- Email: [email protected]
Characterization of Target Volume Changes During Breast Radiotherapy Using Implanted Fiducial Markers and Portal Imaging
Int. J. Radiation Oncology Biol. Phys., Vol. 73, No. 3, pp. 958–966, 2009 Copyright Ó 2009 Elsevier Inc. Printed in the USA. All rights reserved 0360-3016/09/$–see front matter
doi:10.1016/j.ijrobp.2008.10.030
PHYSICS CONTRIBUTION
CHARACTERIZATION OF TARGET VOLUME CHANGES DURING BREAST RADIOTHERAPY USING IMPLANTED FIDUCIAL MARKERS AND PORTAL IMAGING EMMA J. HARRIS, PH.D.,* ELLEN M. DONOVAN, PH.D.,* JOHN R. YARNOLD, M.D.,y CHARLOTTE E. COLES, M.D.,z AND PHILIP M. EVANS, D.PHIL., * ON BEHALF OF THE IMPORT TRIAL MANAGEMENT GROUP y
* Joint Department of Physics, Institute of Cancer Research and Royal Marsden Foundation Trust, Sutton, United Kingdom; Department of Clinical Oncology, Royal Marsden Foundation Trust, Sutton, United Kingdom; and z Department of Oncology, Cambridge University Hospitals NHS Foundation Trust, Cambridge, United Kingdom Purpose: To determine target volume changes by using volume and shape analysis for patients receiving radiotherapy after breast conservation surgery and to compare different methods of automatically identifying changes in target volume, position, size, and shape during radiotherapy for use in adaptive radiotherapy. Methods and Materials: Eleven patients undergoing whole breast radiotherapy had fiducial markers sutured into the excision cavity at the time of surgery. Patients underwent imaging using computed tomography (for planning and at the end of treatment) and during treatment by using portal imaging. A marker volume (MV) was defined by using the measured marker positions. Changes in both individual marker positions and MVs were identified manually and using six automated similarity indices. Comparison of the two types of analysis (manual and automated) was undertaken to establish whether similarity indices can be used to automatically detect changes in target volumes. Results: Manual analysis showed that 3 patients had significant MV reduction. This analysis also showed significant changes between planning computed tomography and the start of treatment for 9 patients, including single and multiple marker movement, deformation (shape change), and rotation. Four of the six similarity indices were shown to be sensitive to the observed changes. Conclusions: Significant changes in size, shape, and position occur to the fiducial marker–defined volume. Four similarity indices can be used to identify these changes, and a protocol for their use in adaptive radiotherapy is suggested. Ó 2009 Elsevier Inc.
Excision cavity, Partial breast irradiation, Fiducial markers, Shape analysis, Breast radiotherapy.
mography (CT) to localize the excision cavity, and margins to encompass possible subclinical spread of disease are added to create the clinical target volume (CTV). Using fiducial markers, the center of mass (CoM) of the markers at the time of treatment can be compared with the CoM at the time of planning to calculate daily shifts in the position of the excision cavity (6, 7). In addition, changes in marker positions may arise because of deformation of the excision cavity (3) and fiducial marker migration (8, 9), which lead to changes in the size, shape, and measured CoM of the measured volume. Deformation may result in underdosing of the CTV or overdosing of healthy adjacent tissues. Marker migration
INTRODUCTION Radiotherapy of the breast after breast conservation surgery is adjuvant treatment because the gross tumor volume has been excised, leaving possible microscopic disease. Standard breast radiotherapy treats the whole breast, but the region with the greatest risk of tumor recurrence is proximal to the excision cavity, and this may be treated with a higher dose of radiation. The excision cavity may be localized by using imaging techniques to visualize seroma (1, 2), changes in tissue architecture (3, 4), or fiducial markers implanted into the breast at the time of surgery (5). The most common method uses a combination of fiducial markers (typically surgical clips) and soft-tissue changes imaged using computed toReprint requests to: Emma J. Harris, Ph.D., Institute of Cancer Research, Joint Department of Physics, Cotswold Road, Sutton, SM2 5PT, UK. Tel: (++44) 208-661-3478; fax: (++44) 208-643-3812; E-mail: [email protected] IMPORT and the Institute of Cancer Research are funded by Cancer Research UK. Charlotte Coles received funding from the
Cambridge National Institute of Health Research Biomedical Research Centre. Conflict of interest: none. Received June 27, 2008, and in revised form Sept 9, 2008. Accepted for publication Oct 21, 2008. 958
Excision cavity changes d E. J. HARRIS et al.
(a)
(b)
959
(c)
Fig. 1. Schematic two-dimensional diagram of the excision cavity boundary (dashed lines) and six fiducial markers (circles) shows the variation in excision cavity volume contained within the marker volume (MV). (a, b) Two possible MVs and (c) the MV containing the excision cavity volume chosen for this work.
will change the shape of the measured volume and its CoM. Consequently, these may differ from the shape and CoM of the excision cavity as defined at the initial CT planning scan. Measurement and characterization of changes in marker positions during treatment may help identify time trends in changes in the excision cavity, characterize deformation, and identify marker migration. Using this information, strategies may be developed so that a response to changes in the excision cavity may be made and the treatment may be tailored to the individual patient. The aims of this study are to: (1) manually analyze target volume, shape, and position based on implanted fiducial markers imaged using projection x-rays to determine what changes occur during treatment; (2) investigate automated methods of comparing target volumes and positions by using similarity indices to automatically detect changes in volume position, shape, and orientation; and (3) recommend an image-guided radiotherapy protocol for breast radiotherapy incorporating the use of similarity indices. METHODS AND MATERIALS The measured positions of implanted fiducial markers were used to define the marker volume (MV), which acts as a surrogate for the excision cavity, which cannot be visualized using portal imaging (PI). This was based on the envelope of the markers to obtain the maximum coverage of the target volume, shown in Fig. 1. We evaluated changes to the MV. Four methods (resulting in six possible similarity indices) were used to compare the MV defined using CT at the time of planning (Volume A) with that defined during treatment, imaged using PI (Volume B). The first method uses the positions of the fiducial markers and the other three use the MV. The methods used were: (1) analysis of the second-order moments of the positions of markers in A and B in the three spatial dimensions (first-order moment gives the CoM and has been studied by Coles et al. [6]); (2) volume analysis, including the measured volume of A and B and a volume index (VI) that gives the fractional change in volume of the MV between planning and treatment images; (3) the Jaccard Index (JI), which describes the volume overlap of A and B; and (4) the Minkowski Index (MI), which is sensitive to changes in volume shape only. If the CoMs of A and B are shifted to a common origin, the analysis will be independent of any translation between the two volumes. For Methods 1 (second-order moments), 2 (volume analysis), and 4 (MI), a CoM shift was applied before the analysis. For Method 3 JI was calculated before (JI1) and after (JI2) the CoM shift.
Patient study At The Royal Marsden Hospital, Sutton, United Kingdom, 11 patients gave informed consent to be entered into the National Cancer Research Network multi-centre ‘‘GOLD SEED’’ feasibility study evaluating the use of implanted fiducial markers for image-guided radiotherapy in patients with early breast cancer (5). After wide local excision of the primary tumor, gold fiducial markers were sutured around the excision cavity following a strict surgical procedure. The markers were gold rings 3 mm in diameter, with a 1-mm central hole and 1-mm thickness. The intended number of markers was six, but there were fewer markers in some patients because of re-excision or surgeon preference. Therefore, of the 11 patients recruited, 7 had six markers inserted, 4 had five markers inserted, and 1 patient had four markers sutured into the cavity. All patients underwent CT scanning before treatment (CT1) and in the last week or just after the end of their treatment (CT2). The CT axial slice separation was 1.25 mm. The CT images were exported to the Pinnacle Treatment planning system (Phillips Medical Systems, Milpitas, CA). Marker positions relative to the treatment isocenter were identified by manually locating the center of each marker on the treatment planning system. Median time between surgery and CT1 was 88 days (range, 30–169 days). Three patients received chemotherapy, with 155, 157, and 169 days between surgery and CT1 for these patients. Ten patients received hormonal therapy. Patients were treated by using three fractionation regimens. These were the current international standard of 50 Gy in 25 fractions and two regimens from randomized controlled trials of 30 Gy in five fractions and 40 Gy in 15 fractions. No changes were made to their treatment based on the fiducial markers. Portal images were acquired on the first treatment day and between 4 and 13 subsequent treatment days. Median time between CT1 and the first day of treatment was 13 days. The images were acquired using an iViewGT electronic portal imaging device (Elekta Oncology Systems Ltd., Crawley, UK). Two orthogonal portal images were acquired to locate the markers in three dimensions. One image was acquired using the anterior tangential treatment beam. The second image was acquired with the gantry placed orthogonal to this beam (using 1 MU). Examples of orthogonal portal images of the implanted fiducial markers are shown in Fig. 2. Each marker position was calculated using triangulation with inhouse software. Observer errors associated with the measurements were evaluated by three physicists locating the markers three times each by using both CT and PI and determining the coefficient of repeatability (10). Changes in MVs throughout treatment were made by analyzing the patient data manually. These observed changes were used as the gold standard with which we compared the changes identified by using the similarity indices.
I. J. Radiation Oncology d Biology d Physics
960
Volume 73, Number 3, 2009
ment of the markers, di, was measured. If we replace d with d in Eqs. 1 and 2, we may calculate the variance in marker movements, m2(AB), and define similarity index Dm, to be: Dm ¼ m2 ðA BÞ þ 1
(Eq. 4)
(where the addition of 1 is used to ensure Dm is 1 if the volumes are identical). This pairwise approach was adopted to detect changes in the position of a single marker between images of A and B as an indication of single-marker migration. As with m, this may be calculated in each spatial direction. Volume analysis. The MVs were calculated for all scans. For two volumes V(A) and V(B), VI is given by: VIðA; BÞ ¼
Fig. 2. (a) Right and (b) left anterior oblique portal image (1 MU). Portal images were acquired at orthogonal angles. Translation was measured using the CoM shifts. A significant translation is defined as a shift greater than 5 mm (the local setup tolerance). The median magnitude of rotation was 5 ; hence, this was used as the threshold to identify MVs with large rotation. Significant volume reduction is defined as a volume reduction outside the 95% confidence intervals (CIs). Single-marker movement is identified if a single marker has moved more than 2.5 mm (three times the mean total observer error) with respect to its CT1 position. Multiplemarker movement is defined as the movement of two or more markers by the same distance. A change was defined as being characteristic for a patient’s MV if observed in more than 50% of cases.
Comparison methods Second-order moment analysis. The second-order central moment of a distribution of N points is its variance and is given by: m2 ¼
N 1X ðdi m1 Þ2 N i¼1
JIðA; BÞ ¼
The position of di is specified in each of the three spatial directions: left-right (LR), superior-inferior (SI), and anterior-posterior (AP), allowing m2 to be determined in each direction (to limit the number of variables, cross-moments were not considered). Two methods of comparing Volumes A and B were considered: first, comparison of the whole volume, and second, pairwise comparison of markers. For the first method, the variance index, m, was defined: m¼
m2 ðBÞ m2 ðAÞ
(Eq. 3)
where m2(A) and m2(B) are the variance in marker positions in A and B, respectively. The m was evaluated in each direction (LR, SI, and AP). This is expected to be sensitive to changes in rotation and deformation. For the pairwise comparison of marker positions, by subtracting marker positions in Volume B from those in Volume A, the move-
jAXBj jAWBj
(Eq. 6)
The JI measures the overlap of the two volumes and a JI of unity indicates that the two volumes are identical in size, shape, position, and orientation and will be less than unity if any of these parameters differ, i.e., the ideal value of the JI is 1. The JI was evaluated both before (JI1) and after (JI2) the CoMs of the markers in Volumes A and B were shifted to a common origin. The JI1 measured differences in volume overlap between the CT and PI volumes without setup correction. Values of JI1 less then 1 may be caused by translational shifts and changes in size, shape, and orientation, whereas JI2 measures changes in size, shape, and orientation of the volumes only. Neither of these indices can be used to determine specific changes, only to indicate that one of these changes has occurred. Minkowski Index. The Minkowski Index (MI) measures the similarity in shape of two volumes (A and B) and is given by (11):
(Eq. 1)
(Eq. 2)
(Eq. 5)
(as with Eq. (4), the addition of 1 ensures identical volumes have an index of 1). The VI is constructed to be sensitive to change in volume only. Jaccard Index. The JI is defined as the intersection divided by the union of two sample sets. For Volumes A and B:
2
1
VðBÞ3 VðAÞ3 MIðA; BÞ ¼ Vm ðA; BÞ
where di is the position of point i and m1 is the CoM, given by: N 1X m1 ¼ ðdi Þ N i¼1
VðAÞ VðBÞ þ1 VðAÞ
(Eq. 7)
where Vm(A,B) is the Minkowski mixed volume for Volumes A and B (12). Volumes A and B are corrected for any difference in position or orientation: Volume B is rotated until its orientation matches that of A as closely as possible. The MI is sensitive to differences in shape only and therefore can be used to distinguish between changes in shape and other changes that occur. If no shape change has occurred, MI will be equal to 1.
Model analysis The effects of stochastic variations in measured marker positions caused by observer variability were modeled to calculate the resulting uncertainty in the volumes and similarity indices measured for the patient data and evaluate CIs to represent threshold values for each index for each patient MV; we define no change as a value of similarity index that lies within the 95% CI. For each patient’s MV, 10,000 sample volumes were generated by adding total observer error to each marker position in the LR, SI, and AP directions. Observer errors were generated by randomly sampling a normal distribution with the standard deviation equal to the total observer error (10). Similarity indices were calculated for
Excision cavity changes d E. J. HARRIS et al.
961
Table 1. Observer coefficients of repeatability for CT and PI measurements Coefficient of repeatability (mm)
Intraobserver CT/PI (Observer 1) CT/PI (Observer 2) CT/PI (Observer 3) Interobserver CT PI Total (mean intra- + interobserver) CT/PI/CT + PI
Left-right
Superior-inferior
Anterior-posterior
0.3/1.0 0.6/0.4 0.8/1.4
0.2/0.4 0.5/0.2 0.9/0.7
0.5/1.1 0.6/0.4 0.8/1.1
1.0/1.8
0.6/1.0
1.1/1.9
1.1/2.0/2.2
0.8/1.1/1.4
1.3/2.1/2.2
Abbreviations: CT = computed tomography; PI = portal imaging.
the 10,000 new volumes. It was found empirically that the resulting non-normal distribution could be transformed into a normal distribution by evaluating the cubed root of the index values. The resulting 95% confidence limits of this distribution were then found. Confidence limits were transformed back to give the 95% CI for each similarity index.
RESULTS Observer error The coefficients of repeatability (SD 1.96) in each direction are listed in Table 1 for both CT and PI for each observer (intraobserver) and all observers (interobserver). The total coefficient of repeatability (CT + PI) is used to calculate error in the similarity indices evaluated for measured MVs. Patient MVs and volume changes during treatment The implanted markers did not define a regularly shaped MV, as shown for Patients 3 and 6 in Fig. 3a and b. For Patient 3, the outer four markers describe the largest MV and therefore two markers lie completely inside the MV (c.f., Fig. 1c). The MV(CT1) for Patient 6 is an example of an MV formed from all six markers. It can be seen that the MV surface comprises a set of triangular faces, and MVs
can be described by using the greatest width and length of the largest face and the distance of the marker farthest from this face. The dimensions of the MV(CT1) and the number of markers implanted for each patient are listed in Table 2. Five types of change were observed: (1) translation, (2) rotation, (3) volume change, (4) single-marker movement, and (5) multiple-marker movement (two or more markers). The observed MV changes between CT1 and (1) first day of treatment (PI first) and (2) all MVs measured during treatment are listed in Table 2. Column 5 in Table 2 lists changes observed between CT1 and PI first. Maximum measured values are given if rotation, marker movement, or volume change is observed. Column 6 of Table 2 lists changes for all MVs measured during treatment. In Column 6, values listed are maximum values for all PI and CT2 MVs. Translations were observed in the majority of patient MVs. These were excluded from further analysis because they may be corrected by using couch shift. Two patients (Patients 4 and 11) had no substantial change between CT1 and treatment, 1 patient (Patient 6) had rotation only, and the remaining patients had some marker movement. The MVs for Patients 1 and 10 were almost planar, with thicknesses less than 1 mm.
Fig. 3. Schematic diagrams of implanted fiducial marker positions and marker volumes (MVs) for Patients (a) 3 and (b) 6. Rings represent the fiducial markers and lines show boundaries of the MVs. Shadow projections of the MVs are shown. AP = anterior-posterior; LR = left-right; SI = superior-inferior.
I. J. Radiation Oncology d Biology d Physics
962
Volume 73, Number 3, 2009
Table 2. Description of patient MV(CT1) and observed changes between CT1 and treatment and during treatment Patient No.
Dimensions of planning MV* (cm)
Volume (cm3) No. of markers
Observed MV changes (comparison of MV[CT1] and MV[PI first])
Observed MV changes during treatment (characteristic of >50% of treatment MVs)
Rotation (6.5 ) Thin MV: large variation in shape & volume, single-marker movement (2.7 mm) Single-marker movement (6.7 mm) No change Volume shrinkage (31%) Rotation (6.1 ) Rotation (8 ) Single-marker movement (3.7 mm)
1 2
2.2/1.5/0.1 (Planar MV) 4/1.8/0.3
0.16 0.64
5 5
3
4/1.3/0.7
1.11
6
4 5 6 7
4/2.2/1.2 0.6/1.5/0.6 2.5/1/1 3.5/3.5/0.5
1.45 0.35 0.52 1.85
6 5 6 6
8 9
4/2/1 6/6/4
2.03 25.62
6 6
Volume shrinkage (13%) Volume shrinkage (47%) Multiple marker movement (9.1 mm)
10
2.8/2.5/0.1 (Planar MV)
0.9
4
Rotation (6 ) Single-marker movement (3.3 mm)
11
2/2.5/0.4
0.6
5
No change
Rotation (maximum, 7.1 ) Single-marker movement (maximum, 4.2 mm) Multiple-marker movement (maximum, 7 mm) No change Volume shrinkage (60%) Rotation (maximum, 11.5 ) Rotation (maximum, 8.4 ) Multiple-marker movement (maximum, 5.2 mm) Volume shrinkage (30%) Rotation (maximum, 10.2 ) Volume shrinkage (65%) Multiple-marker movement (maximum, 13.3 mm) Rotation (maximum, 12 ) Multiple-marker movement (maximum, 4.6 mm) No change
Abbreviations: MV = marker volume; CT1 = computed tomography scanning before treatment; PI = portal imaging. * Dimensions given are length and width (see text for explanation) of largest MV face and distance of the marker farthest from this face.
it lies outside the 95% CI. There is good agreement between CT2 and the last days of treatment, indicating no systematic difference between the two imaging methods. Table 3 lists percentages of MVs with ‘‘true results’’ for each patient, in which true results are defined as a correct prediction of a change or no change using the similarity index and false results give an incorrect prediction. We now consider each of the indices in turn. 30 25 20 15 10 5
Volume of MV (cm3)
Figure 4 shows volume of MV as a function of time. The 95% CIs are shown as error bars for the MV measured for CT1 and CT2 (indicated by using larger data points) and on the first and last PI days. With the exception of Patient 9, who had a large MV, observer error produces large uncertainty in measured volumes because of the relatively small volume of the MVs compared with the magnitude of the observer errors. There was a significant (p < 0.05) change in volume size between CT1 (Day 0) and the first day of treatment for Patients 5, 8, and 9. The MV for Patient 9 decreased between CT and treatment by 47% and by a total of 65% by the end of treatment. This patient had a large specimen size (100 g), and a large volume of seroma is visible in CT1 (this patient did not undergo a second CT scan). The MVs for Patients 5 and 8 shrink in volume by the end of treatment by 60% and 30%, respectively. Patient 2 had a very thin volume that shows a large fluctuation (by >40% on Days 10, 21, and 38 [CT2]). This was a result of the movement of the most superior marker in the SI direction by 2 mm (shown in Fig. 5).
2.5 Patient 1 2 3 4 5 6 7 8 9 10 11
2.0
1.5
1.0
0.5
Similarity indices for patient data Similarity indices have been calculated between MV(CT1) and treatment MVs for all patients. Values for the first PI, last day of PI, and CT2 are shown in Fig. 6. Both m and Dm were calculated in all three spatial directions, and only the spatial direction with the greatest change is given. Patients are grouped by type of MV change. We define a significant change in index value (i.e., significantly different from 1) if
0.0 0
10
20
30
40
50
""
60
Days from CT1
Fig. 4. Volume of patient marker volumes (MVs) as a function of time. Error bars show 95% confidence intervals for computed tomography (CT) scanning before treatment (CT1) (Day 0), MV(portal imaging [PI] first), MV(PI last), and CT in the last week or just after the end of treatment (CT2; large data points).
Excision cavity changes d E. J. HARRIS et al.
963
Fig. 5. (a) Left anterior oblique digitally reconstructed radiograph (Patient 2) from computed tomography (CT) scanning before treatment (CT1) and portal images of markers acquired 8 (b) and 10 (c) days after CT1. Implanted fiducial markers are circled. The change in the most superior marker is seen.
Variance index, m. The m indicates marker movement and volume change. From Fig. 6, m indicates change for MV(PI first), MV(PI last), and MV(CT2) for all patients for whom this happens. The m gives false-positive results for Patient 11 (PI first) and false-negative results for Patients 2 (PI first, PI last, and CT2) and 3 (PI first), both of whom have singlemarker movement. The m also gives false results for Patients 8 and 10 (Table 3). In the case of Patient 2, single-marker movement has been identified (Table 2); however, this marker lies within the MV and as a consequence, variance, shape, and volume do not change. For Patient 3, marker movement is in the direction of the largest dimension and the variance is not affected until one marker moves more than 5 mm. Marker movement index, Dm. The Dm should also indicate marker movement or volume change. This index indicates changes for all MVs shown in Fig. 6 with the exception of Patient 6. False-positive results are given for Patients 2, 4, 6, and 11. Jaccard Indices, JI1, and JI2. The JI should indicate translation, deformation, volume change, or rotation. All patients had true-positive results for JI1. From Fig. 6, JI2 indicates change for MV(Day 1) for Patient 1; this is in agreement with the observed rotation. The JI2 also correctly indicates no change for Patients 4 and 11. False-negative results were seen for Patients 2, 3, 4, 6, and 8 (Table 3). For Patient 2, the single-marker movement is inside the MV and hence does not change the volume. Hence, this single-marker movement was not detected. From the manual analysis, Patient 3 has a high incidence of multiple-marker movement, which was not detected by JI2. This also was the case for MV(CT2) for Patient 4. For Patient 6, rotations of 6.2 and 6.3 were
not detected. For Patient 8, single-marker movement of 2.55 mm was not detected for one MV only. Volume Index. The VI gives an incidence of true results of 100% for all patients. Minkowski Index. The MI should indicate shape change only. For all except Patients 2 and 3, the MI gives an incidence of true results of 100%. False-negative results for Patient 2 occurred because of single-marker movement inside the MV (as described). Marker movement for Patient 3 also was not detected by using the MI because the shape does not change significantly. From this analysis, a single index cannot distinguish between single-marker movement and deformation. It was shown that in some cases, JI2 and the MI do not indicate marker movement when MV shape does not change. However, these changes can be detected by means of Dm. DISCUSSION A significant reduction in the size of the MV was observed for 3 patients. Other studies have measured the reduction in volume of the excision cavity during breast radiotherapy for a larger proportion of patients. Using CT scanning, Jacobson et al. (3) found that of 20 patients, 16 had a greater than 20% decrease in excision cavity volume 4–5 weeks into radiotherapy. Oh et al. (13) performed a similar analysis for 31 patients and found a mean reduction of 22.5% in excision cavity volume. In both studies, excision cavities were defined by using surgical clips, seroma, and anatomic changes seen in the CT. Mean MVs measured by Jacobson et al. (3) and Oh et al. (13) and in this study were 36.1, 36.6, and 3.12 cm3, respectively. Median time between CT1 and surgery
964
I. J. Radiation Oncology d Biology d Physics
Volume 73, Number 3, 2009
PI first PI last CT2 2
3 2 1 0
6
1 0 2
8 7
1
4
0
0 Rotation
3 2 1 0 2
20
10
10 0 20
3
1
10
0
0
2
80
9
1
40
0
0
2
8
5
1
4
0
0
2
8
8
1
4
0
0
2
8
2
1
4
0
0
2
Single Marker Movement
6 4 2 0
4
1 0 2 1
Volume reduction
4 11
No change
2
0
0 JI1
JI2
VI
MI
m
Δm
Fig. 6. Similarity index values for marker volumes (MVs) on the first (portal imaging [PI] first) and last day of treatment (PI last) and for computed tomography (CT) in the last week or just after the end of treatment (CT2). The similarity index Dm bars correspond to the right axis. All other index values correspond to the left axis. Error bars = uncertainty in measurement caused by observer error; solid black boxes = index values that lie outside the threshold values (95% confidence interval). JI = Jaccard Index; VI = volume index; MI = Minkowski Index.
was 88 days for this study and 60.9 for the study by Oh et al. (13) (these data were not given by Jacobson et al. [3]); therefore, shrinkage during this longer period may account for the difference in volume. However, because of the magnitude of the difference in volume (factor > 10), it is more likely that we measured smaller volumes because we use only the marker positions to define volume, and not seroma. Furthermore, our surgeons aim to obliterate the excision cavity to reduce seroma; thus, we would expect MVs to be small. Observer error in marker location by using PI was found to be less than 1 mm. However, the small dimensions of the MVs (2–60 mm) mean that this error creates large uncertainties in volume measurement and evaluation of similarity indices. Therefore, in this study, uncertainties in measured
MV are on the order of 20%, and volume reduction for some patients therefore may be undetectable. However, it is possible that the volumes used in this study do not tend to decrease by significant amounts because they are not defined by using seroma, the volume of which we would expect to decrease with time after surgery. In our study, we use treatment PI to measure marker positions. Our treatment PI had an acquisition time of 4.56 seconds, which is on the order of one breathing cycle. Therefore, any error in marker position measurement caused by infraction motion is included in the observer error (markers were slightly blurred in the images). If shorter PI acquisition times are used, it will be necessary to consider intrafraction motion in the estimation of measurement errors.
Excision cavity changes d E. J. HARRIS et al.
Table 3. Percentage of true results obtained by using similarity indices for marker volumes Patient 1 2 3 4 5 6 7 8 9 10 11
m
Dm
JI1
JI2
VI
MI
100 36 54 100 100 100 100 70 100 60 72
100 93 100 92 100 87 100 100 100 100 63
100 100 100 100 100 100 100 100 100 100 100
100 71 45 83 100 87 100 92 100 100 100
100 100 100 100 100 100 100 100 100 100 100
100 65 43 100 100 100 100 100 100 100 100
Abbreviations: m = variance index; Dm = marker movement index; JI1 = Jaccard Index before the center of mass shift; JI2 = Jaccard Index after the center of mass shift; VI = volume index; MI = Minkowski Index. The 100% indicates that all marker volume changes were identified correctly by using the similarity index.
Migration of surgical clips previously was identified in two studies (7, 8). However, this feasibility study is the first to use markers sutured into the excision cavity wall. If we consider that the greatest amount of single-marker movement detected was 7 mm for Patient 2 and multiple-marker movements of up to 13.3 mm were also observed, it is reasonable to assume that the single-marker movement observed for Patient 2 is caused by anatomic changes, not marker migration. Large
Measure marker positions
965
volume fluctuations observed for Patient 2 were not a result of the 7-mm marker movement (this marker was inside the MV), but the 2-mm movement of the most superior. This marker moved back and forth in the SI direction during treatment. This motion may be caused by slight changes in the patient’s arm position. In practice, the CTV will be derived from the MV, seroma, and any anatomic changes evident on CT. However, the MV can be used in conjunction with the similarity indices to indicate anatomic changes. With the exception of m, similarity indices agreed well with the observed changes. The JI1 was found to be very sensitive to shifts in patient position. These shifts also can be measured by using the CoM. In some cases, marker movements may not be indicated by JI2 or MI. However, Dm will be sensitive to these changes. The MI identifies shape change only. These methods will not allow automatic distinction between single- and multiple-marker movement. To apply these techniques to adaptive radiotherapy, we suggest the verification protocol shown in Fig. 7. Interventions are shown by grey trapezoids and include repositioning, investigation into the need to replan a treatment, and possible marker migration and/or CoM changes. After initial corrections for CoM shifts and rotations, VI is used to detect volume change, JI2 is used to detect changes in rotation or shape, MI detects shape change (if JI2 is positive and MI is negative, changes in JI2 must be caused by rotation), and finally, Dm is used to detect marker movement and therefore possible migration.
Calculate VI, JI2, MI and Δm
VI change?
Calculate COM
yes
Volume change
no
yes >5mm?
Reposition
JI2 change? no
no
MI change?
Calculate Rotation
yes
MI change?
Significant rotation
yes
no
Check for re-plan
yes Reposition
Δm change?
no
Shape change
no >5°?
yes
yes
no Marker migration?
Check for systematic error in CoM / migration
Treat
Fig. 7. Work flow diagram represents suggested protocol for use of the similarity indices volume index (VI), Jaccard Index analysis performed after the center of mass (COM) shift (JI2), Minkowski Index (MI), and similarity index Dm in an adaptive radiotherapy scheme.
966
I. J. Radiation Oncology d Biology d Physics
Volume 73, Number 3, 2009
CONCLUSIONS
that by using a combination of similarity indices, we can iden-
Significant changes in size, shape, and position occur to the fiducial marker–defined volume during radiotherapy. We show
tify when changes in size and shape have occurred and devise a protocol to use these methods for adaptive radiotherapy.
REFERENCES 1. Ringash J, Whelan T, Elliott E, et al. Accuracy of ultrasound in localization of breast boost field. Radiother Oncol 2004;72: 61–66. 2. Wong EK, Truong PT, Kader HA, et al. Consistency in seroma contouring for partial breast radiotherapy: Impact of guidelines. Int J Radiat Oncol Biol Phys 2006;66:372–376. 3. Jacobson G, Betts V, Smith B. Change in volume of lumpectomy cavity during external-beam irradiation of the intact breast. Int J Radiat Oncol Biol Phys 2006;65:1161–1164. 4. Goldberg H, Prosnitz RG, Olson JA, et al. Definition of post lumpectomy tumour bed for radiotherapy boost field planning: CT versus surgical clips. Int J Radiat Oncol Biol Phys 2005; 63:209–213. 5. Coles CE, Wishart G, Donovan E, et al. The IMPORT gold seed study: Evaluation of tumour bed localisation and image-guided radiotherapy for breast cancer. Clin Oncol 2007;19(Suppl.): S26–S27. 6. Coles CE, Wishart GC, Fairfoul J, et al. Implanted gold seeds for tumour bed localisation and image-guided radiotherapy [Abstract]. EJC Supplements 2007;5:6.
7. Coles CE, Cash CJC, Treece GM, et al. High definition three-dimensional ultrasound to localise the tumour bed: A breast radiotherapy planning study. Radiother Oncol 2007;84:233–241. 8. Kass R, Kumar G, Klimberg VS, et al. Clip migration in stereotactic biopsy. Am J Surg 2002;184:325–331. 9. Tersteeg JHA, Roesink JM, Albregts M, et al. Changes in excision cavity volume: Prediction of the reduction in absolute volume during breast irradiation. Int J Radiat Oncol Biol Phys 2007;69(Suppl.):S209–S210. 10. Bland JM, Altman DG. Applying the right statistics: Analyses of measurement studies. Ultrasound Obstet Gynecol 2003;22:85–93. 11. Tuzikov AV, Roerdink J, Heijmans H. Similarity measures for convex polyhedra based on Minkowski addition. Pattern Recognition 2000;33:979–995. 12. Klain DA. The Minkowski problem for polytopes. Adv Math 2004;185:270–288. 13. Oh KS, Kong FM, Griffith KA, et al. Planning the breast tumor bed boost: Changes in the excision cavity volume and surgical scar location after breast-conserving surgery and whole-breast irradiation. Int J Radiat Oncol Biol Phys 2006;66:680–686.
doi:10.1016/j.ijrobp.2008.10.030
PHYSICS CONTRIBUTION
CHARACTERIZATION OF TARGET VOLUME CHANGES DURING BREAST RADIOTHERAPY USING IMPLANTED FIDUCIAL MARKERS AND PORTAL IMAGING EMMA J. HARRIS, PH.D.,* ELLEN M. DONOVAN, PH.D.,* JOHN R. YARNOLD, M.D.,y CHARLOTTE E. COLES, M.D.,z AND PHILIP M. EVANS, D.PHIL., * ON BEHALF OF THE IMPORT TRIAL MANAGEMENT GROUP y
* Joint Department of Physics, Institute of Cancer Research and Royal Marsden Foundation Trust, Sutton, United Kingdom; Department of Clinical Oncology, Royal Marsden Foundation Trust, Sutton, United Kingdom; and z Department of Oncology, Cambridge University Hospitals NHS Foundation Trust, Cambridge, United Kingdom Purpose: To determine target volume changes by using volume and shape analysis for patients receiving radiotherapy after breast conservation surgery and to compare different methods of automatically identifying changes in target volume, position, size, and shape during radiotherapy for use in adaptive radiotherapy. Methods and Materials: Eleven patients undergoing whole breast radiotherapy had fiducial markers sutured into the excision cavity at the time of surgery. Patients underwent imaging using computed tomography (for planning and at the end of treatment) and during treatment by using portal imaging. A marker volume (MV) was defined by using the measured marker positions. Changes in both individual marker positions and MVs were identified manually and using six automated similarity indices. Comparison of the two types of analysis (manual and automated) was undertaken to establish whether similarity indices can be used to automatically detect changes in target volumes. Results: Manual analysis showed that 3 patients had significant MV reduction. This analysis also showed significant changes between planning computed tomography and the start of treatment for 9 patients, including single and multiple marker movement, deformation (shape change), and rotation. Four of the six similarity indices were shown to be sensitive to the observed changes. Conclusions: Significant changes in size, shape, and position occur to the fiducial marker–defined volume. Four similarity indices can be used to identify these changes, and a protocol for their use in adaptive radiotherapy is suggested. Ó 2009 Elsevier Inc.
Excision cavity, Partial breast irradiation, Fiducial markers, Shape analysis, Breast radiotherapy.
mography (CT) to localize the excision cavity, and margins to encompass possible subclinical spread of disease are added to create the clinical target volume (CTV). Using fiducial markers, the center of mass (CoM) of the markers at the time of treatment can be compared with the CoM at the time of planning to calculate daily shifts in the position of the excision cavity (6, 7). In addition, changes in marker positions may arise because of deformation of the excision cavity (3) and fiducial marker migration (8, 9), which lead to changes in the size, shape, and measured CoM of the measured volume. Deformation may result in underdosing of the CTV or overdosing of healthy adjacent tissues. Marker migration
INTRODUCTION Radiotherapy of the breast after breast conservation surgery is adjuvant treatment because the gross tumor volume has been excised, leaving possible microscopic disease. Standard breast radiotherapy treats the whole breast, but the region with the greatest risk of tumor recurrence is proximal to the excision cavity, and this may be treated with a higher dose of radiation. The excision cavity may be localized by using imaging techniques to visualize seroma (1, 2), changes in tissue architecture (3, 4), or fiducial markers implanted into the breast at the time of surgery (5). The most common method uses a combination of fiducial markers (typically surgical clips) and soft-tissue changes imaged using computed toReprint requests to: Emma J. Harris, Ph.D., Institute of Cancer Research, Joint Department of Physics, Cotswold Road, Sutton, SM2 5PT, UK. Tel: (++44) 208-661-3478; fax: (++44) 208-643-3812; E-mail: [email protected] IMPORT and the Institute of Cancer Research are funded by Cancer Research UK. Charlotte Coles received funding from the
Cambridge National Institute of Health Research Biomedical Research Centre. Conflict of interest: none. Received June 27, 2008, and in revised form Sept 9, 2008. Accepted for publication Oct 21, 2008. 958
Excision cavity changes d E. J. HARRIS et al.
(a)
(b)
959
(c)
Fig. 1. Schematic two-dimensional diagram of the excision cavity boundary (dashed lines) and six fiducial markers (circles) shows the variation in excision cavity volume contained within the marker volume (MV). (a, b) Two possible MVs and (c) the MV containing the excision cavity volume chosen for this work.
will change the shape of the measured volume and its CoM. Consequently, these may differ from the shape and CoM of the excision cavity as defined at the initial CT planning scan. Measurement and characterization of changes in marker positions during treatment may help identify time trends in changes in the excision cavity, characterize deformation, and identify marker migration. Using this information, strategies may be developed so that a response to changes in the excision cavity may be made and the treatment may be tailored to the individual patient. The aims of this study are to: (1) manually analyze target volume, shape, and position based on implanted fiducial markers imaged using projection x-rays to determine what changes occur during treatment; (2) investigate automated methods of comparing target volumes and positions by using similarity indices to automatically detect changes in volume position, shape, and orientation; and (3) recommend an image-guided radiotherapy protocol for breast radiotherapy incorporating the use of similarity indices. METHODS AND MATERIALS The measured positions of implanted fiducial markers were used to define the marker volume (MV), which acts as a surrogate for the excision cavity, which cannot be visualized using portal imaging (PI). This was based on the envelope of the markers to obtain the maximum coverage of the target volume, shown in Fig. 1. We evaluated changes to the MV. Four methods (resulting in six possible similarity indices) were used to compare the MV defined using CT at the time of planning (Volume A) with that defined during treatment, imaged using PI (Volume B). The first method uses the positions of the fiducial markers and the other three use the MV. The methods used were: (1) analysis of the second-order moments of the positions of markers in A and B in the three spatial dimensions (first-order moment gives the CoM and has been studied by Coles et al. [6]); (2) volume analysis, including the measured volume of A and B and a volume index (VI) that gives the fractional change in volume of the MV between planning and treatment images; (3) the Jaccard Index (JI), which describes the volume overlap of A and B; and (4) the Minkowski Index (MI), which is sensitive to changes in volume shape only. If the CoMs of A and B are shifted to a common origin, the analysis will be independent of any translation between the two volumes. For Methods 1 (second-order moments), 2 (volume analysis), and 4 (MI), a CoM shift was applied before the analysis. For Method 3 JI was calculated before (JI1) and after (JI2) the CoM shift.
Patient study At The Royal Marsden Hospital, Sutton, United Kingdom, 11 patients gave informed consent to be entered into the National Cancer Research Network multi-centre ‘‘GOLD SEED’’ feasibility study evaluating the use of implanted fiducial markers for image-guided radiotherapy in patients with early breast cancer (5). After wide local excision of the primary tumor, gold fiducial markers were sutured around the excision cavity following a strict surgical procedure. The markers were gold rings 3 mm in diameter, with a 1-mm central hole and 1-mm thickness. The intended number of markers was six, but there were fewer markers in some patients because of re-excision or surgeon preference. Therefore, of the 11 patients recruited, 7 had six markers inserted, 4 had five markers inserted, and 1 patient had four markers sutured into the cavity. All patients underwent CT scanning before treatment (CT1) and in the last week or just after the end of their treatment (CT2). The CT axial slice separation was 1.25 mm. The CT images were exported to the Pinnacle Treatment planning system (Phillips Medical Systems, Milpitas, CA). Marker positions relative to the treatment isocenter were identified by manually locating the center of each marker on the treatment planning system. Median time between surgery and CT1 was 88 days (range, 30–169 days). Three patients received chemotherapy, with 155, 157, and 169 days between surgery and CT1 for these patients. Ten patients received hormonal therapy. Patients were treated by using three fractionation regimens. These were the current international standard of 50 Gy in 25 fractions and two regimens from randomized controlled trials of 30 Gy in five fractions and 40 Gy in 15 fractions. No changes were made to their treatment based on the fiducial markers. Portal images were acquired on the first treatment day and between 4 and 13 subsequent treatment days. Median time between CT1 and the first day of treatment was 13 days. The images were acquired using an iViewGT electronic portal imaging device (Elekta Oncology Systems Ltd., Crawley, UK). Two orthogonal portal images were acquired to locate the markers in three dimensions. One image was acquired using the anterior tangential treatment beam. The second image was acquired with the gantry placed orthogonal to this beam (using 1 MU). Examples of orthogonal portal images of the implanted fiducial markers are shown in Fig. 2. Each marker position was calculated using triangulation with inhouse software. Observer errors associated with the measurements were evaluated by three physicists locating the markers three times each by using both CT and PI and determining the coefficient of repeatability (10). Changes in MVs throughout treatment were made by analyzing the patient data manually. These observed changes were used as the gold standard with which we compared the changes identified by using the similarity indices.
I. J. Radiation Oncology d Biology d Physics
960
Volume 73, Number 3, 2009
ment of the markers, di, was measured. If we replace d with d in Eqs. 1 and 2, we may calculate the variance in marker movements, m2(AB), and define similarity index Dm, to be: Dm ¼ m2 ðA BÞ þ 1
(Eq. 4)
(where the addition of 1 is used to ensure Dm is 1 if the volumes are identical). This pairwise approach was adopted to detect changes in the position of a single marker between images of A and B as an indication of single-marker migration. As with m, this may be calculated in each spatial direction. Volume analysis. The MVs were calculated for all scans. For two volumes V(A) and V(B), VI is given by: VIðA; BÞ ¼
Fig. 2. (a) Right and (b) left anterior oblique portal image (1 MU). Portal images were acquired at orthogonal angles. Translation was measured using the CoM shifts. A significant translation is defined as a shift greater than 5 mm (the local setup tolerance). The median magnitude of rotation was 5 ; hence, this was used as the threshold to identify MVs with large rotation. Significant volume reduction is defined as a volume reduction outside the 95% confidence intervals (CIs). Single-marker movement is identified if a single marker has moved more than 2.5 mm (three times the mean total observer error) with respect to its CT1 position. Multiplemarker movement is defined as the movement of two or more markers by the same distance. A change was defined as being characteristic for a patient’s MV if observed in more than 50% of cases.
Comparison methods Second-order moment analysis. The second-order central moment of a distribution of N points is its variance and is given by: m2 ¼
N 1X ðdi m1 Þ2 N i¼1
JIðA; BÞ ¼
The position of di is specified in each of the three spatial directions: left-right (LR), superior-inferior (SI), and anterior-posterior (AP), allowing m2 to be determined in each direction (to limit the number of variables, cross-moments were not considered). Two methods of comparing Volumes A and B were considered: first, comparison of the whole volume, and second, pairwise comparison of markers. For the first method, the variance index, m, was defined: m¼
m2 ðBÞ m2 ðAÞ
(Eq. 3)
where m2(A) and m2(B) are the variance in marker positions in A and B, respectively. The m was evaluated in each direction (LR, SI, and AP). This is expected to be sensitive to changes in rotation and deformation. For the pairwise comparison of marker positions, by subtracting marker positions in Volume B from those in Volume A, the move-
jAXBj jAWBj
(Eq. 6)
The JI measures the overlap of the two volumes and a JI of unity indicates that the two volumes are identical in size, shape, position, and orientation and will be less than unity if any of these parameters differ, i.e., the ideal value of the JI is 1. The JI was evaluated both before (JI1) and after (JI2) the CoMs of the markers in Volumes A and B were shifted to a common origin. The JI1 measured differences in volume overlap between the CT and PI volumes without setup correction. Values of JI1 less then 1 may be caused by translational shifts and changes in size, shape, and orientation, whereas JI2 measures changes in size, shape, and orientation of the volumes only. Neither of these indices can be used to determine specific changes, only to indicate that one of these changes has occurred. Minkowski Index. The Minkowski Index (MI) measures the similarity in shape of two volumes (A and B) and is given by (11):
(Eq. 1)
(Eq. 2)
(Eq. 5)
(as with Eq. (4), the addition of 1 ensures identical volumes have an index of 1). The VI is constructed to be sensitive to change in volume only. Jaccard Index. The JI is defined as the intersection divided by the union of two sample sets. For Volumes A and B:
2
1
VðBÞ3 VðAÞ3 MIðA; BÞ ¼ Vm ðA; BÞ
where di is the position of point i and m1 is the CoM, given by: N 1X m1 ¼ ðdi Þ N i¼1
VðAÞ VðBÞ þ1 VðAÞ
(Eq. 7)
where Vm(A,B) is the Minkowski mixed volume for Volumes A and B (12). Volumes A and B are corrected for any difference in position or orientation: Volume B is rotated until its orientation matches that of A as closely as possible. The MI is sensitive to differences in shape only and therefore can be used to distinguish between changes in shape and other changes that occur. If no shape change has occurred, MI will be equal to 1.
Model analysis The effects of stochastic variations in measured marker positions caused by observer variability were modeled to calculate the resulting uncertainty in the volumes and similarity indices measured for the patient data and evaluate CIs to represent threshold values for each index for each patient MV; we define no change as a value of similarity index that lies within the 95% CI. For each patient’s MV, 10,000 sample volumes were generated by adding total observer error to each marker position in the LR, SI, and AP directions. Observer errors were generated by randomly sampling a normal distribution with the standard deviation equal to the total observer error (10). Similarity indices were calculated for
Excision cavity changes d E. J. HARRIS et al.
961
Table 1. Observer coefficients of repeatability for CT and PI measurements Coefficient of repeatability (mm)
Intraobserver CT/PI (Observer 1) CT/PI (Observer 2) CT/PI (Observer 3) Interobserver CT PI Total (mean intra- + interobserver) CT/PI/CT + PI
Left-right
Superior-inferior
Anterior-posterior
0.3/1.0 0.6/0.4 0.8/1.4
0.2/0.4 0.5/0.2 0.9/0.7
0.5/1.1 0.6/0.4 0.8/1.1
1.0/1.8
0.6/1.0
1.1/1.9
1.1/2.0/2.2
0.8/1.1/1.4
1.3/2.1/2.2
Abbreviations: CT = computed tomography; PI = portal imaging.
the 10,000 new volumes. It was found empirically that the resulting non-normal distribution could be transformed into a normal distribution by evaluating the cubed root of the index values. The resulting 95% confidence limits of this distribution were then found. Confidence limits were transformed back to give the 95% CI for each similarity index.
RESULTS Observer error The coefficients of repeatability (SD 1.96) in each direction are listed in Table 1 for both CT and PI for each observer (intraobserver) and all observers (interobserver). The total coefficient of repeatability (CT + PI) is used to calculate error in the similarity indices evaluated for measured MVs. Patient MVs and volume changes during treatment The implanted markers did not define a regularly shaped MV, as shown for Patients 3 and 6 in Fig. 3a and b. For Patient 3, the outer four markers describe the largest MV and therefore two markers lie completely inside the MV (c.f., Fig. 1c). The MV(CT1) for Patient 6 is an example of an MV formed from all six markers. It can be seen that the MV surface comprises a set of triangular faces, and MVs
can be described by using the greatest width and length of the largest face and the distance of the marker farthest from this face. The dimensions of the MV(CT1) and the number of markers implanted for each patient are listed in Table 2. Five types of change were observed: (1) translation, (2) rotation, (3) volume change, (4) single-marker movement, and (5) multiple-marker movement (two or more markers). The observed MV changes between CT1 and (1) first day of treatment (PI first) and (2) all MVs measured during treatment are listed in Table 2. Column 5 in Table 2 lists changes observed between CT1 and PI first. Maximum measured values are given if rotation, marker movement, or volume change is observed. Column 6 of Table 2 lists changes for all MVs measured during treatment. In Column 6, values listed are maximum values for all PI and CT2 MVs. Translations were observed in the majority of patient MVs. These were excluded from further analysis because they may be corrected by using couch shift. Two patients (Patients 4 and 11) had no substantial change between CT1 and treatment, 1 patient (Patient 6) had rotation only, and the remaining patients had some marker movement. The MVs for Patients 1 and 10 were almost planar, with thicknesses less than 1 mm.
Fig. 3. Schematic diagrams of implanted fiducial marker positions and marker volumes (MVs) for Patients (a) 3 and (b) 6. Rings represent the fiducial markers and lines show boundaries of the MVs. Shadow projections of the MVs are shown. AP = anterior-posterior; LR = left-right; SI = superior-inferior.
I. J. Radiation Oncology d Biology d Physics
962
Volume 73, Number 3, 2009
Table 2. Description of patient MV(CT1) and observed changes between CT1 and treatment and during treatment Patient No.
Dimensions of planning MV* (cm)
Volume (cm3) No. of markers
Observed MV changes (comparison of MV[CT1] and MV[PI first])
Observed MV changes during treatment (characteristic of >50% of treatment MVs)
Rotation (6.5 ) Thin MV: large variation in shape & volume, single-marker movement (2.7 mm) Single-marker movement (6.7 mm) No change Volume shrinkage (31%) Rotation (6.1 ) Rotation (8 ) Single-marker movement (3.7 mm)
1 2
2.2/1.5/0.1 (Planar MV) 4/1.8/0.3
0.16 0.64
5 5
3
4/1.3/0.7
1.11
6
4 5 6 7
4/2.2/1.2 0.6/1.5/0.6 2.5/1/1 3.5/3.5/0.5
1.45 0.35 0.52 1.85
6 5 6 6
8 9
4/2/1 6/6/4
2.03 25.62
6 6
Volume shrinkage (13%) Volume shrinkage (47%) Multiple marker movement (9.1 mm)
10
2.8/2.5/0.1 (Planar MV)
0.9
4
Rotation (6 ) Single-marker movement (3.3 mm)
11
2/2.5/0.4
0.6
5
No change
Rotation (maximum, 7.1 ) Single-marker movement (maximum, 4.2 mm) Multiple-marker movement (maximum, 7 mm) No change Volume shrinkage (60%) Rotation (maximum, 11.5 ) Rotation (maximum, 8.4 ) Multiple-marker movement (maximum, 5.2 mm) Volume shrinkage (30%) Rotation (maximum, 10.2 ) Volume shrinkage (65%) Multiple-marker movement (maximum, 13.3 mm) Rotation (maximum, 12 ) Multiple-marker movement (maximum, 4.6 mm) No change
Abbreviations: MV = marker volume; CT1 = computed tomography scanning before treatment; PI = portal imaging. * Dimensions given are length and width (see text for explanation) of largest MV face and distance of the marker farthest from this face.
it lies outside the 95% CI. There is good agreement between CT2 and the last days of treatment, indicating no systematic difference between the two imaging methods. Table 3 lists percentages of MVs with ‘‘true results’’ for each patient, in which true results are defined as a correct prediction of a change or no change using the similarity index and false results give an incorrect prediction. We now consider each of the indices in turn. 30 25 20 15 10 5
Volume of MV (cm3)
Figure 4 shows volume of MV as a function of time. The 95% CIs are shown as error bars for the MV measured for CT1 and CT2 (indicated by using larger data points) and on the first and last PI days. With the exception of Patient 9, who had a large MV, observer error produces large uncertainty in measured volumes because of the relatively small volume of the MVs compared with the magnitude of the observer errors. There was a significant (p < 0.05) change in volume size between CT1 (Day 0) and the first day of treatment for Patients 5, 8, and 9. The MV for Patient 9 decreased between CT and treatment by 47% and by a total of 65% by the end of treatment. This patient had a large specimen size (100 g), and a large volume of seroma is visible in CT1 (this patient did not undergo a second CT scan). The MVs for Patients 5 and 8 shrink in volume by the end of treatment by 60% and 30%, respectively. Patient 2 had a very thin volume that shows a large fluctuation (by >40% on Days 10, 21, and 38 [CT2]). This was a result of the movement of the most superior marker in the SI direction by 2 mm (shown in Fig. 5).
2.5 Patient 1 2 3 4 5 6 7 8 9 10 11
2.0
1.5
1.0
0.5
Similarity indices for patient data Similarity indices have been calculated between MV(CT1) and treatment MVs for all patients. Values for the first PI, last day of PI, and CT2 are shown in Fig. 6. Both m and Dm were calculated in all three spatial directions, and only the spatial direction with the greatest change is given. Patients are grouped by type of MV change. We define a significant change in index value (i.e., significantly different from 1) if
0.0 0
10
20
30
40
50
""
60
Days from CT1
Fig. 4. Volume of patient marker volumes (MVs) as a function of time. Error bars show 95% confidence intervals for computed tomography (CT) scanning before treatment (CT1) (Day 0), MV(portal imaging [PI] first), MV(PI last), and CT in the last week or just after the end of treatment (CT2; large data points).
Excision cavity changes d E. J. HARRIS et al.
963
Fig. 5. (a) Left anterior oblique digitally reconstructed radiograph (Patient 2) from computed tomography (CT) scanning before treatment (CT1) and portal images of markers acquired 8 (b) and 10 (c) days after CT1. Implanted fiducial markers are circled. The change in the most superior marker is seen.
Variance index, m. The m indicates marker movement and volume change. From Fig. 6, m indicates change for MV(PI first), MV(PI last), and MV(CT2) for all patients for whom this happens. The m gives false-positive results for Patient 11 (PI first) and false-negative results for Patients 2 (PI first, PI last, and CT2) and 3 (PI first), both of whom have singlemarker movement. The m also gives false results for Patients 8 and 10 (Table 3). In the case of Patient 2, single-marker movement has been identified (Table 2); however, this marker lies within the MV and as a consequence, variance, shape, and volume do not change. For Patient 3, marker movement is in the direction of the largest dimension and the variance is not affected until one marker moves more than 5 mm. Marker movement index, Dm. The Dm should also indicate marker movement or volume change. This index indicates changes for all MVs shown in Fig. 6 with the exception of Patient 6. False-positive results are given for Patients 2, 4, 6, and 11. Jaccard Indices, JI1, and JI2. The JI should indicate translation, deformation, volume change, or rotation. All patients had true-positive results for JI1. From Fig. 6, JI2 indicates change for MV(Day 1) for Patient 1; this is in agreement with the observed rotation. The JI2 also correctly indicates no change for Patients 4 and 11. False-negative results were seen for Patients 2, 3, 4, 6, and 8 (Table 3). For Patient 2, the single-marker movement is inside the MV and hence does not change the volume. Hence, this single-marker movement was not detected. From the manual analysis, Patient 3 has a high incidence of multiple-marker movement, which was not detected by JI2. This also was the case for MV(CT2) for Patient 4. For Patient 6, rotations of 6.2 and 6.3 were
not detected. For Patient 8, single-marker movement of 2.55 mm was not detected for one MV only. Volume Index. The VI gives an incidence of true results of 100% for all patients. Minkowski Index. The MI should indicate shape change only. For all except Patients 2 and 3, the MI gives an incidence of true results of 100%. False-negative results for Patient 2 occurred because of single-marker movement inside the MV (as described). Marker movement for Patient 3 also was not detected by using the MI because the shape does not change significantly. From this analysis, a single index cannot distinguish between single-marker movement and deformation. It was shown that in some cases, JI2 and the MI do not indicate marker movement when MV shape does not change. However, these changes can be detected by means of Dm. DISCUSSION A significant reduction in the size of the MV was observed for 3 patients. Other studies have measured the reduction in volume of the excision cavity during breast radiotherapy for a larger proportion of patients. Using CT scanning, Jacobson et al. (3) found that of 20 patients, 16 had a greater than 20% decrease in excision cavity volume 4–5 weeks into radiotherapy. Oh et al. (13) performed a similar analysis for 31 patients and found a mean reduction of 22.5% in excision cavity volume. In both studies, excision cavities were defined by using surgical clips, seroma, and anatomic changes seen in the CT. Mean MVs measured by Jacobson et al. (3) and Oh et al. (13) and in this study were 36.1, 36.6, and 3.12 cm3, respectively. Median time between CT1 and surgery
964
I. J. Radiation Oncology d Biology d Physics
Volume 73, Number 3, 2009
PI first PI last CT2 2
3 2 1 0
6
1 0 2
8 7
1
4
0
0 Rotation
3 2 1 0 2
20
10
10 0 20
3
1
10
0
0
2
80
9
1
40
0
0
2
8
5
1
4
0
0
2
8
8
1
4
0
0
2
8
2
1
4
0
0
2
Single Marker Movement
6 4 2 0
4
1 0 2 1
Volume reduction
4 11
No change
2
0
0 JI1
JI2
VI
MI
m
Δm
Fig. 6. Similarity index values for marker volumes (MVs) on the first (portal imaging [PI] first) and last day of treatment (PI last) and for computed tomography (CT) in the last week or just after the end of treatment (CT2). The similarity index Dm bars correspond to the right axis. All other index values correspond to the left axis. Error bars = uncertainty in measurement caused by observer error; solid black boxes = index values that lie outside the threshold values (95% confidence interval). JI = Jaccard Index; VI = volume index; MI = Minkowski Index.
was 88 days for this study and 60.9 for the study by Oh et al. (13) (these data were not given by Jacobson et al. [3]); therefore, shrinkage during this longer period may account for the difference in volume. However, because of the magnitude of the difference in volume (factor > 10), it is more likely that we measured smaller volumes because we use only the marker positions to define volume, and not seroma. Furthermore, our surgeons aim to obliterate the excision cavity to reduce seroma; thus, we would expect MVs to be small. Observer error in marker location by using PI was found to be less than 1 mm. However, the small dimensions of the MVs (2–60 mm) mean that this error creates large uncertainties in volume measurement and evaluation of similarity indices. Therefore, in this study, uncertainties in measured
MV are on the order of 20%, and volume reduction for some patients therefore may be undetectable. However, it is possible that the volumes used in this study do not tend to decrease by significant amounts because they are not defined by using seroma, the volume of which we would expect to decrease with time after surgery. In our study, we use treatment PI to measure marker positions. Our treatment PI had an acquisition time of 4.56 seconds, which is on the order of one breathing cycle. Therefore, any error in marker position measurement caused by infraction motion is included in the observer error (markers were slightly blurred in the images). If shorter PI acquisition times are used, it will be necessary to consider intrafraction motion in the estimation of measurement errors.
Excision cavity changes d E. J. HARRIS et al.
Table 3. Percentage of true results obtained by using similarity indices for marker volumes Patient 1 2 3 4 5 6 7 8 9 10 11
m
Dm
JI1
JI2
VI
MI
100 36 54 100 100 100 100 70 100 60 72
100 93 100 92 100 87 100 100 100 100 63
100 100 100 100 100 100 100 100 100 100 100
100 71 45 83 100 87 100 92 100 100 100
100 100 100 100 100 100 100 100 100 100 100
100 65 43 100 100 100 100 100 100 100 100
Abbreviations: m = variance index; Dm = marker movement index; JI1 = Jaccard Index before the center of mass shift; JI2 = Jaccard Index after the center of mass shift; VI = volume index; MI = Minkowski Index. The 100% indicates that all marker volume changes were identified correctly by using the similarity index.
Migration of surgical clips previously was identified in two studies (7, 8). However, this feasibility study is the first to use markers sutured into the excision cavity wall. If we consider that the greatest amount of single-marker movement detected was 7 mm for Patient 2 and multiple-marker movements of up to 13.3 mm were also observed, it is reasonable to assume that the single-marker movement observed for Patient 2 is caused by anatomic changes, not marker migration. Large
Measure marker positions
965
volume fluctuations observed for Patient 2 were not a result of the 7-mm marker movement (this marker was inside the MV), but the 2-mm movement of the most superior. This marker moved back and forth in the SI direction during treatment. This motion may be caused by slight changes in the patient’s arm position. In practice, the CTV will be derived from the MV, seroma, and any anatomic changes evident on CT. However, the MV can be used in conjunction with the similarity indices to indicate anatomic changes. With the exception of m, similarity indices agreed well with the observed changes. The JI1 was found to be very sensitive to shifts in patient position. These shifts also can be measured by using the CoM. In some cases, marker movements may not be indicated by JI2 or MI. However, Dm will be sensitive to these changes. The MI identifies shape change only. These methods will not allow automatic distinction between single- and multiple-marker movement. To apply these techniques to adaptive radiotherapy, we suggest the verification protocol shown in Fig. 7. Interventions are shown by grey trapezoids and include repositioning, investigation into the need to replan a treatment, and possible marker migration and/or CoM changes. After initial corrections for CoM shifts and rotations, VI is used to detect volume change, JI2 is used to detect changes in rotation or shape, MI detects shape change (if JI2 is positive and MI is negative, changes in JI2 must be caused by rotation), and finally, Dm is used to detect marker movement and therefore possible migration.
Calculate VI, JI2, MI and Δm
VI change?
Calculate COM
yes
Volume change
no
yes >5mm?
Reposition
JI2 change? no
no
MI change?
Calculate Rotation
yes
MI change?
Significant rotation
yes
no
Check for re-plan
yes Reposition
Δm change?
no
Shape change
no >5°?
yes
yes
no Marker migration?
Check for systematic error in CoM / migration
Treat
Fig. 7. Work flow diagram represents suggested protocol for use of the similarity indices volume index (VI), Jaccard Index analysis performed after the center of mass (COM) shift (JI2), Minkowski Index (MI), and similarity index Dm in an adaptive radiotherapy scheme.
966
I. J. Radiation Oncology d Biology d Physics
Volume 73, Number 3, 2009
CONCLUSIONS
that by using a combination of similarity indices, we can iden-
Significant changes in size, shape, and position occur to the fiducial marker–defined volume during radiotherapy. We show
tify when changes in size and shape have occurred and devise a protocol to use these methods for adaptive radiotherapy.
REFERENCES 1. Ringash J, Whelan T, Elliott E, et al. Accuracy of ultrasound in localization of breast boost field. Radiother Oncol 2004;72: 61–66. 2. Wong EK, Truong PT, Kader HA, et al. Consistency in seroma contouring for partial breast radiotherapy: Impact of guidelines. Int J Radiat Oncol Biol Phys 2006;66:372–376. 3. Jacobson G, Betts V, Smith B. Change in volume of lumpectomy cavity during external-beam irradiation of the intact breast. Int J Radiat Oncol Biol Phys 2006;65:1161–1164. 4. Goldberg H, Prosnitz RG, Olson JA, et al. Definition of post lumpectomy tumour bed for radiotherapy boost field planning: CT versus surgical clips. Int J Radiat Oncol Biol Phys 2005; 63:209–213. 5. Coles CE, Wishart G, Donovan E, et al. The IMPORT gold seed study: Evaluation of tumour bed localisation and image-guided radiotherapy for breast cancer. Clin Oncol 2007;19(Suppl.): S26–S27. 6. Coles CE, Wishart GC, Fairfoul J, et al. Implanted gold seeds for tumour bed localisation and image-guided radiotherapy [Abstract]. EJC Supplements 2007;5:6.
7. Coles CE, Cash CJC, Treece GM, et al. High definition three-dimensional ultrasound to localise the tumour bed: A breast radiotherapy planning study. Radiother Oncol 2007;84:233–241. 8. Kass R, Kumar G, Klimberg VS, et al. Clip migration in stereotactic biopsy. Am J Surg 2002;184:325–331. 9. Tersteeg JHA, Roesink JM, Albregts M, et al. Changes in excision cavity volume: Prediction of the reduction in absolute volume during breast irradiation. Int J Radiat Oncol Biol Phys 2007;69(Suppl.):S209–S210. 10. Bland JM, Altman DG. Applying the right statistics: Analyses of measurement studies. Ultrasound Obstet Gynecol 2003;22:85–93. 11. Tuzikov AV, Roerdink J, Heijmans H. Similarity measures for convex polyhedra based on Minkowski addition. Pattern Recognition 2000;33:979–995. 12. Klain DA. The Minkowski problem for polytopes. Adv Math 2004;185:270–288. 13. Oh KS, Kong FM, Griffith KA, et al. Planning the breast tumor bed boost: Changes in the excision cavity volume and surgical scar location after breast-conserving surgery and whole-breast irradiation. Int J Radiat Oncol Biol Phys 2006;66:680–686.