Institute of Cancer Research Repository
Please direct all emails to:
This is an author produced version of an article that appears in:
The internet address for this paper is:
INTERNATIONAL JOURNAL OF RADIATION ONCOLOGY
A M Kirby, J R Yarnold, P M Evans, V A Morgan, M A Schmidt,
E D Scurr, N M DeSouza (2009) Tumor Bed Delineation for
Partial Breast and Breast Boost Radiotherapy Planned in the
Prone Position: What Does MRI Add to X-ray CT Localization
of Titanium Clips Placed in the Excision Cavity Wall?,
International Journal of Radiation Oncology Biology
MRCT tumor bed delineation for partial breast RT Kirby page 1
Tumour bed delineation for partial breast and breast boost radiotherapy planned
in the prone position: what does MRI add to x-ray CT localization of titanium clips
placed in the excision cavity wall?
Anna M. Kirby FRCR,* John R. Yarnold FRCR, *φ Philip M. Evans D.Phil, φ Veronica A. Morgan MSc,
ψ Maria A. Schmidt Ph.D.,φ Erica D. Scurr MSc, ψ Nandita M. deSouza FRCR. ψ
*Department of Academic Radiotherapy, Royal Marsden NHS Foundation Trust, Sutton, UK
φJoint Departments of Radiotherapy and Physics, Institute of Cancer Research, Sutton, UK
ψDepartment of Clinical Magnetic Resonance, Institute of Cancer Research, Sutton, UK
Corresponding author: Dr Anna M. Kirby FRCR, Department of Academic Radiotherapy, Royal Marsden
NHS Foundation Trust, Downs Road, Sutton, Surrey, SM2 5PT, United Kingdom. Telephone: +44 208 661
3392. Fax: +44 208 661 3107. email@example.com
Acknowledgements: Dr A Kirby was funded by the Royal College of Radiologists’ Clinical
Oncology Research Fellowship.
Conflict of interest: None declared.
MRCT tumor bed delineation for partial breast RT Kirby page 2
Purpose: To compare tumour bed (TB) volumes delineated using MRI plus CT/clips (MRCT)
with those delineated using CT/clips-alone in post-lumpectomy breast cancer patients positioned
prone, and to determine the value of MRCT for planning partial breast radiotherapy (PBI).
Materials/ Methods: 30 women with breast cancer each had 6-12 titanium-clips secured in the
excision cavity walls at lumpectomy. Patients underwent prone CT-imaging followed by prone
MRI (T1-weighted (standard & fat-suppressed) and T2-weighted sequences). TB volumes were
delineated separately on CT and fused-MRCT datasets. Clinical (CTV=TB+15mm) and planning
target volumes (PTV=CTV+10mm) were generated. Conformity indices between CT- and
MRCT-defined target volumes were calculated (ratio of volume-of-agreement to total-delineated-
volume). Discordance was expressed as a geographical-miss index (GMI=fraction of total-
delineated-volume not defined by CT), and a normal-tissue index (NTI=fraction of total-
delineated-volume designated as normal tissue on MRCT). PBI dose-distributions were
generated to cover CT-defined CTV (CTVCT) with ≥95% of the reference dose. The percentage
of MRCT-defined CTV (CTVMRCT) receiving ≥95% of the reference dose was measured.
Results: Mean conformity indices were 0.54 (TB), 0.84 (CTV) and 0.89 (PTV). For TB volumes,
GMI was 0.37 and NTI was 0.09. Median percentage-volume coverage of CTVCT was 97.1%
(range 95.3-100.0%), and of CTVMRCT was 96.5% (range 89.0-100.0%).
Conclusions: Addition of MR- to CT/clip-data generated TB-volumes that were discordant to
those based on CT/clips-alone. However, clinically satisfactory coverage of CTVMRCT by CTVCT-
based tangential PBI fields provides support for CT/clip-based TB delineation remaining the
method of choice for PBI/ breast boost radiotherapy planned using tangential fields.
MRCT tumor bed delineation for partial breast RT Kirby page 3
Key words: Breast cancer; Partial breast radiotherapy; Magnetic resonance imaging; Target
volume delineation; computed tomography
MRCT tumor bed delineation for partial breast RT Kirby page 4
Whole breast radiotherapy (WBRT) following breast-conserving surgery (BCS) improves local
control and survival (1, 2) but is associated with increased non-breast-cancer-related mortality
and morbidity due to irradiation of non-target tissue (2, 3). A strategy that aims to improve the
therapeutic ratio in women at relatively low risk of local tumour relapse involves limiting high
radiation doses to the index quadrant and reducing or eliminating dose to breast tissue remote
from the tumour bed (TB) (4, 5). An essential prerequisite of partial breast irradiation (PBI) is
accurate localization of the TB. Until recently, TB localization was performed using pre-operative
radiological imaging, surgical annotation, clinical palpation of surgical defect, scar position,
patient recollection of tumour location, and in some cases post-operative ultrasound imaging.
More recently, localization of titanium clips attached to excision cavity walls at surgery has been
shown to reduce risks of geographical miss and unnecessary normal-tissue irradiation (6-9).
Clips provide additional localization information compared to kv-CT-imaging alone, and therefore
CT/clip-based TB delineation is considered the current gold standard (10). However, clips define
a limited number of points on an irregular excision cavity wall surface [see figure 1], and the
complete breast tissue/excision cavity interface is derived by interpolation, taking into account
tissue density and distortion (11). Moreover, CT has limited soft-tissue contrast making it an
unreliable modality for detecting small volumes of seroma between clips and for distinguishing
TB from normal glandular breast tissue (12).
Magnetic resonance (MR) imaging has superior soft-tissue contrast and the potential to
differentiate more clearly between normal tissue and post-operative TB. Studies in post-
operative sarcoma and prostate patients have correlated signal changes on T1- (T1W) and T2-
weighted (T2W) images with seroma, haematoma, haemorrhage and fibrosis on the pathological
specimen (13-16). Following wide-local excision of breast cancer, heterogeneous ellipsoidal
fluid-filled cavities with irregular borders are commonly reported on MR (17), but there are no
MRCT tumor bed delineation for partial breast RT Kirby page 5
studies using these data for radiotherapy planning. Direct comparison of MR- and CT/clip-
defined volumes is difficult in the supine position due to the limited bore size of conventional
closed MR scanners, respiratory motion artefacts, and distortion of breast tissue by overlying
MR-receiver coils. These difficulties are partially overcome by use of a prone position (as per
diagnostic breast MR examinations) and this position is therefore more suitable for comparison
of CT and MR in the context of breast radiotherapy planning.
The purpose of this study was to evaluate the contribution of MRCT to the planning of partial
breast and breast boost radiotherapy. This was done by comparing tumour bed volumes
delineated using MR plus CT/clips (MRCT) with those delineated using CT/clips alone in post-
lumpectomy breast cancer patients positioned prone.
MRCT tumor bed delineation for partial breast RT Kirby page 6
Patients and methods
The study was approved by the Royal Marsden NHS Foundation Trust Committee for Clinical
Research. Patients due to undergo adjuvant breast RT following lumpectomy for unifocal G1-3
invasive ductal carcinoma or high-grade ductal carcinoma-in-situ were eligible. Patients with
claustrophobia or ferrous implants were excluded.
Placement of titanium clips
Patients had titanium clips placed at BCS according to a national protocol (10) in which each of
the six excision cavity boundaries was defined by 1 or 2 clips. Clips were secured at the centre
of the deep boundary, half-way between skin and pectoral fascia (lateral, medial, superior and
inferior walls) and in subcutaneous tissue close to the suture line (anterior boundary). Where
oncoplastic surgery was performed, clips were placed before re-modelling. All patients had clear
margins of ≥2mm around the microscopic limits of tumour.
Patient positioning and image acquisition
Each patient underwent CT-imaging in a prone position not less than three weeks post-surgery
using an in-house designed platform with an aperture through which the index breast could fall
away from chest wall. The platform was compatible with both CT and MR scanners. The
contralateral breast was supported on a foam wedge. Three multi-modality markers were placed
at different points on the index breast surface in proximity to the scar. Patients had bilateral
tattoos in the mid-axillary line (placed as part of their standard supine radiotherapy treatment-
planning). The distances from each lateral tattoo to the surface and inferior edge of the prone
platform were recorded. Photographs were taken of arm and head position. Non-contrast CT
images were acquired in each position (slice thickness 1.5mm, from C6 to below diaphragm).
Patients proceeded directly to the MR scanner where their prone position was reproduced on
the same platform using measurements from tattoos and photographs. Patients were imaged
using T1W 3-D sequences without fat suppression (TR 6.1ms, TE 1.7ms, flip angle 12) and with
MRCT tumor bed delineation for partial breast RT Kirby page 7
fat suppression (TR 4.8ms, TE 2.4ms, flip angle 10, 100ms inversion delay pulse), and using
T2W sequences (TR 10195ms, TE 100ms, flip angle 90). Each sequence had a field of view of
192/100mm, slice thickness 1mm, and matrix 192*192mm. Imaging acquisition took 20 minutes.
Image transfer and fusion
MR data was imported into the radiotherapy-planning system (Pinnacle version 8.0, ADAC
systems) and co-registered with CT data. Matching was achieved using regions of interest (ROI)
corresponding to the midpoint of each of the surgical clips, manually identified on each of the CT
and MR datasets. Misalignment between CT and MR clips was calculated in terms of the total
conjugate deviation (TCD), defined as the square root of the sum of the squares of the deviation
between each pair of ROIs. The mean misalignment per clip was calculated as √(TCD2/ number
of clips) in 3-dimensions.
Target volume definition
TB was delineated on the prone CT data by a single observer (AK), encompassing clips, seroma
and architectural distortion. Window level and width were fixed at 0 and 500 Hounsfield Units
respectively. The observer was blinded to MR findings. Each CT-defined TB (TBCT) was
assigned a cavity visualization score (CVS) (12) in which 1=no cavity visible, 2=heterogeneous
cavity with indistinct margins, 3=heterogeneous cavity with some distinct margins, 4=mildly
heterogeneous cavity with mostly distinct margins, and 5= homogeneous cavity with clearly
identified margins. TB was outlined on the MR data at least two weeks after CT-outlining by the
same observer in consensus with an experienced MR radiologist (NdS) (both blinded to CT
findings), encompassing seroma, fibrosis and clips. Tissue that produced heterogeneous signal
on all three MR sequences was considered to be haemorrhage/haematoma and was also
included. The MR- and CT-defined TB volumes were fused to create an MRCT-defined TB
(TBMRCT). This fused volume was edited to exclude tissue that MR did not classify as seroma,
fibrosis, haematoma, or haemorrhage. Clinical target volumes were created for each of the CT
MRCT tumor bed delineation for partial breast RT Kirby page 8
(CTVCT), and fused-MRCT (CTVMRCT) datasets by adding a uniform 15mm margin to the tumour
bed in 3D, limited deeply by chest wall and superficially by 5mm beneath skin surface. Planning
target volumes were created for each of the CT (PTVCT) and MRCT (PTVMRCT) datasets by
addition of a 10mm margin to the CTV (limited by skin). Volumes were recorded and amount of
overlap and underlap between CT- and MRCT-defined target volumes calculated.
Conformity and discordance between volumes
Conformity between TBCT and TBMRCT was expressed as a conformity index (CI). Discordance
was expressed as geographical miss (GMI) and normal tissue (NTI) indices. Definitions for these
indices are given in figure 2. The same indices were calculated for CT- versus MRCT-defined
CTV and PTV.
PBI dose distributions were generated using multiple static tangential fields with the aim of
covering CT-clip-defined target volumes according to UK IMPORT LOW (Intensity Modulated
Partial Organ Radiotherapy) trial criteria (18) i.e. >95% of CTVCT should be covered by >95% of
isocentre dose (50Gy in 2Gy fractions). Plans fulfilled ICRU dose homogeneity criteria (19).
Based on these plans, the percentage of CTVMRCT receiving 95% of isocentre dose was
measured, a value of ≥95% being deemed adequate coverage and a value of <95% being
deemed inadequate coverage. Statistical analyses were performed using the two-tailed Student
t-test for significance.
MRCT tumor bed delineation for partial breast RT Kirby page 9
Thirty-five patients gave informed consent to participate in the study. Two patients did not fit into
the MR-scanner on the in-house platform. Three patients were unable to tolerate MR-scanning
due to claustrophobia. Thirty patients had evaluable data. Median age was 54 years (range 34
to 76), and median UK cup size was C (range A to FF) (equivalent to median US cup size B
(range A to G). Median time from surgery to imaging was 47 (22-210) days. Twenty patients had
full-thickness closure of their excision cavities following lumpectomy (apposed cavities) and 10
patients did not (unapposed cavities). 23/30 patients had cavity visualisation scores (CVS) (12)
of 1 (n=7) or 2 (n=16). 7/30 patients had more clearly-visualised cavities: CVS=3 (n=2); CVS=4
(n=3); CVS=5 (n=2).
Mean clip misalignment across all 30 cases was 0.8mm (medial-lateral), 0.6mm (superior-
inferior) and 1.0mm (anterior to posterior). The largest mean clip misalignment seen was 2.6mm
Comparison of findings on imaging sequences
Table 1 summarizes the features described on CT and on the individual MR sequences. These
features are illustrated in Figure 3. Titanium clip and skin-surface markers were demonstrated
on all datasets. Clip-related artefact was minimal on CT and did not affect visualisation of
surrounding tissue. Seroma was visualised on T2-W images, enabling MR to distinguish TB from
normal glandular breast tissue (figure 3a(i-iv)). T1-W MR sequences most clearly demonstrated
titanium clips (figure3b(iv)). Clips appeared as voids and those closest to the breast/ chest wall
interface were most difficult to visualize. Heterogeneity corresponding to haemorrhage or
haematoma was visualized on T1W and T2W MR sequences (case 3, images 3b-d). Wrap
artefact at lateral edge of image made it difficult to visualize lateral TB on MR in some cases
MRCT tumor bed delineation for partial breast RT Kirby page 10
The presence of seroma on MR was related to the time interval from surgery to scanning:
patients with visible seroma had a median time to scan of 39 (range 22-210) days whilst those
with no seroma had a median time to scan of 154 (31-196) days. The presence of fibrosis on MR
was also associated with time from surgery to scan. Median time to imaging for patients with
fibrosis was 154 (44-196) days and for those without fibrosis was 42 (22-210) days.
Target volumes and the differences between them are summarized in table 2. In 28/30 cases,
the addition of MR to CT data increased the TB volume. Median percentage volume increases
for MRCT- versus CT-defined CTV and PTV were proportionally less than for TB because these
volumes are truncated at skin or lung/chest-wall interface. CTVCT and CTVMRCT correlate well
(Pearson correlation coefficient= 0.957, p<0.001). (figure 4).
Table 3 summarizes values for conformity and discordance between CT- and MRCT-target
volumes. Concordance between TB volumes was low but increased for CTV and PTV due to
truncation of target volumes at skin and lung. Mean differences in centres-of-mass for TBCT
versus TBMRCT were 1.5mm in the medial-lateral plane, 2.0mm in the anterior-posterior plane
and 2.2mm in the superior-inferior plane.
Target volume coverage by standard tangential PBI plans
Median percentage volume encompassed by the 95% isodose was 97.1% for CTVCT (range
95.3-100.0%), and 96.5% for CTVMRCT (range 89.0-100.0%). The 95% isodose covered the
CTVCT in 30/30 cases and covered the CTVMRCT in 26/30. In 3/4 cases with inadequately-covered
CTVMRCT, the percentage of CTVMRCT encompassed by the 95% isodose was >93%. In the
remaining case, the percentage of CTVMRCT encompassed by the 95% isodose was 89.0% (95%
of the CTVMRCT was covered by the 87% isodose). In 2/4 of the inadequately-covered cases,
TBMRCT extended inferiorly to TBCT, and 4/4 cases had tumours located at the extreme lateral or
medial edges of breast tissue. Mean CTV conformity index of the inadequately covered cases
was significantly lower than that of the adequately covered-cases (0.69 vs. 0.86, p=0.001), but
MRCT tumor bed delineation for partial breast RT Kirby page 11
there was no difference in mean number of clips (6 in both groups), CVS (2 in both groups) or
TBCT volume (7.9 vs. 8.5cm3, p=0.9) between the adequately and inadequately covered groups.
MRCT tumor bed delineation for partial breast RT Kirby page 12
This study found that tumour bed volumes delineated using fused MR and CT/clip data were
discordant to those delineated using CT/clips alone. However, resulting clinical and planning
target volumes were sufficiently concordant to ensure adequate coverage of CTVMRCT using
CTVCT-based tangential partial breast radiotherapy fields in most cases.
CT alone is able to clearly visualise only seromas that are large enough to be under tension,
producing a convex border (equivalent to CVS 4 and 5) (12). These were an infrequent finding in
our population, of whom two-thirds had their excision cavities apposed at surgery. In all of our
patients, titanium clips, clearly visible on CT, defined points on the TB/ breast tissue interface.
However, uncertainty remained over how to join these points together. Intervening soft tissue
abnormalities, described as “architectural distortion”, were seen in 23/30 of our cases on CT.
Distortion can represent post-operative change (small-volume seroma, fibrosis, haemorrhage or
oedema) but is difficult on CT alone to distinguish from normal glandular breast tissue. A more
common problem was that clips were separated by apparently normal fatty tissue. Indeed, 7/30
cases in our study had no abnormalities at all on CT apart from clips. We assumed that tissue in
between clips was TB but could not with any certainty decide on the true location of the excision
cavity wall between clips. MR imaging, on the other hand, visualised seroma, haemorrhage,
haematoma and fibrosis in association with titanium clips. The TBMRCT extended outside the
TBCT in most cases resulting in a median GMI of 0.37. This finding agrees with previous work
reporting MR-defined TB volumes to be larger than those defined on CT (20). MR was also able
to distinguish post-operative change from normal glandular breast tissue, albeit with a median
NTI of only 0.07. Thus, the principal cause of discordance between TBCT and TBMRCT volumes
was the finding of soft-tissue abnormalities on MR in regions where CT defined apparently
normal tissue. Although MR identified a larger volume of abnormal tissue than CT, it was difficult
on MR imaging alone to identify clips close to the breast tissue/ chest wall interface. T1-weighted
MRCT tumor bed delineation for partial breast RT Kirby page 13
sequences without fat suppression were better than the other two sequences, as the signal
voids left by clips were more clearly visible against the high-signal fat, but still only detected 77%
of clips. Inclusion of gradient echo sequences might improve clip visualisation in future studies.
Following expansion of TB to CTV, CI between CT - and MRCT-defined volumes improved from
0.54 to 0.89 due to the size of the margin in relation to the magnitude of discordance and to
limits on expansion presented by skin, chest wall and breast tissue boundaries. In the majority of
cases, the addition of MR to CT/clip-data generated target volumes which were adequately
encompassed by tangential PBI fields based on CTVCT. In only 4/30 cases was coverage
inadequate according to our criteria, and in only two of these was inadequate coverage due to
discordance between volumes, both inferiorly. In the other two cases, inadequate coverage was
related to the target volumes being located at the peripheries of breast tissue where coverage is
difficult to achieve due to the complex 3-dimensional shape of breast tissue. Differences in COM
positions for TBCT versus TBMRCT confirmed that discordance was greatest in the SI plane
(2.2mm) but were again small in the context of a TB-PTV margin of 25mm, thus explaining why
coverage of TBMRCT by the 95% isodose was achieved in such a high proportion of cases.
Even in the case with the least adequate coverage, 89% of the CTVMRCT was encompassed by
the 95% isodose, and 95% of the CTVMRCT was covered by the 87% isodose. A recent large
study of dose-fractionation in breast radiotherapy has found the gradient (γ value) of the dose-
response curve (measured as the percentage increase in effect per percentage increase in total
dose delivered in 2Gy fractions) to be only 0.2 (1). Assuming this γ value to be correct, a 13%
underdosage of 11% of the partial breast CTV would not be expected to impact measurably
upon local tumour control.
One criticism of this study is that the different imaging modalities are delineating different
targets. Following excision of breast cancer, the boundaries of TB have been defined in 3D as
the interface between fluid and breast tissue, demonstrated on CT by a change in soft-tissue
MRCT tumor bed delineation for partial breast RT Kirby page 14 Download full-text
density. A margin (standardly 10-15mm) is then added to encompass tissue considered to be at
risk of local recurrence (CTV). Our MR-based delineation protocol is likely to overestimate the
true TB by including haematoma and haemorrhage that is not necessarily within the cavity itself
but may instead represent pericavity post-operative changes. The resulting “post-operative
complex” (17) would therefore include part of the tissue-volume at risk of microscopic spread
and adding 15mm to this structure would likely overestimate CTV. However, without including
post-operative haemorrhage and haematoma it would have been difficult to standardise our
approach to MR-delineation of TB. Two-thirds of our study population underwent full-thickness
closure of excision cavities, decreasing the volume of intra-cavity fluid, and resulting in a cavity-
tissue interface that was difficult to define, even on MR. Also, reports suggest that granulation
tissue may be laid down within the original excision cavity (17), and we did not want to
underestimate TB on MR by only outlining seroma. Our approach of including any tissue on MR
that might be part of the cavity/ tissue interface produced a “worst case scenario” by which to
test the current CT/clip-based method. Our finding that addition of MR to CT/clips did not
significantly increase target volumes is reassuring that use of the CT/clip method is unlikely to
result in a geographical miss.
The co-registration of CT and MR datasets is another potential source of error, but the use of TB
clips as match points minimized changes in breast shape from CT to MR as a variable (21).
Clips had other advantages over chest wall as a matching structure: they overcame the problem
of accurately identifying bony boundaries on MR, the smaller field-of-view required reduced
system-related image distortion (22), and clips were within the region that we were interested in
matching most accurately. Previous work suggests that, for fusion to be considered satisfactory,
the total conjugate deviation should be <3mm (consistent with a mean misalignment between
imaging modalities of <1.74mm for each clip) (21). Our mean misalignment was better than this.
Only 3/30 cases had mean misalignment in a single plane of >2mm (none of whom had