- Email: [email protected]
Intensity-modulated radiotherapy of breast cancer using direct aperture optimization
Radiotherapy and Oncology 79 (2006) 162–169 www.thegreenjournal.com
Breast IMRT
Intensity-modulated radiotherapy of breast cancer using direct aperture optimization Bram van Asselen, Marco Schwarz, Corine van Vliet-Vroegindeweij, Joos V. Lebesque, Ben J. Mijnheer, Eugene M.F. Damen* Department of Radiation Oncology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
Abstract Background and purpose: To design a clinically reliable and efficient step-and-shoot IMRT delivery technique for the treatment of breast cancer using direct aperture optimization (DAO). Using DAO, segments are created and optimized within the same optimization process. Patients and methods: The DAO technique implemented in the Pinnacle treatment planning system, which is called direct machine parameter optimization (DMPO), was used to generate IMRT plans for twelve breast cancer patients. The prescribed dose was 50 Gy. Two DMPO plans were generated. The first approach uses DMPO only; the second technique combines DMPO with two predefined segments (DMPOsegm), having shapes identical to the conventional tangential fields. The weight of these predefined segments is optimized simultaneously with DMPO. The DMPO plans were compared with normal two-step (TS) IMRT, creating segments after optimizing the intensity. Results: Dose homogeneity within the target volume was 4.8G0.6, 4.3G0.5 and 3.8G0.5 Gy for the TS, DMPO and DMPOsegm plans, respectively. Comparing the IMRT plans with an idealized dose distribution obtained using only beamlet optimization, the degradation of the dose distribution was less for the DMPO plans compared with the two-step IMRT approach. Furthermore, this degradation was similar for all patients, while for the two-step IMRT approach it was patient specific. Conclusions: An efficient step-and-shoot IMRT solution was developed for the treatment of breast cancer using DMPO combined with two predefined segments. q 2006 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 79 (2006) 162–169. Keywords: IMRT; Breast; Direct aperture optimisation; Step-and-shoot
Radiotherapy is an effective treatment modality for the treatment of breast cancer [16,21] and is a widely accepted treatment modality for early-stage breast cancer. Patients are usually treated using two tangential wedged beams. The use of wedged fields can result in a heterogeneous distribution, particularly in cranial and caudal parts of the breast, where low and high dose areas can occur. An overdosage may result in worse cosmetic results after irradiation [14,15,23] and underdosage may result in a lower tumor control probability. Using intensity modulated radiation therapy (IMRT), the homogeneity of the dose distribution in the breast can be improved compared with the use of wedged beams [1,5,9–11,13,20]. Promising data were published on cosmetic results of patients treated with IMRT [22] and a better breast appearance using IMRT was reported in another study [25]. Although it has been shown that IMRT can improve the dose distribution, a number of institutes still hesitate to introduce IMRT for breast cancer treatment for several
reasons. First, a planning CT scan has to be performed of each patient, and a target volume has to be delineated. This may increase the workload, since breast patients are a relatively large patient group in many departments. Furthermore, individual delineation may result in uncertainties in target volume definition [8]. Finally, in most treatment planning systems relatively simple dose calculation algorithms are used for inverse planning, while after conversion the dose is usually recalculated using a more accurate dose calculation. Due to different modeling of the tissue inhomogeneity of the lung, missing tissue and oblique incidence, the quality of the plan may degrade after recalculation. The conversion of the intensity profile to deliverable segments may also cause a degradation of the plan, since the intensity profile delivered by the segments is not exactly the same as the optimized intensity map. This can be the result of a limited number of intensity levels used for sampling the intensity profile, but also due to the fact that leakage, head scatter and tongue-and-groove effect are
0167-8140/$ - see front matter q 2006 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2006.04.010
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
not taken into account during optimization. Furthermore, the leaf-sequencing process should be efficient, since many segments and monitor units (MUs) might not be clinically acceptable. Too many MUs introduce a larger scattered dose component, resulting in an increase in dose to surrounding tissues such as the contralateral breast. Institutes treating breast cancer with IMRT developed their own IMRT methods to solve one or more of the abovementioned issues [1,5,10,11,13,20]. For step-and-shoot IMRT techniques, the segments shapes are usually created using isodose surfaces, anatomy or radiological thickness. The weight of these segments is then optimized to achieve a homogeneous dose distribution. Segments are thus created before the optimization starts and are not part of the optimization itself. In this article, we will focus on step-and-shoot IMRT techniques for the irradiation of the breast. The main focus of this paper was on the use of direct aperture optimization (DAO). DAO is a method for step-and-shoot IMRT and has shown to be an efficient method for IMRT delivery [2,3,12,18,19]. Using DAO, the leaf positions and the weight of each segment are parameters of the optimization, therefore eliminating a separate leaf sequencing step after fluence optimization. The segment shapes and weights are thus part of the optimization cycle. It was the purpose of this study to design a simple, efficient and clinically acceptable IMRT technique for irradiation of the breast using DAO. The DAO plans will be tested separately as well as in combination with predefined segments. The results will be compared with the more classical two-step approach of IMRT, where segments are created after the optimization process. Since, the dose calculation algorithm used during optimization may be an important factor for IMRT of the breast, the effect of using a convolution–superposition algorithm during the inverse planning process was also tested relative to the use of a simple pencil model for the classical two-step approach.
Methods and materials Patients Twelve patients, previously treated for left-sided breast cancer, were selected for this study. All patients underwent a computer tomography (CT) scan, with a 5 mm slice interval. The images were obtained with the patient positioned supine with their arms placed above the heads in an arm-rest, equivalent to the treatment position. The scan was transferred to the Pinnacle treatment planning system (Philips Medical Systems, Best, the Netherlands; version 7.4). Before CT-scanning the palpated breast tissue was outlined using a lead wire. The CT scan included the heart and both lungs.
Volume definition For large-scale implementation of an IMRT technique, the planning time should not significantly increase compared to conventional treatment planning. Delineation of the breast tissue for each individual patient by a physician may therefore not be desirable. Furthermore, the interobserver variation has shown to be large [8]. However, using an
163
Fig. 1. TVIMRT delineated in a transversal CT image. The field borders are also delineated.
inverse planning system, a volume is needed to specify the planning objectives. In this study, we used the volume that is conventionally irradiated, i.e. the tissue within the conventional tangential fields excluding the thorax wall and with a margin towards the skin and field borders. In order to minimize the delineation time, this volume was generated automatically. First, the skin surface and the lung were delineated using the auto-contouring tool available in our treatment planning system while the heart was delineated manually. Secondly, using the conventional beam settings obtained from virtual simulation, the volume within the treatment fields, the treated volume (TV), was automatically constructed and contracted with a margin of 7 mm to exclude the build-up region. Next, the heart and lung were subtracted from the TV, after an expansion of 10 and 5 mm, respectively, resulting in the treated volume used to optimize the dose distribution (TVIMRT; Fig. 1). The lung and heart were expanded to obtain a volume including the adjacent thorax wall. After automatic generation, the TVIMRT was visually checked to ensure that the result was clinically acceptable.
Treatment planning Conventional plans Treatment planning was performed using an 8 MV photon beam of an Elekta accelerator. The treatment was virtually simulated using the CT-data to establish the collimator angle, gantry angle and field borders. The field borders of the rectangular beams were designed to encompass the whole breast. A dose of 50 Gy (2 Gy/per fraction; five times weekly) was prescribed to the breast tissue. The dose distribution for the conventional wedged beam plans was optimized in a transversal slice in the center of the field by varying the wedge angle. These dose calculations were performed using a convolution–superposition algorithm.
164
Direct aperture optimization for breast IMRT
After calculation, the dose was normalized to the mean dose of the TVIMRT.
IMRT techniques For the IMRT plans, the beam parameters (gantry angle, collimator angle, energy) were identical to those of the wedged beam plans. The IMRT plans were calculated using the inverse planning module of the Pinnacle treatment planning system. A quadratic objective cost function was used to optimize the dose distribution. A minimum (47.5 Gy), maximum (52.5 Gy) and uniform (50 Gy) dose objective were assigned to the TVIMRT. For each patient, the same objective and weights were used for TVIMRT. For the heart and the lung, a maximum dose objective was set at 47.5 Gy, setting the weights individually for each patient. All IMRT plans were optimized during 25 iterations and a final dose calculation was performed using the convolution– superposition algorithm. The DAO method implemented in our planning system is called direct machine parameter optimization (DMPO; Fig. 2) developed by RaySearch Laboratories (Stockholm, Sweden). Using DMPO, the segments are created after initial optimization of the fluence map using a pencil beam model during the first iterations. The number of initial iterations was set to 10. In order to obtain the segments, the optimized fluence map is first divided into equidistant intensity levels. Next, the dose distribution is calculated with the convolution–superposition algorithm for all beams including segments. In the remaining iterations segment shape and weights are optimized, resulting in deliverable and optimized segments after optimization. The maximum number of segments is a soft constraint of the optimization process, i.e. it can be violated. Two IMRT plans were generated using DMPO. The first plan was generated using DMPO as described above. In order to obtain acceptable dose distributions, the maximum total number of segments was set to 18 for all patients (DMPO18). The second DMPO plan combined two predefined segments with DMPO (DMPOsegm). These predefined segments were
identical to the fields used for the conventional wedged plan. During optimization, the shape of these two segments was fixed, while the weight was optimized simultaneously with DMPO optimization for the other segments. The maximum number of segments for DMPO was set to eight. The initial weight of the two predefined segments was 80% relative to number of monitor units (MU) and for the DMPO segments 20%, to ensure a relatively large number of MUs for the two predefined segments after optimization. The two DMPO plans were compared with two conventional two-step (TS) IMRT approaches (Fig. 2). For the first plan (TScs), the optimization started using a simple pencil beam algorithm to get close to the optimal solution, and after 10 iterations a convolution–superposition algorithm was used to calculate the dose for all beams. For subsequent iterations, the pencil beam algorithm was used to calculate the difference in dose due to changes in intensity. These dose differences were than used to recalculate the dose at each iteration. In order to determine the effect of a more accurate dose calculation algorithm during the optimization process on the final dose distribution, a second plan (TSpb) was generated using the same approach but only using a pencil beam algorithm during optimization, thus excluding the convolution–superposition calculation after 10 iterations. The resulting mean dose of the TVIMRT was on average 51.7 Gy for the TSpb plans. The dose distribution was therefore normalized to 50 Gy for comparison with the other techniques. In the two-step approach, the k-means clustering algorithm [24] was used for the leaf-sequencing step. This algorithm was applied to minimize the difference between the intensity profile resulting from fluence optimization and the discrete intensity levels needed for conversion to segments, without violating the user specific deviation tolerance (set to 2%). The maximum number of intensity levels was set to five. Finally, besides the two DMPO plans and the two TS plans an extra IMRT plan was generated using the same approach as the TScs plan, but without the leaf-sequencing step after optimizing the intensity. This full fluence (FF) optimization was considered as the ‘optimal’ treatment for the patient and was used for comparison.
Analysis
Fig. 2. Flow chart of the classical two-step approach and direct machine parameter optimization (DMPO).
Quantative data were extracted from dose–volume histograms. In order to analyze the under- and overdosed areas in the TVIMRT for the different techniques, the volume receiving more than 95% of the prescribed dose (V95) and the volume receiving more than 105% of the prescribed dose (V105) were determined. The mean dose was also calculated for TVIMRT. Instead of the maximum and minimum dose, the dose received by 99% of the volume (D99), and the dose received by 1% (D01), were calculated. The difference between D99 and D01 (D99–01) was used to quantify the homogeneity of the dose distribution within the TVIMRT. D99–01 of each technique was compared with D99–01 of the FF calculation. This will show how close the dose homogeneity of an individual technique is to the optimal solution and thus, how much is lost by generating a step-and-shoot technique. The relative gain (RG) of each technique was calculated by:
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
165
Fig. 3. Example of a differential dose volume histogram of TVIMRT calculated for the DMPOsegm, DMPO18, TSpb, TScs and the wedged beam plans.
RGðimrtÞ Z
D99K01 ðwedgeÞKD99K01 ðimrtÞ D99K01 ðwedgeÞKD99K01 ðFFÞ
In order to quantify the effect on heart toxicity [7], the maximum dose (Dmax) was determined for all plans. Dmax is defined as the dose delivered to 1 cm3. For the lung, the mean dose of both lungs was calculated to quantify the risk of lung toxicity [4,17]. For step-and-shoot IMRT techniques, the number of MUs and segments are important factors and indicate the efficiency of a treatment. These parameters were therefore determined for all techniques. Paired two-sided student t-tests were performed to test if differences in calculated parameters were statistically significant. In the remainder of this paper, differences were considered to be statistically significant when P!0.05.
TS plans. The relative gain was therefore patient specific for both TS plans, resulting in a relatively large SD (Table 2). Compared with the two-step IMRT plans, the use of DMPO could significantly improve the dose homogeneity within the TVIMRT (Fig. 3). A remarkable difference compared with the TS plans was that dose homogeneity was close to the dose homogeneity of the optimal plans for all patients (Fig. 4(b)). This was not only reflected in an improved relative gain of the DMPO plan, but also in a much smaller SD (Table 2). When using DMPO, the gain was thus similar for all patients. Using DMPO in combination with predefined segments could even further improve the homogeneity of the dose distribution (Table 1), as could also be observed from the dose–volume histograms (Fig. 3) and the relative gain (Table 2). The DMPOsegm plan showed the least degradation of the dose distribution compared to the FF plan.
Organs at risk
Results Target volume Dose homogeneity within TVIMRT improved for all IMRT techniques compared to the use of wedged beams for the irradiation of the breast (Fig. 3, Table 1). The homogeneity (D99–01) did vary between patients, especially for the wedged beam plans, with a range of 5.6–8.5 Gy. The high-dose areas were somewhat decreased, but especially the low-dose areas were improved using IMRT (Table 2). IMRT dose distributions obtained with a simple pencil beam model during the optimization did improve the homogeneity of the dose distribution with respect to the wedged plan, but they were on average worse than those obtained with the TScs and DMPO techniques (Fig. 3, Table 1). Fig. 4 shows that the dose homogeneity for the TScs plans was in most cases better than for the TSpb plan, resulting in an improved relative gain for the TScs plan. The points in Fig. 4(a) are, however, relatively scattered for both
The average of the mean lung dose for all patients and all techniques was 3.4 Gy, including the lung dose for the wedged beam plan. Differences in mean dose between the techniques were within 0.5 Gy. Even for the wedged beam
Table 1 Average, minimum (min) and maximum (max) value of the dose homogeneity (D99–01) of the TVIMRT for the various plans of the 12 patients Dose homogeneity (D99–01) Wedge TSpb TScs DMPO18 DMPOsegm FF
Average (Gy)
Min (Gy)
Max (Gy)
6.4 5.2 4.8 4.3 3.8 3.4
5.6 4.6 3.8 3.4 3.0 2.2
8.5 6.7 5.8 5.4 5.3 4.8
166
Direct aperture optimization for breast IMRT
Table 2 Average value and standard deviation (SD) of V95 (%) and V105 (%) of the TVIMRT for the various plans of the 12 patients and the average value and SD of the relative gain of each IMRT technique V95 (%) Wedge TSpb TScs DMPO18 DMPOsegm FF
V105 (%)
Relative gain
Average
SD
Average
SD
Average
SD
94.4 98.2 98.8 99.1 99.4 99.6
3.5 1.1 0.5 0.4 0.4 0.3
2.4 0.9 0.5 0.3 0.1 0.1
2.7 0.8 0.6 0.4 0.3 0.3
– 0.38 0.51 0.69 0.85 1
– 0.18 0.18 0.09 0.08 –
plans, the high dose areas were already pushed towards the thorax wall. However, due to the differences in patient anatomy, the difference in mean lung dose between the patients was large. Small differences were found in the high dose areas, which occur in the parts of the lung included in the treatment fields. The dose for the wedged beam plan was for some patients slightly higher compared with the IMRT plans in these parts of the lung. On average, however, the differences are small; the volume of the lung receiving
more than 47.5 Gy was on average 2.6% for the wedged beam plans compared with approximately 1.5% for the IMRT plans. The maximum heart dose was on average 40.6 Gy for all patients and techniques. The differences between the techniques were within 1 Gy. Due to the patient and beam geometry, some dose delivery to the heart was unavoidable without decreasing the dose to the TVIMRT.
Delivery The number of segments was on average smaller than 10 (Table 3), except for the DMPO18 plan. Although DMPO could result in a smaller number of segments by setting the maximum number of segments to a smaller value, the homogeneity of the dose distribution would decrease. For instance, when decreasing the maximum number of segments from 18 to 10, the average value of D99–01 increased from 4.3 Gy (range 3.4–5.4 Gy) to 5.4 Gy (range 4.5–6.5 Gy). Increasing the number to higher values (O18) would hardly have any benefit. For example, when the maximum number of segments was set to 25, the average value of D99–01 was 4.2 Gy (range 3.7–5.5 Gy). Since, for the DMPO plans the maximum number of segments is a constraint, the resulting number of segments is similar for all patients. The resulting number of segments for the TS plans was varying more compared with the DMPO plans but resulted on average in slightly less segments (Table 3). The two-step IMRT techniques resulted in general in a large conformal segment, delivering 80–90% of the dose and covering the TVIMRT completely (Fig. 5). The remaining
Table 3 Average and standard deviation (SD) of the total number of segments and the relative number of monitor units, normalized to the wedged beam plan
Fig. 4. Dose homogeneity of the ‘optimal’ FF dose distribution plotted against the dose homogeneity of the wedged, TSpb and TScs plans (a) and against the DMPOsegm, DMPO18 plans (b). The xZy line is plotted to indicate a plan associated to an ideal segmentation that does not deteriorate the optimized dose distribution.
Wedge TSpb TScs DMPO18 DMPOsegm
Number of segments
Relative MUs
Average
SD
Average
SD
– 8.8 8.3 17.5 9.7
– 1.8 1.7 0.9 0.5
1.00 0.67 0.68 0.74 0.66
0.17 0.05 0.06 0.05 0.04
For the wedged beam plans the number of MUs was on average 380.
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
167
Fig. 5. Example of the segment shapes of a single beam obtained with the DMPOsegm technique.
segments were usually smaller and were used to modulate the intensity. For the DMPO18 technique, this was not the case and the bulk of the dose was often delivered by a few large segments with similar number of MUs, but different shapes. This might be a problem concerning intrafraction motion. The weight of the predefined segments for the DMPOsegm plans was initially set to 80% relative to the number of MUs. After optimization the weight was changed due to the optimization process itself, however, the dose delivered by these two predefined segments was approximately 80%. The leaf-sequencing procedure was efficient for all IMRT techniques considering the number of MUs. The resulting number of MUs was significantly lower for all IMRT plans compared with the number of MUs needed for the wedged beam plan (Table 3), which was on average 380.
Discussion In this paper, direct aperture optimization was used to generate step-and-shoot IMRT plans for the irradiation of breast cancer. IMRT plans were obtained using DMPO alone as well as in combination with predefined segments. Both approaches were compared with two traditional two-step IMRT approaches, using only a pencil beam algorithm or a pencil beam algorithm in combination with a convolution– superposition algorithm during the optimization. In general, the dose homogeneity for the DMPO plans was close to the optimal solution for all patients, i.e. the degradation of the dose distribution relative to the optimal solution was small and similar for each patient. This is different with respect to the two-step IMRT plans, which showed some degree of patient dependent variation in dose homogeneity. The most homogeneous dose distributions were obtained using DMPOsegm. These results are in agreement with results reported for several DAO techniques tested for prostate and head-and-neck cases. Using DAO, it was shown that the dose distribution was similar or better compared with a traditional two-step IMRT [2,3,12,18,19]. Only small differences in mean lung dose and maximum heart dose were observed between the IMRT techniques and the wedged beam technique. Even for the conventional technique, the high dose areas were already pushed towards the thorax wall, resulting in most cases in a satisfactory dose distribution in the lung. The dose to the lung and heart could probably reduced by changing the parameters in the cost function. However, the dose in the medial and lateral area of
the target volume next to the lung will than decrease as well, since only two opposed beams are used. This might be clinically relevant for some patients. The number of MUs used for the two-step IMRT and the DMPO plans were similar for all patients, showing approximately a 30% decrease relative to the wedged beam plan. Consequently, all IMRT techniques resulted in highly efficient treatment delivery in terms of number of MUs. For a breast IMRT plan, Mayo et al. [13] reported that the ratio of the number of MUs for IMRT relative to the wedged beam plans was 2.3. This ratio could be reduced to 1.1, by using a hybrid IMRT plan, which was optimized on top of an existing dose distribution produced by two tangential open beams. For the simplified IMRT technique developed by Chui et al. [1], the number of MUs was similar to the wedged beam plans. Using DAO for other treatments sites, such as head-and-neck cases, a reduction of 70–90% in number of MUs [18,19] was reported relative to a two-step IMRT approach. Using the DAO technique implemented in our planning system, DMPO, for the breast such a large reduction could not be achieved. This is probably due to the fact that IMRT plans for head-and-neck cased are more complex compared with the breast. Although other DAO techniques reported the use of less segments relative to a normal two-step IMRT technique [2,3,12,18,19], this was not the case for the DMPO18 technique. For this technique, more segments were needed compared with the other techniques in order to achieve a clinical acceptable dose distribution. This might be the result of the leaf-sequencing process during the optimization process. Using DMPO the intensity is first optimized to get close to an optimal solution. Before segmentation, the intensity profile is divided into equidistant levels and the segments are created. Since, the dose inhomogeneity for two tangential fields without modulation is only 10–20%, the bulk of the dose can be delivered using a large conformal segment. By using equidistant levels, this bulk is divided into several steps resulting in segments with similar shape and weight. Although the weight of the segments was optimized, the system was not able to produce two segments delivering the bulk of the dose when only using DMPO. By applying more segments, a better sampling of the intensity profile was performed, not only including the bulk part but also the more modulated part of the intensity profile. Using DMPO in combination with two open fields with a high weight at the beginning of the optimization gives a better starting position. In this case, the DMPO part is not used to deliver the bulk of the dose but to modulate the intensity and to
168
Direct aperture optimization for breast IMRT
generate some subfields. For breast IMRT, a more advanced conversion process could probably reduce the number of segments for DMPO. Step-and-shoot IMRT techniques developed for breast cancer treatment usually deliver 70–90% of the dose by a large open field [5,10,11,20]. A set of smaller segments is then added to improve the dose homogeneity. The total number of segments used for the treatment is generally between 5 and 10, which was also achieved using DMPOsegm and both TS techniques. The IMRT plans obtained with only a pencil beam algorithm resulted in an improved dose distribution compared with the wedged beam plans. However, the dose distribution had to be normalized after final calculation with the convolution–superposition algorithm, since these TSpb plans resulted on average in a mean dose of 51.7 Gy instead of 50 Gy. By using the convolution– superposition algorithm during optimization, the accuracy of the dose calculation was maintained and resulted in an improved dose distribution compared with optimizing only with a pencil beam algorithm. The use of a more accurate dose calculation during the optimization of the dose distribution for a complex geometry such as the breast is thus advisable. The use of DMPOsegm has several advantages besides the improved dose distribution and the efficiency of the delivery. Due to the use of two segments identical to the conventional tangential fields, the conventional field set-up procedure can be maintained and since, these segments deliver a large amount of dose to the breast the resulting irradiation technique is probably less sensitive to intrafraction motion than DMPO18 [6]. Instead of delineation of the clinical target volume, we decided to generate the IMRT plans using the conventional target volume. Delineation studies are therefore not necessary for the clinical implementation of IMRT for breast cancer and the generation of the TVIMRT is a relative fast procedure. Due to these findings, the DMPOsegm technique can be implemented relatively easily following existing clinical procedures.
Conclusion Using direct aperture optimization, the degradation of the quality of the dose distribution relative to the optimal solution was small and similar for each patient compared with a classical two-step IMRT approach. The combination of DMPO with two segments identical to those used in conventional 3D-CRT, resulted in a simple and efficient treatment technique for the irradiation of the breast that achieves a highly homogeneous dose distribution throughout the target volume. The DMPOsegm technique described in this paper is currently being implemented in our clinic.
Acknowledgements This project was financially supported by The Dutch Cancer Society (KNB Grant NKI 2000–2212).
*
`ne M.F. Damen, Department of Corresponding author. Euge Radiation Oncology, The Netherlands Cancer Institute, Antoni van
Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands. E-mail address: [email protected] Received 7 September 2005; received in revised form 18 April 2006; accepted 24 April 2006
References [1] Chui CS, Hong L, Hunt M, McCormick B. A simplified intensity modulated radiation therapy technique for the breast. Med Phys 2002;29:522–9. [2] Cotrutz C, Xing L. Segment-based dose optimization using a genetic algorithm. Phys Med Biol 2003;48:2987–98. [3] De Gersem W, Claus F, De Wagter C, De Neve W. An anatomybased beam segmentation tool for intensity-modulated radiation therapy and its application to head-and-neck cancer. Int J Radiat Oncol Biol Phys 2001;51:849–59. [4] De Jaeger K, Hoogeman MS, Engelsman M, et al. Incorporating an improved dose-calculation algorithm in conformal radiotherapy of lung cancer: re-evaluation of dose in normal lung tissue. Radiother Oncol 2003;69:1–10. [5] Evans PM, Donovan EM, Partridge M, et al. The delivery of intensity modulated radiotherapy to the breast using multiple static fields. Radiother Oncol 2000;57:79–89. [6] Frazier RC, Vicini FA, Sharpe MB, et al. Impact of breathing motion on whole breast radiotherapy: a dosimetric analysis using active breathing control. Int J Radiat Oncol Biol Phys 2004;58:1041–7. [7] Gagliardi G, Lax I, Soderstrom S, Gyenes G, Rutqvist LE. Prediction of excess risk of long-term cardiac mortality after radiotherapy of stage I breast cancer. Radiother Oncol 1998;46: 63–71. [8] Hurkmans CW, Borger JH, Pieters BR, Russell NS, Jansen EP, Mijnheer BJ. Variability in target volume delineation on CT scans of the breast. Int J Radiat Oncol Biol Phys 2001;50: 1366–72. [9] Hurkmans CW, Cho BC, Damen E, Zijp L, Mijnheer BJ. Reduction of cardiac and lung complication probabilities after breast irradiation using conformal radiotherapy with or without intensity modulation. Radiother Oncol 2002;62:163–71. [10] Kestin LL, Sharpe MB, Frazier RC, et al. Intensity modulation to improve dose uniformity with tangential breast radiotherapy: initial clinical experience. Int J Radiat Oncol Biol Phys 2000;48: 1559–68. [11] Li JS, Freedman GM, Price R, et al. Clinical implementation of intensity-modulated tangential beam irradiation for breast cancer. Med Phys 2004;31:1023–31. [12] Li Y, Yao J, Yao D. Genetic algorithm based deliverable segments optimization for static intensity-modulated radiotherapy. Phys Med Biol 2003;48:3353–74. [13] Mayo CS, Urie MM, Fitzgerald TJ. Hybrid IMRT plans— concurrently treating conventional and IMRT beams for improved breast irradiation and reduced planning time. Int J Radiat Oncol Biol Phys 2005;61:922–32. [14] Moody AM, Mayles WP, Bliss JM, et al. The influence of breast size on late radiation effects and association with radiotherapy dose inhomogeneity. Radiother Oncol 1994;33:106–12. [15] Neal AJ, Ton M, Helyer S, Yarnold JR. Correlation of breast dose heterogeneity with breast size using 3D CT planning and dose–volume histograms. Radiother Oncol 1995;34: 210–8. [16] Poggi MM, Danforth DN, Sciuto LC, et al. Eighteen-year results in the treatment of early breast carcinoma with mastectomy versus breast conservation therapy: the National Cancer Institute Randomized Trial. Cancer 2003;98:697–702.
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169 [17] Seppenwoolde Y, Lebesque JV, de Jaeger K, et al. Comparing different NTCP models that predict the incidence of radiation pneumonitis. Normal tissue complication probability. Int J Radiat Oncol Biol Phys 2003;55: 724–35. [18] Shepard DM, Earl MA, Li XA, Naqvi S, Yu C. Direct aperture optimization: a turnkey solution for step-and-shoot IMRT. Med Phys 2002;29:1007–18. [19] Siebers JV, Lauterbach M, Keall PJ, Mohan R. Incorporating multi-leaf collimator leaf sequencing into iterative IMRT optimization. Med Phys 2002;29:952–9. [20] van Asselen B, Raaijmakers CPJ, Hofman P, Lagendijk JJW. An improved breast irradiation technique using three-dimensional geometrical information and intensity modulation. Radiother Oncol 2001;58:341–7. [21] van Dongen JA, Voogd AC, Fentiman IS, et al. Long-term results of a randomized trial comparing breast-conserving therapy with
[22]
[23]
[24]
[25]
169
mastectomy: European Organization for Research and Treatment of Cancer 10801 trial. J Natl Cancer Inst 2000;92:1143–50. Vicini FA, Sharpe M, Kestin L, et al. Optimizing breast cancer treatment efficacy with intensity-modulated radiotherapy. Int J Radiat Oncol Biol Phys 2002;54:1336–44. Vrieling C, Collette L, Fourquet A, et al. The influence of patient, tumor and treatment factors on the cosmetic results after breast-conserving therapy in the EORTC ‘boost vs. no boost’ trial. EORTC Radiotherapy and Breast Cancer Cooperative Groups. Radiother Oncol 2000;55:219–32. Wu Y, Yan D, Sharpe MB, Miller B, Wong JW. Implementing multiple static field delivery for intensity modulated beams. Med Phys 2001;28:2188–97. Yarnold J, Donovan E, Bleackly N, et al. Randomised trial of standard 2D radiotherapy (RT) versus 3D intensity modulated radiotherapy (IMRT) in patients prescribed breast radiotherapy. Radiother Oncol 2005;64:S15.
Breast IMRT
Intensity-modulated radiotherapy of breast cancer using direct aperture optimization Bram van Asselen, Marco Schwarz, Corine van Vliet-Vroegindeweij, Joos V. Lebesque, Ben J. Mijnheer, Eugene M.F. Damen* Department of Radiation Oncology, The Netherlands Cancer Institute, Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands
Abstract Background and purpose: To design a clinically reliable and efficient step-and-shoot IMRT delivery technique for the treatment of breast cancer using direct aperture optimization (DAO). Using DAO, segments are created and optimized within the same optimization process. Patients and methods: The DAO technique implemented in the Pinnacle treatment planning system, which is called direct machine parameter optimization (DMPO), was used to generate IMRT plans for twelve breast cancer patients. The prescribed dose was 50 Gy. Two DMPO plans were generated. The first approach uses DMPO only; the second technique combines DMPO with two predefined segments (DMPOsegm), having shapes identical to the conventional tangential fields. The weight of these predefined segments is optimized simultaneously with DMPO. The DMPO plans were compared with normal two-step (TS) IMRT, creating segments after optimizing the intensity. Results: Dose homogeneity within the target volume was 4.8G0.6, 4.3G0.5 and 3.8G0.5 Gy for the TS, DMPO and DMPOsegm plans, respectively. Comparing the IMRT plans with an idealized dose distribution obtained using only beamlet optimization, the degradation of the dose distribution was less for the DMPO plans compared with the two-step IMRT approach. Furthermore, this degradation was similar for all patients, while for the two-step IMRT approach it was patient specific. Conclusions: An efficient step-and-shoot IMRT solution was developed for the treatment of breast cancer using DMPO combined with two predefined segments. q 2006 Elsevier Ireland Ltd. All rights reserved. Radiotherapy and Oncology 79 (2006) 162–169. Keywords: IMRT; Breast; Direct aperture optimisation; Step-and-shoot
Radiotherapy is an effective treatment modality for the treatment of breast cancer [16,21] and is a widely accepted treatment modality for early-stage breast cancer. Patients are usually treated using two tangential wedged beams. The use of wedged fields can result in a heterogeneous distribution, particularly in cranial and caudal parts of the breast, where low and high dose areas can occur. An overdosage may result in worse cosmetic results after irradiation [14,15,23] and underdosage may result in a lower tumor control probability. Using intensity modulated radiation therapy (IMRT), the homogeneity of the dose distribution in the breast can be improved compared with the use of wedged beams [1,5,9–11,13,20]. Promising data were published on cosmetic results of patients treated with IMRT [22] and a better breast appearance using IMRT was reported in another study [25]. Although it has been shown that IMRT can improve the dose distribution, a number of institutes still hesitate to introduce IMRT for breast cancer treatment for several
reasons. First, a planning CT scan has to be performed of each patient, and a target volume has to be delineated. This may increase the workload, since breast patients are a relatively large patient group in many departments. Furthermore, individual delineation may result in uncertainties in target volume definition [8]. Finally, in most treatment planning systems relatively simple dose calculation algorithms are used for inverse planning, while after conversion the dose is usually recalculated using a more accurate dose calculation. Due to different modeling of the tissue inhomogeneity of the lung, missing tissue and oblique incidence, the quality of the plan may degrade after recalculation. The conversion of the intensity profile to deliverable segments may also cause a degradation of the plan, since the intensity profile delivered by the segments is not exactly the same as the optimized intensity map. This can be the result of a limited number of intensity levels used for sampling the intensity profile, but also due to the fact that leakage, head scatter and tongue-and-groove effect are
0167-8140/$ - see front matter q 2006 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.radonc.2006.04.010
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
not taken into account during optimization. Furthermore, the leaf-sequencing process should be efficient, since many segments and monitor units (MUs) might not be clinically acceptable. Too many MUs introduce a larger scattered dose component, resulting in an increase in dose to surrounding tissues such as the contralateral breast. Institutes treating breast cancer with IMRT developed their own IMRT methods to solve one or more of the abovementioned issues [1,5,10,11,13,20]. For step-and-shoot IMRT techniques, the segments shapes are usually created using isodose surfaces, anatomy or radiological thickness. The weight of these segments is then optimized to achieve a homogeneous dose distribution. Segments are thus created before the optimization starts and are not part of the optimization itself. In this article, we will focus on step-and-shoot IMRT techniques for the irradiation of the breast. The main focus of this paper was on the use of direct aperture optimization (DAO). DAO is a method for step-and-shoot IMRT and has shown to be an efficient method for IMRT delivery [2,3,12,18,19]. Using DAO, the leaf positions and the weight of each segment are parameters of the optimization, therefore eliminating a separate leaf sequencing step after fluence optimization. The segment shapes and weights are thus part of the optimization cycle. It was the purpose of this study to design a simple, efficient and clinically acceptable IMRT technique for irradiation of the breast using DAO. The DAO plans will be tested separately as well as in combination with predefined segments. The results will be compared with the more classical two-step approach of IMRT, where segments are created after the optimization process. Since, the dose calculation algorithm used during optimization may be an important factor for IMRT of the breast, the effect of using a convolution–superposition algorithm during the inverse planning process was also tested relative to the use of a simple pencil model for the classical two-step approach.
Methods and materials Patients Twelve patients, previously treated for left-sided breast cancer, were selected for this study. All patients underwent a computer tomography (CT) scan, with a 5 mm slice interval. The images were obtained with the patient positioned supine with their arms placed above the heads in an arm-rest, equivalent to the treatment position. The scan was transferred to the Pinnacle treatment planning system (Philips Medical Systems, Best, the Netherlands; version 7.4). Before CT-scanning the palpated breast tissue was outlined using a lead wire. The CT scan included the heart and both lungs.
Volume definition For large-scale implementation of an IMRT technique, the planning time should not significantly increase compared to conventional treatment planning. Delineation of the breast tissue for each individual patient by a physician may therefore not be desirable. Furthermore, the interobserver variation has shown to be large [8]. However, using an
163
Fig. 1. TVIMRT delineated in a transversal CT image. The field borders are also delineated.
inverse planning system, a volume is needed to specify the planning objectives. In this study, we used the volume that is conventionally irradiated, i.e. the tissue within the conventional tangential fields excluding the thorax wall and with a margin towards the skin and field borders. In order to minimize the delineation time, this volume was generated automatically. First, the skin surface and the lung were delineated using the auto-contouring tool available in our treatment planning system while the heart was delineated manually. Secondly, using the conventional beam settings obtained from virtual simulation, the volume within the treatment fields, the treated volume (TV), was automatically constructed and contracted with a margin of 7 mm to exclude the build-up region. Next, the heart and lung were subtracted from the TV, after an expansion of 10 and 5 mm, respectively, resulting in the treated volume used to optimize the dose distribution (TVIMRT; Fig. 1). The lung and heart were expanded to obtain a volume including the adjacent thorax wall. After automatic generation, the TVIMRT was visually checked to ensure that the result was clinically acceptable.
Treatment planning Conventional plans Treatment planning was performed using an 8 MV photon beam of an Elekta accelerator. The treatment was virtually simulated using the CT-data to establish the collimator angle, gantry angle and field borders. The field borders of the rectangular beams were designed to encompass the whole breast. A dose of 50 Gy (2 Gy/per fraction; five times weekly) was prescribed to the breast tissue. The dose distribution for the conventional wedged beam plans was optimized in a transversal slice in the center of the field by varying the wedge angle. These dose calculations were performed using a convolution–superposition algorithm.
164
Direct aperture optimization for breast IMRT
After calculation, the dose was normalized to the mean dose of the TVIMRT.
IMRT techniques For the IMRT plans, the beam parameters (gantry angle, collimator angle, energy) were identical to those of the wedged beam plans. The IMRT plans were calculated using the inverse planning module of the Pinnacle treatment planning system. A quadratic objective cost function was used to optimize the dose distribution. A minimum (47.5 Gy), maximum (52.5 Gy) and uniform (50 Gy) dose objective were assigned to the TVIMRT. For each patient, the same objective and weights were used for TVIMRT. For the heart and the lung, a maximum dose objective was set at 47.5 Gy, setting the weights individually for each patient. All IMRT plans were optimized during 25 iterations and a final dose calculation was performed using the convolution– superposition algorithm. The DAO method implemented in our planning system is called direct machine parameter optimization (DMPO; Fig. 2) developed by RaySearch Laboratories (Stockholm, Sweden). Using DMPO, the segments are created after initial optimization of the fluence map using a pencil beam model during the first iterations. The number of initial iterations was set to 10. In order to obtain the segments, the optimized fluence map is first divided into equidistant intensity levels. Next, the dose distribution is calculated with the convolution–superposition algorithm for all beams including segments. In the remaining iterations segment shape and weights are optimized, resulting in deliverable and optimized segments after optimization. The maximum number of segments is a soft constraint of the optimization process, i.e. it can be violated. Two IMRT plans were generated using DMPO. The first plan was generated using DMPO as described above. In order to obtain acceptable dose distributions, the maximum total number of segments was set to 18 for all patients (DMPO18). The second DMPO plan combined two predefined segments with DMPO (DMPOsegm). These predefined segments were
identical to the fields used for the conventional wedged plan. During optimization, the shape of these two segments was fixed, while the weight was optimized simultaneously with DMPO optimization for the other segments. The maximum number of segments for DMPO was set to eight. The initial weight of the two predefined segments was 80% relative to number of monitor units (MU) and for the DMPO segments 20%, to ensure a relatively large number of MUs for the two predefined segments after optimization. The two DMPO plans were compared with two conventional two-step (TS) IMRT approaches (Fig. 2). For the first plan (TScs), the optimization started using a simple pencil beam algorithm to get close to the optimal solution, and after 10 iterations a convolution–superposition algorithm was used to calculate the dose for all beams. For subsequent iterations, the pencil beam algorithm was used to calculate the difference in dose due to changes in intensity. These dose differences were than used to recalculate the dose at each iteration. In order to determine the effect of a more accurate dose calculation algorithm during the optimization process on the final dose distribution, a second plan (TSpb) was generated using the same approach but only using a pencil beam algorithm during optimization, thus excluding the convolution–superposition calculation after 10 iterations. The resulting mean dose of the TVIMRT was on average 51.7 Gy for the TSpb plans. The dose distribution was therefore normalized to 50 Gy for comparison with the other techniques. In the two-step approach, the k-means clustering algorithm [24] was used for the leaf-sequencing step. This algorithm was applied to minimize the difference between the intensity profile resulting from fluence optimization and the discrete intensity levels needed for conversion to segments, without violating the user specific deviation tolerance (set to 2%). The maximum number of intensity levels was set to five. Finally, besides the two DMPO plans and the two TS plans an extra IMRT plan was generated using the same approach as the TScs plan, but without the leaf-sequencing step after optimizing the intensity. This full fluence (FF) optimization was considered as the ‘optimal’ treatment for the patient and was used for comparison.
Analysis
Fig. 2. Flow chart of the classical two-step approach and direct machine parameter optimization (DMPO).
Quantative data were extracted from dose–volume histograms. In order to analyze the under- and overdosed areas in the TVIMRT for the different techniques, the volume receiving more than 95% of the prescribed dose (V95) and the volume receiving more than 105% of the prescribed dose (V105) were determined. The mean dose was also calculated for TVIMRT. Instead of the maximum and minimum dose, the dose received by 99% of the volume (D99), and the dose received by 1% (D01), were calculated. The difference between D99 and D01 (D99–01) was used to quantify the homogeneity of the dose distribution within the TVIMRT. D99–01 of each technique was compared with D99–01 of the FF calculation. This will show how close the dose homogeneity of an individual technique is to the optimal solution and thus, how much is lost by generating a step-and-shoot technique. The relative gain (RG) of each technique was calculated by:
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
165
Fig. 3. Example of a differential dose volume histogram of TVIMRT calculated for the DMPOsegm, DMPO18, TSpb, TScs and the wedged beam plans.
RGðimrtÞ Z
D99K01 ðwedgeÞKD99K01 ðimrtÞ D99K01 ðwedgeÞKD99K01 ðFFÞ
In order to quantify the effect on heart toxicity [7], the maximum dose (Dmax) was determined for all plans. Dmax is defined as the dose delivered to 1 cm3. For the lung, the mean dose of both lungs was calculated to quantify the risk of lung toxicity [4,17]. For step-and-shoot IMRT techniques, the number of MUs and segments are important factors and indicate the efficiency of a treatment. These parameters were therefore determined for all techniques. Paired two-sided student t-tests were performed to test if differences in calculated parameters were statistically significant. In the remainder of this paper, differences were considered to be statistically significant when P!0.05.
TS plans. The relative gain was therefore patient specific for both TS plans, resulting in a relatively large SD (Table 2). Compared with the two-step IMRT plans, the use of DMPO could significantly improve the dose homogeneity within the TVIMRT (Fig. 3). A remarkable difference compared with the TS plans was that dose homogeneity was close to the dose homogeneity of the optimal plans for all patients (Fig. 4(b)). This was not only reflected in an improved relative gain of the DMPO plan, but also in a much smaller SD (Table 2). When using DMPO, the gain was thus similar for all patients. Using DMPO in combination with predefined segments could even further improve the homogeneity of the dose distribution (Table 1), as could also be observed from the dose–volume histograms (Fig. 3) and the relative gain (Table 2). The DMPOsegm plan showed the least degradation of the dose distribution compared to the FF plan.
Organs at risk
Results Target volume Dose homogeneity within TVIMRT improved for all IMRT techniques compared to the use of wedged beams for the irradiation of the breast (Fig. 3, Table 1). The homogeneity (D99–01) did vary between patients, especially for the wedged beam plans, with a range of 5.6–8.5 Gy. The high-dose areas were somewhat decreased, but especially the low-dose areas were improved using IMRT (Table 2). IMRT dose distributions obtained with a simple pencil beam model during the optimization did improve the homogeneity of the dose distribution with respect to the wedged plan, but they were on average worse than those obtained with the TScs and DMPO techniques (Fig. 3, Table 1). Fig. 4 shows that the dose homogeneity for the TScs plans was in most cases better than for the TSpb plan, resulting in an improved relative gain for the TScs plan. The points in Fig. 4(a) are, however, relatively scattered for both
The average of the mean lung dose for all patients and all techniques was 3.4 Gy, including the lung dose for the wedged beam plan. Differences in mean dose between the techniques were within 0.5 Gy. Even for the wedged beam
Table 1 Average, minimum (min) and maximum (max) value of the dose homogeneity (D99–01) of the TVIMRT for the various plans of the 12 patients Dose homogeneity (D99–01) Wedge TSpb TScs DMPO18 DMPOsegm FF
Average (Gy)
Min (Gy)
Max (Gy)
6.4 5.2 4.8 4.3 3.8 3.4
5.6 4.6 3.8 3.4 3.0 2.2
8.5 6.7 5.8 5.4 5.3 4.8
166
Direct aperture optimization for breast IMRT
Table 2 Average value and standard deviation (SD) of V95 (%) and V105 (%) of the TVIMRT for the various plans of the 12 patients and the average value and SD of the relative gain of each IMRT technique V95 (%) Wedge TSpb TScs DMPO18 DMPOsegm FF
V105 (%)
Relative gain
Average
SD
Average
SD
Average
SD
94.4 98.2 98.8 99.1 99.4 99.6
3.5 1.1 0.5 0.4 0.4 0.3
2.4 0.9 0.5 0.3 0.1 0.1
2.7 0.8 0.6 0.4 0.3 0.3
– 0.38 0.51 0.69 0.85 1
– 0.18 0.18 0.09 0.08 –
plans, the high dose areas were already pushed towards the thorax wall. However, due to the differences in patient anatomy, the difference in mean lung dose between the patients was large. Small differences were found in the high dose areas, which occur in the parts of the lung included in the treatment fields. The dose for the wedged beam plan was for some patients slightly higher compared with the IMRT plans in these parts of the lung. On average, however, the differences are small; the volume of the lung receiving
more than 47.5 Gy was on average 2.6% for the wedged beam plans compared with approximately 1.5% for the IMRT plans. The maximum heart dose was on average 40.6 Gy for all patients and techniques. The differences between the techniques were within 1 Gy. Due to the patient and beam geometry, some dose delivery to the heart was unavoidable without decreasing the dose to the TVIMRT.
Delivery The number of segments was on average smaller than 10 (Table 3), except for the DMPO18 plan. Although DMPO could result in a smaller number of segments by setting the maximum number of segments to a smaller value, the homogeneity of the dose distribution would decrease. For instance, when decreasing the maximum number of segments from 18 to 10, the average value of D99–01 increased from 4.3 Gy (range 3.4–5.4 Gy) to 5.4 Gy (range 4.5–6.5 Gy). Increasing the number to higher values (O18) would hardly have any benefit. For example, when the maximum number of segments was set to 25, the average value of D99–01 was 4.2 Gy (range 3.7–5.5 Gy). Since, for the DMPO plans the maximum number of segments is a constraint, the resulting number of segments is similar for all patients. The resulting number of segments for the TS plans was varying more compared with the DMPO plans but resulted on average in slightly less segments (Table 3). The two-step IMRT techniques resulted in general in a large conformal segment, delivering 80–90% of the dose and covering the TVIMRT completely (Fig. 5). The remaining
Table 3 Average and standard deviation (SD) of the total number of segments and the relative number of monitor units, normalized to the wedged beam plan
Fig. 4. Dose homogeneity of the ‘optimal’ FF dose distribution plotted against the dose homogeneity of the wedged, TSpb and TScs plans (a) and against the DMPOsegm, DMPO18 plans (b). The xZy line is plotted to indicate a plan associated to an ideal segmentation that does not deteriorate the optimized dose distribution.
Wedge TSpb TScs DMPO18 DMPOsegm
Number of segments
Relative MUs
Average
SD
Average
SD
– 8.8 8.3 17.5 9.7
– 1.8 1.7 0.9 0.5
1.00 0.67 0.68 0.74 0.66
0.17 0.05 0.06 0.05 0.04
For the wedged beam plans the number of MUs was on average 380.
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169
167
Fig. 5. Example of the segment shapes of a single beam obtained with the DMPOsegm technique.
segments were usually smaller and were used to modulate the intensity. For the DMPO18 technique, this was not the case and the bulk of the dose was often delivered by a few large segments with similar number of MUs, but different shapes. This might be a problem concerning intrafraction motion. The weight of the predefined segments for the DMPOsegm plans was initially set to 80% relative to the number of MUs. After optimization the weight was changed due to the optimization process itself, however, the dose delivered by these two predefined segments was approximately 80%. The leaf-sequencing procedure was efficient for all IMRT techniques considering the number of MUs. The resulting number of MUs was significantly lower for all IMRT plans compared with the number of MUs needed for the wedged beam plan (Table 3), which was on average 380.
Discussion In this paper, direct aperture optimization was used to generate step-and-shoot IMRT plans for the irradiation of breast cancer. IMRT plans were obtained using DMPO alone as well as in combination with predefined segments. Both approaches were compared with two traditional two-step IMRT approaches, using only a pencil beam algorithm or a pencil beam algorithm in combination with a convolution– superposition algorithm during the optimization. In general, the dose homogeneity for the DMPO plans was close to the optimal solution for all patients, i.e. the degradation of the dose distribution relative to the optimal solution was small and similar for each patient. This is different with respect to the two-step IMRT plans, which showed some degree of patient dependent variation in dose homogeneity. The most homogeneous dose distributions were obtained using DMPOsegm. These results are in agreement with results reported for several DAO techniques tested for prostate and head-and-neck cases. Using DAO, it was shown that the dose distribution was similar or better compared with a traditional two-step IMRT [2,3,12,18,19]. Only small differences in mean lung dose and maximum heart dose were observed between the IMRT techniques and the wedged beam technique. Even for the conventional technique, the high dose areas were already pushed towards the thorax wall, resulting in most cases in a satisfactory dose distribution in the lung. The dose to the lung and heart could probably reduced by changing the parameters in the cost function. However, the dose in the medial and lateral area of
the target volume next to the lung will than decrease as well, since only two opposed beams are used. This might be clinically relevant for some patients. The number of MUs used for the two-step IMRT and the DMPO plans were similar for all patients, showing approximately a 30% decrease relative to the wedged beam plan. Consequently, all IMRT techniques resulted in highly efficient treatment delivery in terms of number of MUs. For a breast IMRT plan, Mayo et al. [13] reported that the ratio of the number of MUs for IMRT relative to the wedged beam plans was 2.3. This ratio could be reduced to 1.1, by using a hybrid IMRT plan, which was optimized on top of an existing dose distribution produced by two tangential open beams. For the simplified IMRT technique developed by Chui et al. [1], the number of MUs was similar to the wedged beam plans. Using DAO for other treatments sites, such as head-and-neck cases, a reduction of 70–90% in number of MUs [18,19] was reported relative to a two-step IMRT approach. Using the DAO technique implemented in our planning system, DMPO, for the breast such a large reduction could not be achieved. This is probably due to the fact that IMRT plans for head-and-neck cased are more complex compared with the breast. Although other DAO techniques reported the use of less segments relative to a normal two-step IMRT technique [2,3,12,18,19], this was not the case for the DMPO18 technique. For this technique, more segments were needed compared with the other techniques in order to achieve a clinical acceptable dose distribution. This might be the result of the leaf-sequencing process during the optimization process. Using DMPO the intensity is first optimized to get close to an optimal solution. Before segmentation, the intensity profile is divided into equidistant levels and the segments are created. Since, the dose inhomogeneity for two tangential fields without modulation is only 10–20%, the bulk of the dose can be delivered using a large conformal segment. By using equidistant levels, this bulk is divided into several steps resulting in segments with similar shape and weight. Although the weight of the segments was optimized, the system was not able to produce two segments delivering the bulk of the dose when only using DMPO. By applying more segments, a better sampling of the intensity profile was performed, not only including the bulk part but also the more modulated part of the intensity profile. Using DMPO in combination with two open fields with a high weight at the beginning of the optimization gives a better starting position. In this case, the DMPO part is not used to deliver the bulk of the dose but to modulate the intensity and to
168
Direct aperture optimization for breast IMRT
generate some subfields. For breast IMRT, a more advanced conversion process could probably reduce the number of segments for DMPO. Step-and-shoot IMRT techniques developed for breast cancer treatment usually deliver 70–90% of the dose by a large open field [5,10,11,20]. A set of smaller segments is then added to improve the dose homogeneity. The total number of segments used for the treatment is generally between 5 and 10, which was also achieved using DMPOsegm and both TS techniques. The IMRT plans obtained with only a pencil beam algorithm resulted in an improved dose distribution compared with the wedged beam plans. However, the dose distribution had to be normalized after final calculation with the convolution–superposition algorithm, since these TSpb plans resulted on average in a mean dose of 51.7 Gy instead of 50 Gy. By using the convolution– superposition algorithm during optimization, the accuracy of the dose calculation was maintained and resulted in an improved dose distribution compared with optimizing only with a pencil beam algorithm. The use of a more accurate dose calculation during the optimization of the dose distribution for a complex geometry such as the breast is thus advisable. The use of DMPOsegm has several advantages besides the improved dose distribution and the efficiency of the delivery. Due to the use of two segments identical to the conventional tangential fields, the conventional field set-up procedure can be maintained and since, these segments deliver a large amount of dose to the breast the resulting irradiation technique is probably less sensitive to intrafraction motion than DMPO18 [6]. Instead of delineation of the clinical target volume, we decided to generate the IMRT plans using the conventional target volume. Delineation studies are therefore not necessary for the clinical implementation of IMRT for breast cancer and the generation of the TVIMRT is a relative fast procedure. Due to these findings, the DMPOsegm technique can be implemented relatively easily following existing clinical procedures.
Conclusion Using direct aperture optimization, the degradation of the quality of the dose distribution relative to the optimal solution was small and similar for each patient compared with a classical two-step IMRT approach. The combination of DMPO with two segments identical to those used in conventional 3D-CRT, resulted in a simple and efficient treatment technique for the irradiation of the breast that achieves a highly homogeneous dose distribution throughout the target volume. The DMPOsegm technique described in this paper is currently being implemented in our clinic.
Acknowledgements This project was financially supported by The Dutch Cancer Society (KNB Grant NKI 2000–2212).
*
`ne M.F. Damen, Department of Corresponding author. Euge Radiation Oncology, The Netherlands Cancer Institute, Antoni van
Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands. E-mail address: [email protected] Received 7 September 2005; received in revised form 18 April 2006; accepted 24 April 2006
References [1] Chui CS, Hong L, Hunt M, McCormick B. A simplified intensity modulated radiation therapy technique for the breast. Med Phys 2002;29:522–9. [2] Cotrutz C, Xing L. Segment-based dose optimization using a genetic algorithm. Phys Med Biol 2003;48:2987–98. [3] De Gersem W, Claus F, De Wagter C, De Neve W. An anatomybased beam segmentation tool for intensity-modulated radiation therapy and its application to head-and-neck cancer. Int J Radiat Oncol Biol Phys 2001;51:849–59. [4] De Jaeger K, Hoogeman MS, Engelsman M, et al. Incorporating an improved dose-calculation algorithm in conformal radiotherapy of lung cancer: re-evaluation of dose in normal lung tissue. Radiother Oncol 2003;69:1–10. [5] Evans PM, Donovan EM, Partridge M, et al. The delivery of intensity modulated radiotherapy to the breast using multiple static fields. Radiother Oncol 2000;57:79–89. [6] Frazier RC, Vicini FA, Sharpe MB, et al. Impact of breathing motion on whole breast radiotherapy: a dosimetric analysis using active breathing control. Int J Radiat Oncol Biol Phys 2004;58:1041–7. [7] Gagliardi G, Lax I, Soderstrom S, Gyenes G, Rutqvist LE. Prediction of excess risk of long-term cardiac mortality after radiotherapy of stage I breast cancer. Radiother Oncol 1998;46: 63–71. [8] Hurkmans CW, Borger JH, Pieters BR, Russell NS, Jansen EP, Mijnheer BJ. Variability in target volume delineation on CT scans of the breast. Int J Radiat Oncol Biol Phys 2001;50: 1366–72. [9] Hurkmans CW, Cho BC, Damen E, Zijp L, Mijnheer BJ. Reduction of cardiac and lung complication probabilities after breast irradiation using conformal radiotherapy with or without intensity modulation. Radiother Oncol 2002;62:163–71. [10] Kestin LL, Sharpe MB, Frazier RC, et al. Intensity modulation to improve dose uniformity with tangential breast radiotherapy: initial clinical experience. Int J Radiat Oncol Biol Phys 2000;48: 1559–68. [11] Li JS, Freedman GM, Price R, et al. Clinical implementation of intensity-modulated tangential beam irradiation for breast cancer. Med Phys 2004;31:1023–31. [12] Li Y, Yao J, Yao D. Genetic algorithm based deliverable segments optimization for static intensity-modulated radiotherapy. Phys Med Biol 2003;48:3353–74. [13] Mayo CS, Urie MM, Fitzgerald TJ. Hybrid IMRT plans— concurrently treating conventional and IMRT beams for improved breast irradiation and reduced planning time. Int J Radiat Oncol Biol Phys 2005;61:922–32. [14] Moody AM, Mayles WP, Bliss JM, et al. The influence of breast size on late radiation effects and association with radiotherapy dose inhomogeneity. Radiother Oncol 1994;33:106–12. [15] Neal AJ, Ton M, Helyer S, Yarnold JR. Correlation of breast dose heterogeneity with breast size using 3D CT planning and dose–volume histograms. Radiother Oncol 1995;34: 210–8. [16] Poggi MM, Danforth DN, Sciuto LC, et al. Eighteen-year results in the treatment of early breast carcinoma with mastectomy versus breast conservation therapy: the National Cancer Institute Randomized Trial. Cancer 2003;98:697–702.
B. van Asselen et al. / Radiotherapy and Oncology 79 (2006) 162–169 [17] Seppenwoolde Y, Lebesque JV, de Jaeger K, et al. Comparing different NTCP models that predict the incidence of radiation pneumonitis. Normal tissue complication probability. Int J Radiat Oncol Biol Phys 2003;55: 724–35. [18] Shepard DM, Earl MA, Li XA, Naqvi S, Yu C. Direct aperture optimization: a turnkey solution for step-and-shoot IMRT. Med Phys 2002;29:1007–18. [19] Siebers JV, Lauterbach M, Keall PJ, Mohan R. Incorporating multi-leaf collimator leaf sequencing into iterative IMRT optimization. Med Phys 2002;29:952–9. [20] van Asselen B, Raaijmakers CPJ, Hofman P, Lagendijk JJW. An improved breast irradiation technique using three-dimensional geometrical information and intensity modulation. Radiother Oncol 2001;58:341–7. [21] van Dongen JA, Voogd AC, Fentiman IS, et al. Long-term results of a randomized trial comparing breast-conserving therapy with
[22]
[23]
[24]
[25]
169
mastectomy: European Organization for Research and Treatment of Cancer 10801 trial. J Natl Cancer Inst 2000;92:1143–50. Vicini FA, Sharpe M, Kestin L, et al. Optimizing breast cancer treatment efficacy with intensity-modulated radiotherapy. Int J Radiat Oncol Biol Phys 2002;54:1336–44. Vrieling C, Collette L, Fourquet A, et al. The influence of patient, tumor and treatment factors on the cosmetic results after breast-conserving therapy in the EORTC ‘boost vs. no boost’ trial. EORTC Radiotherapy and Breast Cancer Cooperative Groups. Radiother Oncol 2000;55:219–32. Wu Y, Yan D, Sharpe MB, Miller B, Wong JW. Implementing multiple static field delivery for intensity modulated beams. Med Phys 2001;28:2188–97. Yarnold J, Donovan E, Bleackly N, et al. Randomised trial of standard 2D radiotherapy (RT) versus 3D intensity modulated radiotherapy (IMRT) in patients prescribed breast radiotherapy. Radiother Oncol 2005;64:S15.