Int. J. Radiation Oncology Biol. Phys., Vol. 53, No. 2, pp. 453–463, 2002
`Copyright © 2002 Elsevier Science Inc.
`Printed in the USA. All rights reserved
`0360-3016/02/$–see front matter
`
`PII S0360-3016(02)02777-3
`
`PHYSICS CONTRIBUTION
`
`CLINICAL IMPLEMENTATION OF INTENSITY-MODULATED ARC THERAPY
`
`CEDRIC X. YU, D.SC., X. ALLEN LI, PH.D., LIJUN MA, PH.D., DONGJUN CHEN, PH.D.,
`SHAHID NAQVI, PH.D., DAVID SHEPARD, PH.D., MEHRDAD SARFARAZ, PH.D.,
`TIMOTHY W. HOLMES, PH.D., MOHAN SUNTHARALINGAM, M.D., AND CARL M. MANSFIELD, M.D.
`Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD
`
`Purpose: Intensity-modulated arc therapy (IMAT) is a method for delivering intensity-modulated radiation
`therapy (IMRT) using rotational beams. During delivery, the field shape, formed by a multileaf collimator
`(MLC), changes constantly. The objectives of this study were to (1) clinically implement the IMAT technique, and
`(2) evaluate the dosimetry in comparison with conventional three-dimensional (3D) conformal techniques.
`Methods and Materials: Forward planning with a commercial system (RenderPlan 3D, Precision Therapy
`International, Inc., Norcross, GA) was used for IMAT planning. Arcs were approximated as multiple shaped
`fields spaced every 5–10° around the patient. The number and ranges of the arcs were chosen manually. Multiple
`coplanar, superimposing arcs or noncoplanar arcs with or without a wedge were allowed. For comparison,
`conventional 3D conformal treatment plans were generated with the same commercial forward planning system
`as for IMAT. Intensity-modulated treatment plans were also created with a commercial inverse planning system
`(CORVUS, Nomos Corporation). A leaf-sequencing program was developed to generate the dynamic MLC
`prescriptions. IMAT treatment delivery was accomplished by programming the linear accelerator (linac) to
`deliver an arc and the MLC to step through a sequence of fields. Both gantry rotation and leaf motion were
`enslaved to the delivered MUs. Dosimetric accuracy of the entire process was verified with phantoms before
`IMAT was used clinically. For each IMAT treatment, a dry run was performed to assess the geometric and
`dosimetric accuracy. Both the central axis dose and dose distributions were measured and compared with
`predictions by the planning system.
`Results: By the end of May 2001, 50 patients had completed their treatments with the IMAT technique. Two to
`five arcs were needed to achieve highly conformal dose distributions. The IMAT plans provided better dose
`uniformity in the target and lower doses to normal structures than 3D conformal plans. The results varied when
`the comparison was made with fixed gantry IMRT. In general, IMAT plans provided more uniform dose
`distributions in the target, whereas the inverse-planned fixed gantry treatments had greater flexibility in
`controlling dose to the critical structures. Because the field sizes and shapes used in the IMAT were similar to
`those used in conventional treatments, the dosimetric uncertainty was very small. Of the first 32 patients treated,
`the average difference between the measured and predicted doses was ⴚ0.54 ⴞ 1.72% at isocenter. The
`80%–95% isodose contours measured with film dosimetry matched those predicted by the planning system to
`within 2 mm. The planning time for IMAT was slightly longer than for generating conventional 3D conformal
`plans. However, because of the need to create phantom plans for the dry run, the overall planning time was
`doubled. The average time a patient spent on the table for IMAT treatment was similar to conventional
`treatments.
`Conclusion: Initial results demonstrated the feasibility and accuracy of IMAT for achieving highly conformal
`dose distributions for different sites. If treatment plans can be optimized for IMAT cone beam delivery, we expect
`IMAT to achieve dose distributions that rival both slice-based and fixed-field IMRT techniques. The efficient
`delivery with existing linac and MLC makes IMAT a practical choice. © 2002 Elsevier Science Inc.
`
`Intensity-modulated radiation therapy (IMRT), Dynamic radiation therapy, Intensity modulation, Arc therapy.
`
`INTRODUCTION
`
`Stemming from the increasing evidence that improved local
`tumor control may enhance long-term survival (1, 2) and
`reduce the cost of cancer treatments (3), intensity-modu-
`
`Reprint requests to: Cedric X. Yu, D.Sc., Department of Radi-
`ation Oncology, University of Maryland School of Medicine, 22 S.
`Greene Street, Baltimore, MD 21201. Tel: (410) 328-0324; Fax:
`(410) 328-2618; E-mail: Cyu002@umaryland.edu
`This work was
`supported in part by NIH Grant No.
`R29CA66075. The authors appreciate the support and technical
`
`lated radiation therapy (IMRT) is receiving increasing in-
`terest and acceptance in radiation oncology. Presently, sev-
`eral IMRT techniques have been proposed. One method is
`to use multiple coplanar and noncoplanar beams at different
`orientations, each beam having spatially modulated intensi-
`
`assistance from Elekta Oncology Systems, Inc.
`Acknowledgment—The authors would like to thank Dr. Matt Earl
`and Dr. Lance Weems for their contributions.
`Received Mar 2, 2001, and in revised form Feb 8, 2002. Ac-
`cepted for publication Feb 15, 2002.
`
`Varian (Ex. 1007)
`IPR of U.S. Pat. No. 7,961,843
`
`453
`
`

`

`454
`
`I. J. Radiation Oncology ● Biology ● Physics
`
`Volume 53, Number 2, 2002
`
`ties (4–15, 19). Another approach, referred to as tomo-
`therapy, delivers the treatment in multiple slices, with each
`slice of the target volume treated with temporally modulated
`fan beams rotating around the patient (16, 17). Each of these
`two methods has its advantages and disadvantages in dose
`conformity and in efficiency of dose delivery, as discussed
`by Webb (18), Brahme (19), and Yu (20). In general, the
`tomotherapy approach spreads the normal tissue dose over a
`greater volume and produces a tighter dose conformation to
`the target.
`We have implemented a new technique, intensity-modu-
`lated arc therapy, or IMAT, to deliver highly conformal
`dose distributions by combining gantry rotation and dy-
`namic multileaf collimation. Instead of delivering intensity-
`modulated beams with fixed gantry angles, IMAT delivers
`optimized dose distributions by rotating the radiation beam
`around the patient. During delivery, the field shape, which is
`formed by a multileaf collimator (MLC), changes continu-
`ously as determined by the treatment plan. Intensity distri-
`butions at all angles around the patient are achieved with
`multiple overlapping arcs, with each arc having a different
`set of field apertures. The weight of the arcs, or total MUs
`delivered in different arcs, are typically different. Therefore,
`IMAT is also different from tomotherapy (16, 17), which
`uses intensity-modulated fan beams rotating around the
`patient, delivering the treatment slice by slice. As with
`tomotherapy, IMAT combines intensity modulation and ro-
`tational delivery. A detailed description of the technique
`was reported by Yu in a previous article (20).
`A Phase I clinical trial using the dynamic MLC and
`rotational delivery technique was approved by the institu-
`tional review board to assess the feasibility and safety of the
`technique. From November 1999 to May 2001, 50 patients
`with cancers of the central nervous system, head and neck,
`and prostate were treated in our clinic using the IMAT
`technique. This article describes the issues in the implemen-
`tation and clinical usage of this technique. Clinical exam-
`ples of IMAT treatments will be presented to illustrate the
`dosimetric advantages of rotational delivery.
`
`METHODS AND MATERIALS
`As with conventional treatment techniques, IMAT in-
`volves treatment planning and delivery. It has been demon-
`strated that
`treatment plans developed for tomotherapy
`treatment delivery can be converted into multiple arcs and
`delivered with IMAT (20). The same inverse treatment
`planning system has also been adapted to MLC delivery
`(CORVUS, NOMOS Corp., Sewickley, PA) (17). However,
`because the treatment plans are optimized with a simulated
`annealing algorithm with little constraint on smoothness of
`the beam intensities, the beam intensity distributions are
`overly modulated. Because we approximate an arc delivery
`as equally spaced beams, the number of beams is usually
`large. Intuitively, as the number of beams increases, the
`degree of intensity modulation required to meet the dosi-
`metric constraints should be reduced. With simulated an-
`
`nealing algorithm and without any constraint on the smooth-
`ness of the intensity maps, however, the result is just the
`opposite. More beams generally increase the randomness of
`the intensity patterns. As a result, a plan with two or three
`intensity levels would typically require more than 10 arcs to
`deliver.
`To overcome such inefficiency, and as the first step in
`using rotational delivery with dynamic MLC, we imple-
`mented IMAT into clinical use with forward planning. From
`simulation CT images, the target and surrounding normal
`structures are delineated on a commercial three-dimensional
`(3D) treatment planning system (RenderPlan 3D, Precision
`Therapy, Inc., Norcross, GA). Arcs are approximated as
`multiple shaped fields spaced every 5–10° around the pa-
`tient. The ranges of the arcs are chosen manually to give the
`desired dose distributions. Multiple coplanar or noncoplanar
`arcs are allowed. Wedges are often used in combination
`with dynamic field shaping to achieve a more uniform dose
`distribution in the planning target volume. At each beam
`angle,
`irregular field shapes are defined based on the
`beam’s-eye–view (BEV) of the planning target and normal
`critical structures. Depending on the normal structure toler-
`ance, the regions in the BEV where the projection of the
`target and the normal critical structure overlap may be
`blocked at some or all beam angles. When such overlap
`region is in the center of the BEV of the target and blocking
`is desired, the MLC-shaped fields cover only the part of the
`target on one side of the critical structure. The other side
`will be irradiated with another arc. Superimposing arcs are
`often used. For example, one arc may cover the BEV of the
`target, including the region where the projections of the
`target and critical structure overlap, and a second overlap-
`ping arc that excludes the overlap region may be used to
`provide the required sparing for the critical structure. Typ-
`ically, two to five arcs spanning an angular range 40–180°
`are used. For dose calculation, each arc is approximated
`with fixed beams equally spaced at 10° intervals. The MLC
`field shapes of these fixed beams are arranged in the order
`of delivery to form the MLC leaf sequence. To keep the
`gantry speed constant for smooth delivery, the weights of
`the beams, i.e., the relative contributions of different beams
`to the dose prescription point, are determined for each arc
`such that each beam angle delivers the same number of
`MUs. This automatically allows the beams with shallower
`radiologic depth to the prescription point to have greater
`dose contributions. For most of the treatment plans, the
`weights of different arcs are adjusted manually to achieve
`acceptable target uniformity and critical structure sparing.
`Once a satisfactory dose distribution is generated, the plan
`is analyzed, as with conventional 3D conformal plans.
`For all patients intended to receive IMAT treatment, a
`conventional 3D conformal plan was independently gener-
`ated by a different planner. Comparisons of dose distribu-
`tions and dose–volume histograms (DVHs) were made by
`the physician. The IMRT technique was used only when the
`physician determined that there was an advantage of IMAT
`treatment over 3D conformal treatment. This comparison
`
`

`

`Clinical implementation of IMAT ● C. X. YU et al.
`
`455
`
`also allowed us to gain experience in the types of cases that
`were more suitable for rotational delivery. Once IMAT
`technique was chosen over conventional 3D conformal
`technique, the plan was read by a leaf-sequencer developed
`at our institution. Because the plan already contained the
`field shapes at all angles, the leaf sequencer simply con-
`verted the shapes into MLC field segments. Because of the
`ways the field shapes at each beam angle were determined,
`the field shapes at neighboring angles typically did not
`differ significantly. As a result, MLC leaves were not re-
`quired to travel large distances from one angle to the next.
`For most cases, gantry rotation speed, rather than leaf trav-
`eling speed, was the factor limiting the dose rate. The MLC
`prescriptions generated by the leaf sequencer were then sent
`to the MLC controller for dynamic delivery through a local
`network link.
`The IMAT delivery is implemented on an MLC system
`equipped on a digitally controlled linear accelerator (SL-20
`linear accelerator with MLCi, Elekta Oncology Systems,
`Inc., Norcross, GA) (21). It consists of 40 pairs of opposing
`leaves, each free to move along its length and projecting 1
`cm in width in the isocenter plane at 100 cm from the
`source. Complementary to the 80 leaves are two pairs of
`backup diaphragms (solid tungsten jaws) in the x and y
`directions, respectively. Both the leaves and the backup
`diaphragms are used for defining the dynamic MLC seg-
`ments. During beam delivery, the linac is programmed to
`deliver arc treatments, and the MLC is programmed to
`dynamically step through a sequence of field shapes. Both
`gantry rotation and leaf motion are coupled to the delivered
`MUs. As a result, although fluctuations in machine dose rate
`can cause the gantry to rotate with changing speed, the
`effect on dose delivery is minimal. It is important to under-
`stand that the field shapes are only defined at a set of beam
`angles spaced 10° apart. In between two successive beam
`angles, the MLC controller linearly interpolates the leaf
`positions. Therefore, the leaves are moving continuously
`throughout the delivery, unless the leaf positions of two
`successive segments are the same.
`To verify the clinical value of such a simplified process,
`we compared the treatment plans using two to five forward
`planned arcs with the plans generated by conventional tech-
`niques and by a commercial inverse planning system (COR-
`VUS by NOMOS Corp., Inc., Sewickley, PA) for treatment
`of head-and-neck cancers, central nervous system tumors,
`and the prostate. Both the conventional plans and the IMAT
`plans are presented to the physician.
`An anthropomorphic phantom (Alderson Rando Phan-
`tom, Alderson Research Laboratories, Inc., Stamford, CT)
`was modified for dosimetric verification. Multiple original
`slices of the phantom at different sites were replaced with
`two sheets of water-equivalent plastic material of the same
`shape, each with half the original slice thickness. Grooves
`were made on each of the two plastic sheets, so that an ion
`chamber could be placed at various positions along the
`horizontal axis. Radiographic film (XV-2, Kodak, Roches-
`ter, NY) cut to the exact shape of the plastic sheets could
`
`also be placed between the two plastic sheets without the
`chamber groove for relative dose measurement. The modi-
`fied phantom was scanned on our CT-simulator unit, and the
`images were imported to the planning system. The use of
`the anthropomorphic phantom for the dry run allowed us not
`only to verify the absolute dose quantitatively, but also to
`make a visual comparison of the dose pattern to the patient
`plan and to identify setup problems or difficulties before
`treatment.
`Dosimetric accuracy of the entire process was verified
`with phantoms before IMAT was used clinically. Shaped
`fields ranging from 4 cm ⫻ 4 cm to 30 cm ⫻ 30 cm spaced
`5–20° were used for approximating an arc. The sequence of
`shaped fields with drastically changing field shapes was
`delivered to a cubical phantom both individually as calcu-
`lations were carried out in the plan and in arc fashion.
`When approximating an arc with multiple fixed fields,
`each field was essentially a sample within the range of the
`arc. Each sample should be treated as though it were at the
`center of the interval. That is, if a 10° interval is used, an arc
`should start 5° ahead of the first field and end 5° beyond the
`last field. Because the field shapes were not defined beyond
`the angles at both ends of the arc, we chose not to have the
`arc extending beyond the angles of the first and last fields.
`To keep the plan and delivery consistent, we set the MUs of
`the first beam and the last beam to be one-half of those of
`the other beams within the same arc.
`Because the planning system was not designed for IMAT
`planning, the MUs provided by the planning system had to
`be adjusted to achieve accurate delivery. Although the pri-
`mary arc commonly used large fields, the overlapping arcs
`were generally small and possibly off axis. A dual source
`model that accounted for the 3D geometry of the collimat-
`ing system (22) was used to calculate the head scatter. The
`ratio of our calculation result to the value predicted by the
`planning system was used to adjust the total MUs. For
`IMAT treatment, a plan might include more than 50 beams.
`The number of MUs for the beams in the overlapping arcs
`could be very small, in the range 3–5 MUs per beam.
`Because no fractional MU is allowed in the planning sys-
`tem, the rounding error for each beam could cause an arc to
`give a dose contribution that differs from the intended value.
`The result would be a total dose to the prescription point
`that deviates from the prescription dose by a small percent-
`age (1%–2%). To correct for such a rounding error, the total
`MU of an arc was adjusted also by the ratio of the intended
`contribution to the contribution used by the plan.
`For all IMAT treatments, a dry run was conducted to
`assess the geometric and dosimetric accuracy and to elim-
`inate possible technical problems, such as setup difficulties
`and MLC movement constraints. The dry run was per-
`formed by copying the patient plan parameters, including
`field shapes and MUs, to the same site of the modified
`humanoid phantom. When it was required to measure dose
`distributions in sagittal planes, common rectangular phan-
`toms consisting of water-equivalent plastic slabs were also
`used. Both the central axis dose and the dose distribution on
`
`

`

`456
`
`I. J. Radiation Oncology ● Biology ● Physics
`
`Volume 53, Number 2, 2002
`
`a plane were measured and compared to those generated by
`the planning system. The accelerator output variation at the
`time of verification was factored out by either performing an
`output reading with the same ion chamber or by using the
`output reading obtained in daily quality assurance. Because
`the speed of leaf travel is limited, it is important that the
`field shapes of adjacent beam angles not differ too much.
`For most clinical cases, the field shapes varied slowly be-
`tween angles. Treatments could be delivered with the high-
`est machine dose rate determined by the linac based on the
`maximum speed of gantry rotation, resulting in very short
`delivery time. There were cases where large leaf travel was
`required, and the leaf motion lagged radiation delivery.
`Once the leaf was behind the desired position for the deliv-
`ered MU by a preset amount, 1 mm in our case, either
`radiation pause, which could recover automatically if leaf
`reaches position within a second, or termination, which
`could not recover by itself, would occur. In such cases, a
`reduced nominal accelerator dose rate setting was used.
`Although treatment could be resumed after an interruption,
`such radiation pause or termination could increase delivery
`time. Although we could predict such occurrences and se-
`lect a dose rate by using the maximum leaf speed, the total
`MU, and the fastest gantry speed, we did not do so in the
`trial because of the rarity of such occurrence and because of
`the use of dry runs for dosimetric verification. If the need for
`a reduced machine dose rate was observed during the dry
`run quality assurance procedure, the right machine dose rate
`setting would be selected for treatment delivery.
`Detailed linac prescriptions for treatment were entered in
`the linac control after the dry run. The linac was pro-
`grammed to deliver an arc treatment with or without
`wedges, and the MLC was programmed to step through a
`sequence of field shapes. Each arc was treated as a separate
`beam and delivered separately.
`
`RESULTS
`Between November 1999 and May 2001, a total of 50
`patients were treated using the IMAT technique. Of the 50
`patients, 13 had cancers of the central nervous system, 16
`had prostate cancer, two had thoracic cancer, three had
`gastrointestinal cancers, and 16 patients had head-and-neck
`cancers. For 31 of the 50 patients, IMAT was used for the
`final boost, with the total number of fractions ranging from
`8 to 12. For 19 of the 50 patients, IMAT was used for the
`entire course of treatment.
`For complex cases, the time needed to create a satisfac-
`tory treatment plan was found to be longer than for conven-
`tional 3D conformal plans. This is largely because of the
`number of fields that the planner must specify and outline
`for IMAT treatment plans. For cases where a precalculated
`treatment plan template could be used, such as for the
`treatment of prostate cancer, the planning times were similar
`to those for conventional planning. However, because of the
`need to create phantom plans for the dry run and 3D
`conformal plans for comparison, the overall planning time
`
`tripled. The dry run quality assurance procedure also takes
`an additional 1–2 h per course of treatment. We have been
`continually modifying our quality assurance procedures to
`speed up the dry runs. Alternative quality assurance proce-
`dures, such as the use of electronic portal imaging systems,
`are being investigated.
`To determine the acceptable spacing of fields used for
`approximating an arc, we performed measurements using
`different field shapes and different angular spacing between
`fields. It was found that spacing of the fields from 5° to 20°
`did not change the central axis dose or the target dose
`coverage for the same total MUs. However, dose distribu-
`tions outside the target, especially at low isodose levels near
`the surface, differed between calculations with fixed fields
`and those delivered with arc beams as angular spacing
`increased. Figures 1a–d illustrate planning results using a
`fixed field width of 5 cm but different angular spacing to
`approximate a 150° arc. In all four figures, the isodose
`levels from the center outward are 95%, 80%, 50%, 30%,
`20%, and 10%, respectively. Figure 1a is with fields spaced
`every 3°, which most closely approximates a continuous arc
`delivery. All isodose lines from 10% to 95% are smooth, as
`expected in an arc delivery. Figures 1b–d are dose distri-
`butions with fields spaced 5°, 10°, and 15°, respectively. As
`the angular spacing increased, lower isodose levels started
`to show ripples. However, for all angular spacing used, the
`isodose lines of 80% and 95% remained the same. The
`rippling appearance on isodose lines near the surface can be
`explained by the gaps in geometric overlap of the fixed
`beams. We found that a spacing of 10° represents a good
`compromise. If the target is small and the lower dose areas
`coincide with critical structures, a finer angular spacing of
`5° should be used.
`Preclinical dose verifications were conducted, starting
`with simple spherical targets in a water-equivalent cubic
`phantom. Plans were generated to test the dosimetric accu-
`racy of the entire process from planning to delivery. For
`these simple and well-controlled cases, we expected perfect
`agreement between the calculations and measurements. It
`was quickly realized that the treatment planning system did
`not properly model the MLC for head scatter. Because the
`MLC, which replaces the upper jaw of the secondary col-
`limator, is used for shaping the fields, the head scatter
`should be determined based on the irregular MLC-shaped
`fields. However, the treatment planning system uses the
`rectangle circumscribing the irregular field to estimate the
`head scatter factor. For small field sizes, an overestimation
`of the equivalent field size by 2 cm can cause 2%–3%
`errors. To correct for the modeling deficiency of the plan-
`ning system, we used a dual source model with consider-
`ation of leaf thickness and shape (22) to calculate the head
`scatter factors of all the fields approximating an arc. The
`total MU of the arc was then adjusted upward by the ratio of
`the head scatter factors predicted by the planning system to
`that obtained with our dual source model. After the correc-
`tions were made for all field shapes, the agreements between
`plan calculation and measurements were all within 1% for
`
`

`

`Clinical implementation of IMAT ● C. X. YU et al.
`
`457
`
`Fig. 1. An illustration of the effect of angular spacing on the accuracy of using multiple fixed fields to approximate an
`arc delivery. The dose distributions show 95%, 80%, 50%, 30%, 20%, and 10% levels from the innermost to the
`outermost isodose lines. (a–d) The isodose distributions shown were obtained with the angular spacing of 3°, 5°, 10°,
`and 15°, respectively.
`
`the simple test cases. For patient-specific verifications, all
`absolute dose measurements were found to be within ⫾3%
`of the calculated values, except for one case, where one of
`the arcs was at sharp angles to the stem of the ion chamber,
`and the effect was not corrected. Figure 2 shows the scat-
`tered plot of the discrepancies between the plan predicted
`
`and the absolute dose measurements for the first 32 patients
`treated with the IMAT technique. The quantities are ex-
`pressed as (Measured ⫺ Predicted)/Measured ⫻ 100%. The
`mean error was found to be ⫺0.54%, and the standard
`deviation was 1.72%.
`Isodose distributions were also measured with films. Fig-
`
`

`

`458
`
`I. J. Radiation Oncology ● Biology ● Physics
`
`Volume 53, Number 2, 2002
`
`ures 3a–d show a clinical example of the planned and
`measured isodose distributions for the treatment of an
`ependymoma. Figure 3a is the dose distribution predicted
`by the planning system in the patient. Unlike the traditional
`treatment with anterior-posterior setup, IMAT uses two
`wedged posterior arcs and one anterior arc overlapped with
`two fixed fields. By spreading lower doses to a greater
`volume, the 90% isodose covers the tumor uniformly and
`the dose dropoff from 90% to 80% within 4 mm, except at
`the superior and inferior ends. The plan is applied to a
`plastic phantom for the dry run. Figure 3b shows the isodose
`distribution calculated for the phantom by applying the
`same beam arrangements and weightings as for the patient
`plan. It shows a dosimetric pattern similar to that of the
`patient plan. The white area is encompassed by the 90%
`isodose line. Figure 3c is the grayscale image of the film
`used for relative isodose measurements. The film was ex-
`posed by sandwiching it between two slices of the plastic
`phantom and delivering the same treatment as for the patient
`with scaled-down total MUs. The grayscale image for the
`film was converted to dose with a precalibrated relationship.
`
`Fig. 2. Absolute dose discrepancies between plan predictions and
`measurements for the first 32 patients. The quantities are expressed
`in (Measured ⫺ Predicted)/Measured ⫻ 100%.
`
`Fig. 3. A clinical example of the planned and measured dose distributions for the treatment of an ependymoma showing
`(a) dose distribution predicted by the planning system in the patient, (b) dose distribution calculated for the phantom by
`applying the same beam arrangements and weightings as for the patient plan, (c) the inverted grayscale image of the film
`used for relative isodose measurements, and (d) resultant measured dose distribution after converting grayscales to dose.
`
`

`

`Clinical implementation of IMAT ● C. X. YU et al.
`
`459
`
`Fig. 4. An example comparing (a) the conventional 3D plan, (b) the IMAT plan, and (c) the inverse plan generated with
`the CORVUS system for an adenoid cystic carcinoma of the maxillary sinus. The DVHs generated by the three plans
`are shown in (d), where the solid lines are for the IMAT plan, the dashed lines are for the 3D plan, and the dotted lines
`are for the inverse plan. DVHs for four structures, including the PTV, the brainstem, and the left and right eyes, are
`compared.
`
`The resultant measured isodose distribution is shown on
`Fig. 3d). As compared with what was predicted by the
`planning system (Fig. 3b), for isodose lines at levels higher
`than 80%, the agreement with calculation is within 2 mm.
`Because of the simplifications in scatter dose calculation
`used by the treatment planning system, the spacing of the
`isodose lines from 80% to 100% is consistently tighter than
`that measured with film. For lower dose levels, such as
`those at or below the 50% level, which typically lie in the
`shallow gradient region with rotational delivery, the dis-
`agreement can be larger than 5 mm in some areas. This is
`caused by both the simplified scatter dose calculation algo-
`rithms and the intrinsic inaccuracy of film dosimetry for
`photon beams. We set an acceptance criterion that
`the
`predicted and measured isodose lines at levels greater than
`
`90% in the phantom must match within 2 mm. When doses
`to critical structures needed to be verified more accurately,
`additional point dose measurements had to be carried out.
`An example comparing the IMAT plan, the inverse plan
`by the CORVUS system, and the conventional 3D plan for
`an adenoid cystic carcinoma of the maxillary sinus is shown
`in Fig. 4. Isodose distributions are shown on the transverse
`and sagittal planes with the regions covered by the 95%
`isodose contour highlighted (darker). Figure 4a is the con-
`ventional plan with five fields, including anterior, left lateral
`and posterior, and vertex fields. Wedges were used in all the
`fields. Figure 4b is the IMAT plan with five coplanar arcs.
`They include a primary arc from 355° to 175°, two over-
`lapping arcs from 45° to 95°, one overlapping arc from 115°
`to 175°, and a primary arc from 225° to 275°. In the primary
`
`

`

`460
`
`I. J. Radiation Oncology ● Biology ● Physics
`
`Volume 53, Number 2, 2002
`
`Fig. 4. (Cont’d)
`
`arcs, the radiation field was shaped as the beam’s-eye–view
`of the target, excluding the beam’s-eye–view of the left eye.
`In all other arcs, the field shapes excluded the beam’s-eye–
`view of both eyes and brainstem, as well as areas in the
`target that already received a hot spot from the primary arcs.
`Figure 4c shows the inverse plan with seven coplanar in-
`tensity-modulated fields, each with 20 intensity levels.
`The DVHs for four structures, including the planning
`target volume (PTV), the brainstem, and the left and right
`eyes, generated by the three plans are shown as Fig. 4d. The
`solid lines represent results of the IMAT plan. The dashed
`lines represent results of the conventional 3D plan. The
`dotted lines are the results of the inverse plan by the COR-
`VUS system. All three plans showed similar dose coverage
`
`to the PTV, with the inverse plan showing a slightly more
`uniform dose. The inverse plan showed an overall lower
`dose to the brainstem but a maximum dose slightly higher
`than that found in the other two plans, because of a slight
`increase in the dose to both eyes. The 3D plan was not able
`to spare the left eye. Both the IMAT plan and the inverse
`plan were clinically acceptable. The IMAT plan was chosen
`because the delivery of five coplanar arcs was much less
`time-consuming than the delivery of seven intensity-modu-
`lated fields with our mixed dynamic and step-and-shoot
`sliding window technique.
`Figure 5 shows the same type of comparison of DVHs as
`in Fig. 4d for a prostate cancer treatment. For the IMAT
`plan and the 3D conformal plan, a 1-cm margin was used
`
`

`

`Clinical implementation of IMAT ● C. X. YU et al.
`
`461
`
`Fig. 5. Comparison of DVHs generated by conventional 3D plan (thin solid lines), the IMAT plan (thick solid lines),
`and the inverse plan by CORVUS (dashed lines) for the treatment of prostate cancer. Two structures, the GTV and
`rectum, are compared.
`
`around the gross tumor volume to derive the planning
`target volume. In the inverse plan using a commercial
`inverse planning system (CORVUS, NOMOS Corp.,
`Sewickley, PA), a 1-cm margin was assigned at
`the
`“prescription” stage, so the optimizer could get better
`target coverage. The IMAT treatment used four arcs. In
`two of the four arcs, an internal 60° wedge was used,
`together with dynamic MLC field shaping. At each gantry
`angle, the field covers only part of the planning target
`BEV on one side of the rectum. Therefore, the rectum
`was shielded in these two arcs. The other two arcs use the
`MLC-shaped irregular fields to cover the BEV of the
`entire planning target at all beam angles. The weighting
`between the partial target BEV arcs with wedge and the
`open arcs is 1:4. By blocking the BEV projection of the
`critical organ during beam rotation, it is