiat Oncol Biol Phys 78: 1244–1252.
Cryptic Critical Structures in High-Grade Glioma
PLoS ONE | www.plosone.org5 March 2012 | Volume 7 | Issue 3 | e32098