Planning quality and delivery efficiency of sMLC delivered IMRT treatment of oropharyngeal cancers evaluated by RTOG H‐0022 dosimetric criteria

The time required to deliver intensity‐modulated radiation therapy (IMRT) treatments can be significantly longer than conventional treatments, especially for the segmented multileaf collimator (sMLC) delivery system with a large record and verification (R&V) overhead. In this work, we evaluate the impact of the number of intensity‐modulated beams (IMBs) and the number of intensity levels (ILs) on the quality and delivery efficiency of IMRT plans, generated by the Corvus planning system for sMLC delivery on a Siemens LINAC with the Lantis R&V system. Detailed studies were performed for three image data sets of previously treated oropharyngeal patients. Treatment plans for patient 1 were developed using 5, 7, 9, or 15 evenly spaced axial IMBs as well as one with 7 axial IMBs whose directions were user‐selected, each using ILs of 3, 5, 10, or 20. For patients 2 and 3, plans with 15 IMBs and 20 ILs were not attempted. A total of 42 plans were developed using three oropharyngeal cancer CT image data sets. Plan quality was evaluated by assessing compliance with the Radiation Therapy Oncology Group (RTOG) H‐0022 protocol criteria and the physician's clinical judgment. Plan efficiency was accessed by the number of segments of each plan. We found that for our treatment‐planning and delivery system, an IMRT plan that uses a moderate number of IMBs and ILs, such as 7 or 9 IMBs with 3 or 5 ILs, would appear to be the optimal approach when both quality of the plan and delivery efficiency are considered. Based on this study, we have routinely used 9 IMBs with 3 ILs or 7 IMBs with 5 ILs for head and neck patients. A retrospective comparison indicates that delivery efficiency is improved on the order of 30% compared to plans generated with 9 IMBs with 5 ILs. PACS number: 87.53.Tf


I. INTRODUCTION
In recent years, there has been great interest in implementing intensity-modulated radiation therapy (IMRT) in external beam radiation therapy. IMRT employs nonuniform beam intensity to deliver highly conformal radiation to the targets while minimizing doses to normal tissues and critical organs. (1,2) Head and neck cancer is one of the attractive sites for IMRT because of the complexity of the anatomy in this region, with many critical and radiation-sensitive tissues in close proximity to the targeted tumor. (3) Recently, the Radiation Therapy Oncology Group (RTOG) activated the first IMRT protocol for phase I/II study of oropharyngeal cancer, H-0022. (4) Conventional radiation therapy for advanced oropharyngeal tumors typically delivers high dose to the major salivary glands (parotid, submandibular, and sublingual) bilaterally. In most cases, this a

A. Target volumes and normal structures
Three CT image data sets of oropharyngeal (two left tonsil, one right tonsil) cancer patients, who were previously treated with IMRT, were used for this study; see Table 1 for the summary of the patient information.

A.1 Target volumes
All target volumes were defined according to RTOG H-0022 protocol. The gross target volume (GTV) included gross disease (tonsil) and palpable lymph nodes in the neck. The clinical target volume (CTV) for the GTV, CTV-66, was equal to GTV plus at least 5-mm margins. The planning target volume (PTV) for the CTV-66, PTV-66, was created with CTV-66 plus 5-mm margins to account for the setup uncertainties. The secondary target is the CTV of lymph node groups or surgical neck levels at risk of subclinical disease, CTV-54. CTV-54 defined for image data sets includes the right second, third, and fifth echelon nodes, the left second, third, and fifth echelon modes, retropharyngeal nodes, and submandibular nodes (level 1B). The PTV for the CTV-54, PTV-54, was created with CTV-54 plus 5-mm margins. RTOG H-0022 also allows an optional target volume (CTV-60) to be defined at the discretion of the treating physician. CTV-60 was defined for patients 2 and 3 ( Table 1). An example of target volumes and normal structures (patient 1) is shown in Fig. 1.

A.2 Critical normal structures
The defined critical normal structures include the spinal cord, parotid salivary glands, mandible, brainstem, glottic larynx, submandibular salivary glands, and skin surface. The spinal cord is expanded by a 5-mm margin to create a planning organ at risk (POAR) for this structure. The tissue within the skin surface and outside all other critical normal structures and PTVs is defined as nontarget tissue. The CT image data set for patient 1 was used for the "dry run" for the RTOG H-0022 protocol for our institution. All the contours defined, including both targets and normal structures, were reviewed and approved by the study chair of the protocol as well as the attending radiation oncologist.

B. Treatment planning
The treatment-planning software used in this work was Corvus (v.5, NOMOS Corporation, Swickley, PA). The input prescription data required by the Corvus treatment planning included the following: (1) 4-point prescription dose-volume histogram (DVH); (2) uncertainties for setup and targets/structures definition (for PTV generation); and (3) treatment unit, beam energy, number of beams, beam directions, and number of level of intensity modulations. The PTVs were assigned a dose goal, a percent volume that may be less than the goal dose, as well as minimum and maximum doses to be delivered. Each normal structure was assigned a dose limit, a percent volume that may receive more than the limit, as well as minimum and maximum doses. A simulated annealing optimization algorithm with the "Continuous Annealer" option was used in this work. This option was chosen because it gives the best resultant dose distributions for plans with complex prescriptions, according to the vendor. (24) All plans were done with inhomogeneity correction turned on during optimization and dose calculation. The initial plan for each image data set was executed using 9 IMBs and 5 ILs because it was the standard technique at that time in our clinic.
Treatment plans were developed using 5, 7, 9, and 15 evenly spaced and planner-selected 7 IMBs using 3, 5, 10, or 20 ILs for patient 1. For 15 IMBs, the plans with 10 and 20 ILs could not be generated because the planning system requires fraction monitor units, but Siemens LINACs do not support it. For patients 2 and 3, plans were developed using 5, 7, and 9 evenly spaced and planner-selected 7 IMBs using 3, 5, or 10 ILs. A total of 42 plans were developed using the three oropharyngeal cancer CT image data sets. The dosimetric criteria from RTOG H-0022 used include the following: 1. The prescription dose of 66 Gy (2.2 Gy × 30 fractions) encompasses at least 95% of the PTV of the gross target (PTV-66), no more than 20% (less than 25% for minor violation) of the PTV-66 receives greater than 110% of the prescribed dose (72.6 Gy), and no more than 1% of the PTV-66 receives less than 93% of the prescribed dose (61.4 Gy). 2. The prescription dose of 54 Gy (1.8 Gy × 30 fractions) encompasses at least 95% of the PTV of the neck lymph nodes (PTV-54), and no more than 20% of the PTV-54 receives greater than 110% of the prescribed dose to PTV-54 (59.4 Gy). For minor variation, the 47 Gy isodose surface covers no less than 99% of the PTV-54, the 54 Gy isodose surface covers no less than 90% of the PTV-54, and the 72.6 Gy isodose surface (110% of the PTV66 prescription dose) covers no more than 20% of the PTV-54. 3. Dose to the spinal cord plus a 5-mm margin is less than 45 Gy. 4. The mean dose to either period less than 26 Gy or at least 50% of the either parotid gland receives less than 30 Gy, or at least 20 cm 3 of the combined volume of both parotid glands receives less than 20 Gy.
The plans were generated for each image data set using the same optimization constraints but different numbers of IMB and IL. All plans were normalized so that 95% of the PTV-66 received at least 66 Gy.

C. Plan evaluation
The treatment plan quality was determined by assessing compliance with RTOG protocol H-0022 criteria. For plans meeting the dosimetric criteria of H-0022, the plans with more homogenous dose to the targets and lower doses to the structures were considered to have better quality. When DVH is used to assess the dose homogeneity to the target, the plans with larger percent of PTV receiving prescription dose and smaller percent of volume exceeding the limit dose (e.g., 110% of the prescription dose) were considered to have better quality. In an ideal situation, the DVH plots for target volumes should be a step function, 100% of volume receiving the prescription dose and no volume receiving doses greater than the prescription dose. In practice, this ideal DVH does not exist.
The treatment plan efficiency was accessed by number of segments of each plan. For the LINACs and R&V system used in this work, the average delivery time of each segment, not including the patient setup time, is approximately 0.2 min. Delivery time can be very quickly estimated based on the number of segments for each plan.

D. Treatment unit
All IMRT plans were generated for delivery on a Siemens digital Mevatron (Primus) with an sMLC and Lantis R&V using 6 MV X-rays. The dual-focused MLC consists of 29 pairs leaves, the inner 27 pairs projecting to 1.0-cm width at isocenter. (25)

III. RESULTS AND DISCUSSION
RTOG H-0022 for oropharyngeal cancer is the first multi-institutional prospective IMRT study for head and neck cancer. The prescription dose is 66 Gy (2.2 Gy × 30 fractions) to PTV-66, 54 Gy (1.8 Gy × 30 fractions) to PTV-54, and optionally 60 Gy (2.0 Gy × 30 fractions) to the PTV-60. The IMRT plans were initially conceived and designed to be delivered as a "simultaneous integrated boost." (8) Shown in Fig. 1 are images for patient 1 in axial, coronal, and sagittal planes with PTVs and normal structures. Also included in Fig. 1 are isodose lines for the plan generated with 9 IMBs and 3 ILs. Figure 2 shows DVHs for PTV-66 of patient 1 for plans with various IMBs and ILs. It should be emphasized that for comparison purposes, all plans were normalized such that 95% of the PTV-66 received the prescription dose, 66 Gy. In general, the more ILs, the better the plan, although there is very little difference between plans with 5, 10, and 20 ILs for 9 IMBs. There are exceptions to the observation for this particular image data set with the treatment-planning software (Corvus v.5). For example, as shown in Fig. 2(a), the plan with 9 IMBs and 20 ILs is not better than one with 10 ILs in terms of percent of volume exceeding the 110% of the prescription dose (72.6 Gy). This is an interesting observation and requires further investigation. The plans with 3 ILs for both 7 and 9 IMBs are not as good as plans with larger numbers of ILs in terms of percent of volume exceeding the 110% of the prescription dose (72.6 Gy). But both plans meet the dosimetric criteria of RTOG H-0022 for PTV-66 (i.e., less than 20% of the PTV-66 can receive 110% of the prescription dose) because the volume exceeding 72.6 Gy is 14.6% and 10.8%, respectively, for 7 and 9 IMBs, as listed in Table 2. In fact, for patient 1, all plans meet the dosimetric criteria for PTV-66, except the plan with 5 IMBs and 3 ILs. This plan fails because 40% of the PTV-66 received a dose of 72.6 Gy or more. For a given number of ILs, the larger number of IMBs, the better the plan, as shown in Figs. 2(c) and (d). The plans with 5 IMBs are clearly the worst in terms of DVH for PTV-66. There is another dosimetric requirement for PTVs, namely, no more than 1% of any PTV will receive less than 93% of its prescribed dose. For PTV-66, the 93% of the prescribed dose is 61.4 Gy. All plans generated this study meet this particular requirement. Similar results are observed for patients 2 and 3 (see Tables 2B and C).  * For 7 IMBs, the numbers in the front of slash are for evenly spaced 7 beams, and the numbers after the slash are for planner-selected 7 beams. ** For 15 IMBs, the plans with 10 and 20 ILs could not be generated due to the limitation of the combined delivery system and treatment-planning system.

B. PTV-60
DVHs for PTV-60 of patient 2 are displayed in Fig. 3. Again, for a given number of IMBs, the more ILs, the better the plan; for a given number of ILs, the more IMBs, the better plan. Similar results are observed for patient 3. C. PTV-54 Figure 4 displays DVHs for PTV-54 of patient 1 for plans with various IMBs and ILs. The dosimetric criterion for PTV-54, at least 95% of PTV-54 receives prescription dose, 54 Gy, and less than 20% of PTV-54 receives of 110% of the prescription dose, 59.4 Gy, is very difficult to meet. In fact, for each plan generated for patient 1, more than 80% of the PTV-54 received at least 59.4 Gy, which is significantly larger than 20%. The reason for this is that PTV-54 is in close proximity to the critical normal structures, such as the salivary glands and the spinal cord. To achieve the sharp dose drop-off at the critical normal structures, dose inhomogeneity in the PTV is larger. Recognizing this difficulty, RTOG H-0022 allows minor variation for PTV-54.
The criteria for such a minor protocol variation are as follows: 47 Gy isodose surface covers no less than 99% of the PTV-54, and the 54 Gy isodose surface covers no less than 90% of the PTV-54; the 72.6 Gy isodose surface (110% of PTV-66 prescription dose) covers no more than 20% of the PTV-54. All plans generated can meet the criteria for PTV-54 with minor variation, as shown in Fig. 4 and Table 3.    IMBs, the numbers in the front of slash are for evenly spaced 7 beams, and the numbers after the slash are for planner-selected 7 beams. ** For 15 IMBs, the plans with 10 and 20 ILs could not be generated due to the limitation of the combined delivery system and treatment-planning system.

D. Normal tissues
Shown in Fig. 5 are DVHs for the parotid salivary glands of patient 1. The dosimetric criteria for parotid glands are as follows: (1) the mean dose to either parotid less than 26 Gy or (2) at least 50% of the either parotid gland receive less than 30 Gy or (3) at least 20 cm 3 of the combined volume of both parotid glands receive less than 20 Gy. All plans meet requirement (2), as listed in Tables 4 and 5. Displayed in Fig. 6 are DVHs for the POAR of the spinal cord of patient 1. The dosimetric criterion is the maximum dose less than 45 Gy. Figure 6 clearly demonstrates that all plans meet this requirement. Similar results are observed for spinal cord of patients 2 and 3.   * For 7 IMBs, the numbers in the front of slash are for evenly spaced 7 beams, and the numbers after the slash are for planner-selected 7 beams. ** For 15 IMBs, the plans with 10 and 20 ILs could not be generated due to the limitation of the combined delivery system and treatment-planning system. IMBs, the numbers in the front of slash are for evenly spaced 7 beams, and the numbers after the slash are for planner-selected 7 beams. ** For 15 IMBs, the plans with 10 and 20 ILs could not be generated due to the limitation of the combined delivery system and treatment-planning system. * For 7 IMBs, the numbers in the front of slash are for evenly spaced 7 beams, and the numbers after the slash are for planner-selected 7 beams. ** For 15 IMBs, the plans with 10 and 20 ILs could not be generated due to the limitation of the combined delivery system and treatment-planning system. RTOG H-0022 requires that no more than 1% or 1 cm 3 of the tissue outside the PTVs receives more than 72.6 Gy (110% of the prescribed dose to PTV-66). Table 6 lists the dose and volume parameters for nontarget tissue. Each plan meets this requirement because each has less than 1%, but more than 1 cm 3 , of the tissue exceeding 72.6 Gy. Other critical structures include glottic larynx (2/3 below 50 Gy), brainstem (54 Gy), and mandible (70 Gy). The protocol encourages the participants to remain within these limits. However, some plans have a significant volume of nontarget tissue (still within 1%) receiving a dose greater than 72.6 Gy, despite meet-ing all the dosimetric criteria. For example, the plan for patient 1 with 7 IMBs and 3 ILs may be clinically unacceptable due to the larger volume (43 cm 3 ) of nontarget tissue receiving a dose greater than 72.6 Gy.

E. User-selected beam angle
The above discussion is focused on plans generated with IMBs with equally spaced gantry angles. It has been reported that for 9 IMBs and more, gantry angle selection may be less important, (8) while for other cases gantry angle selection may still be beneficial. (14) We therefore also tested user-selected beam angles using 7 IMBs. The general angle selection method was to use a beam's-eye view planning tool to maximize the target exposure and minimize normal structures exposure while avoiding parallel-opposed beams. Shown in Fig. 7 are DVHs for the plan generated by 7 user-selected IMBs compared with plans for 7 and 9 evenly spaced IMBs with 3 ILs. The plan with 7 user-selected IMBs is nearly identical to the plan with 9 evenly spaced beams and outperformed the plan with 7 evenly spaced beam in terms of PTV-66, as shown in Fig. 7(a). Similar DVHs are obtained for PTV-54, left parotid, and spinal cord between these three plans, as illustrated in Figs. 7(b) to (d). For patient 2, there is at least a 5% reduction in volume of each parotid gland receiving at least 30 Gy for the user-selected 7 beam plans compared with plans generated with 7 evenly spaced beams. For patient 3, only the right parotid gland shows significant (at least 5%) reduction (see Tables 4 and 5). This result demonstrates that for number of beams fewer than 9, beam angle selection and optimization could be important, which is consistent with the literature. (8,14)

F. Treatment efficiency
Treatment efficiency is governed by the number of segments for the delivery as well as the R&V and system used in this work. Table 7 shows a number of segments for each plan. It is obvious that the plans with the least number of IMBs and ILs has the fewest segments and therefore is the most efficient one to deliver. However, for patient 1, it is also the plan that fails to meet the dosimetric criteria of the H-0022 protocol and has the largest volume of the nontar-get tissue receiving a dose greater than 72.6 Gy. The plans with 7 or 9 IMBs and 3 or 5 ILs may be the optimal choices when both plan quality and delivery efficiency are considered. When number of beams used is fewer than 9, user-selected beam angles could be helpful for improving the quality of plan, while keeping the plan delivery efficient.
The delivery times for plans of patient 1 listed in Table 7 are estimated to be 8 min to 55 min, corresponding to 42 to 276 segments using the average value of 0.2 min each segment for our delivery and R&V system. The total delivery time includes beam-on time, the time needed for the MLC to move to the next shape, R&V system overhead time, and the time needed for the gantry to rotate to the next position. The R&V system has a relatively large overhead for the delivery system studied. For this reason, the total delivery time is best estimated by the number of segments, not the number of monitor units. For a given of number of segments, reducing the time for the other above-mentioned components would improve treatment efficiency. A significant gain in treatment efficiency could be achieved by reducing the overhead of the R&V system. This would necessarily require the LINAC and R&V system manufacturers to work together to provide such a solution. For the delivery and R&V system studied here, reducing the number of segments is the most effective way of improving delivery efficiency. For example, using the image data set of patient 1, one could select the plan with 7 user-selected IMBs and 3 ILs. This plan meets the dosimetric criteria of H-0022 with minor variation yet requires approximately 13 min to be delivered, which is perhaps not significantly longer than some conventional head and neck treatment approaches, particularly when posterior electron and nodal boosting required.
As a result of this study, we have routinely used 9 IMBs with 3 ILs or 5 IMBs with 7 ILs for head and neck cancers, including oropharynx (tonsil, base of tongue, and palate), nasopharynx, unknown primary, pyrimform sinus, and hypopharynx. Reviewing selected head and neck patients treated using 9 IMBs with 5 ILs (8 patients) or 3 ILs (24 patients), we found that, on average, the number of segments is reduced from 143±10 to 101±10 and the number of monitor units from 1983±341 to 1470±247, as shown in Table 8. This represents a reduction of approxi- mately 30% in either the number of segments or the number of monitor units. With this approach, the treatment time in our clinic for these types of patients has been reduced from approximately 30 min to 20 min. Patients included in this comparison were those who had at least one primary target treating to dose 60 Gy to 70 Gy and the bilateral neck nodal regions, including the second, third, and fifth echelon nodes, to 45 Gy to 54 Gy.

G. Delivery accuracy
For Siemens LINACs such as the one used in this studied, only integer monitor units can be delivered. As the numbers of beams and intensity levels are increased, the relative dosimetric error on each segment may increase due to the effect of monitor unit linearity. To estimate this error, we have analyzed the plan with 9 IMBs and 20 ILs, which has the most segments for patient 1. The histogram of monitor units for this plan is shown in Fig. 8. It is obvious that there are a significant number of segments with 2 or 3 monitor units. Assuming the monitor unit linearity error is about 5% for 1 monitor unit, 2% for 2 to 4 monitor units, and 1% for 5 and more monitor units , (26) the largest error is about 1.5% for one of the IMBs and the average error over 9 IMBs is 1.3%. In this work, we attempted to reduce the number of segments to improve delivery efficiency. With fewer segments there are fewer segments with very small monitor units. Therefore the error is smaller. For example, for the plans with 9 IMBs and 3 ILs and 7 IMBs and 5 ILs of patient 1, the minimum monitor units are 12 and 11, respectively. Thus the error is expected to be about 1% or less.

IV. CONCLUSION
We have performed a detailed comparative treatment-planning study for three CT image data sets of oropharyngeal cancer patients. Treatment-planning quality and delivery efficiency for sMLC delivered IMRT treatments are evaluated using RTOG H-0022 dosimetric criteria. Both number of IMBs and ILs have a significant influence on the quality and efficiency of IMRT treatment using sMLC delivery. A treatment plan that uses a moderate number of IMBs and ILs, such as 7 or 9 IMBs with a 3 or 5 ILs, appears to be the optimal approach when both plan quality and delivery efficiency are considered. It should be emphasized that this study is specifically for the Corvus planning system combined with the sMLC on a Siemens LINAC and R&V system (Lantis). The conclusions drawn from this study should not be applied directly to a different planning, delivery, and R&V system. Even for an identical system, the conclusions should not be used without careful analysis and verification by the treating physician and physicist. Nevertheless, the methodology presented here could be adopted by clinics with similar hardware and software attempting to improve delivery efficiency but cannot access more advanced techniques such as beam angle optimization and efficient leaf-sequencing algorithms.