Online daily assessment of dose change in head and neck radiotherapy without dose‐recalculation

Abstract Background Head and neck cancers are commonly treated with radiation therapy, but due to possible volume changes, plan adaptation may be required during the course of treatment. Currently, plan adaptations consume significant clinical resources. Existing methods to evaluate the need for plan adaptation requires deformable image registration (DIR) to a new CT simulation or daily cone beam CT (CBCT) images and the recalculation of the dose distribution. In this study, we explore a tool to assist the decision for plan adaptation using a CBCT without re‐computation of dose, allowing for rapid online assessment. Methods This study involved 18 head and neck cancer patients treated with CBCT image guidance who had their treatment plan modified based on a new CT simulation (ReCT). Dose changes were estimated using different methods and compared to the current gold standard of using DIR between the planning CT scan (PCT) and ReCT with recomputed dose. The first and second methods used DIR between the PCT and daily CBCT with the planned dose or recalculated dose from the ReCT respectively, with the dose transferred to the CBCT using rigid registration. The necessity of plan adaptation was assessed by the change in dose to 95% of the planning target volume (D95) and mean dose to the parotids. Results The treatment plans were adapted clinically for all 18 patients but only 7 actually needed an adaptation yielding 11 unnecessary adaptations. Applying a method using the daily CBCT with the planned dose distribution would have yielded only four unnecessary adaptations and no missed adaptations: a significant improvement from that done clinically. Conclusion Using the DIR between the planning CT and daily CBCT can flag cases for plan adaptation before every fraction while not requiring a new re‐planning CT scan and dose recalculation.


| INTRODUCTION
Radiation therapy is a standard treatment option for a variety of cancers, where the precise geometric targeting of tumors can be exploited for achieving better tumor control while limiting healthy tissue damage. The specific targeting and attenuation of radiation are unique to the patient's anatomy at the time of the planning CT (PCT) simulation, but these conditions are difficult to maintain throughout an entire course of treatment due to changes in anatomy. [1][2][3][4] To account for changes in patient anatomy, plan modification may be required during the treatment course to ensure accurate targeting. Plan adaptation has been shown to improve treatment outcomes by promoting better tumor control and limiting toxicities, 5,6 but this procedure entails additional costs of re-imaging, replanning, and additional quality assurance. Although the potential benefits of plan adaptation are obvious, no guidelines on decisionmaking and optimal time for re-planning are available.
Plan adaptation has been reported for various treatment sites including lung, 7 prostate, [8][9][10] and head and neck cancers. 11,12 Across all treatment sites, adaptation is necessary due to tumor shrinkage, weight loss or other significant anatomical changes that impact the dose distribution (e.g., lung collapse or re-inflation). Specifically for head and neck cancers, large volume changes are common and often detected by external examination or through poor fitting of immobilization devices, but minor changes can go unnoticed. However, relatively minor anatomy changes may still have a significant effect on the dose distribution and are more difficult to discern by visual inspection of anatomy alone.
More precise and conformal radiation treatments available with modern techniques may need more plan adaptations to provide consistent target dose coverage and healthy tissue sparing with a changing anatomy. For making a decision on the necessity of plan adaptation in clinical practice, efficient daily evaluation of the delivered dose distribution on the modified anatomy is required. Different methods have been presented on detecting volume changes 13 and landmark movements, 14 but most rely solely on visual inspection by clinicians. These visual inspections may not be consistent as shown by inter-observer studies. 15 Several groups have presented adaptation strategies and schedules throughout treatment. 16,17 A recent study using the same dataset as in this study has produced a method of detecting anatomical differences to flag consideration of plan re-evaluation without considering the dose distribution. 18 Currently, cone beam CT (CBCT) imaging is routinely used for patient alignment and anatomy monitoring, but can also be used for dosimetric assessment of actual radiation delivery. Dose calculations on CBCTs are possible with the results varying between reported studies 19-21 because of inferior image quality and tissue densitometry. Performing reliable analysis of the dose to the target and organs at risk would require contouring of relevant structures on the daily CBCT image. An attractive alternative is to employ deformable image registration (DIR) to transfer contour information from the planning CT study for analysis. DIR has been shown to produce a variety of results depending on the algorithm used, original contouring accuracy and imaging modalities (i.e., CT simulation, MRI or CBCT).
Unfortunately, registration between different imaging modalities has been shown to have worse accuracy 22 especially for CBCT images due to limited image quality and artifacts.
There are two primary effects of anatomical deformations on a radiation treatment: 1) movement of voxels and regions on interest (ROI) relative to the planned dose distribution and 2) change in the dose distribution itself due to re-arrangement of voxels or density changes therein. The current gold standard (GS) for determining whether to adapt a treatment plan involves a new CT simulation (ReCT), dose calculation and DIR to map contours from the PCT. This procedure is time-consuming and expensive but accounts for both effects of anatomical deformation and is applied when gross anatomical changes are suspected.
The best alternative without a new CT simulation involves using DIR to warp the planning CT to match the daily anatomy from the CBCT and perform dose calculation as proposed by Veiga et al. 23 and accounts for both effects of anatomical deformations. However, the dose recalculation practically can be difficult and time-consuming. It is usually performed off-line which limits its routine daily use at the treatment unit. What if you could determine the necessity of plan adaptation without a new CT scan and dose calculation? Without the re-computation of the dose, only the movement of voxels and ROI's relative to the planned dose distribution are considered, but not the change to the dose distribution. The dose distribution is assumed to be robust and only mildly affected by the re-arrangement of the voxels. In this study, we explore the results of using the CBCT without a dose calculation and a CBCT with a dose calculation and compare both to the current gold standard. The goal is to see if assessing the movement of ROI relative to the planned dose distribution provides enough dose information to properly trigger the plan adaptation process, when compared to current clinical practice of visual inspection.

2.A | Patient studies
For this study, 18 patients who received multi-fractionated radiotherapy for head and neck cancer and had plan adaptation during treatment course were selected. Each patient had a CT scan taken before treatment (range 4-30 days) and used for planning (i.e., PCT), daily pre-treatment CBCT studies and another CT re-taken during treatment (ReCT) when anatomy changes were deemed significant (day "X"). were performed with software from MIM Maestro (version 6.5 MIM Software Inc., Cleveland, OH, USA) using the default DIR algorithm applying an intensity based free form algorithm, with a sum of squared differences similarity metric. 25 The mean registration error using MIM Maestro between two kVCT's was shown to be 1.7 mm by Kirby et al. 22 using a deformable Head and Neck phantom.

2.B | Dose distribution estimation
To determine the necessity of plan adaptation, an estimation of the dose distribution "of the day" was required and three estimation methods are presented and compared to the current gold standard which requires a re-planning CT. The first method (CBCT P ) used DIR to map the contours from the PCT to the daily CBCT with the planned dose distribution rigidly registered to the daily CBCT as shown in Fig. 1. The second method (CBCT R ) used the DIR to map the contours from the PCT to the daily CBCT with the recalculated dose (from the ReCT) rigidly registered to the daily CBCT. The third method (ReCT P ) used the DIR to map the contours from planning CT to the ReCT with the planned dose distribution rigidly registered to the ReCT. The gold standard method (ReCT R ) applied DIR to map contours from the PCT to ReCT with the recalculated dose on the ReCT. Both dose distributions (planned and recalculated) were obtained using the original treatment plan parameters and beam; the plan was not re-optimized. The rigid registration process used 6 degrees of freedom and simulated the alignment of the CBCT study to PCT (or ReCT) performed by the radiation therapists in the clinic before each fraction. In total, four separate methods estimated the daily dose distribution using the CBCT or ReCT as the secondary CT study, with the planned or recomputed dose. For clarity, each method was referred to by the secondary image used (CBCT or ReCT) and if the planned or ReCT dose was used, denoted by subscript P or R, respectively. All dose estimation methods are illustrated in Fig. 2, showing all four investigated combinations.

2.C | Voxel-to-voxel dose comparison
The clinically relevant comparison of the dose results obtained by different estimations requires evaluation on a voxel-to-voxel basis.
Every voxel in the PCT study can have a different dose value in fraction X (when ReCT was ordered), depending on the secondary CT study for image registration and the dose distribution. Comparison with any other method is done by calculating the relative dose difference to the GS (RD j ) for a specific structure j across each individual voxel i: between a test method (T) and GS averaged over all N j voxels within all 18 patients p.
Voxel-to-voxel analysis was performed for the right and left parotids because they were present in all image studies, incurred significant deformation and are frequently positioned close to the target volume. The analysis was also performed for the spinal cord because it is a clinically important structure. For the PTV, the dose was determined at each voxel using CBCT P and CBCT R methods. The threshold criterion for adaptation was for 95% of the volume (D95) to be below the prescribed dose.

2.D | Test for the necessity of plan adaptation
The D95 parameter was selected following recommendations for evaluating the target coverage. 26 Only the primary PTV was analyzed for each patient.

3.A | Voxel-wise dose comparison
The relative dose difference RD j given by Eq. (1) for each method are shown in Table 1 for the ipsilateral and contralateral parotids and spinal cord. The error caused by only the changed dose distribution is presented by the ReCT P row and the CBCT R row represents the error caused only by the DIR between different imaging modalities. CBCT P row represents the error when both effects were present.

| Test for necessity of adaptation
The parotid mean dose estimates using CBCT P and CBCT R are com-

| DISCUSSION
In a standard workflow, the only dose distribution always available is the one calculated using the initial CT simulation for planning purposes. Theoretically, the dose gradients from the planned dose distribution indicate what dose differences may occur due to specific anatomical changes. Dose gradients are mainly defined by the original beam geometry relative to the planned iso-center, which is not affected by deformation. Without extensive deformation, these gradients can be maintained and could predict dose change, when combined with a deformation field. However, with large volume or density reductions within the beams path significant changes to the F I G . 2. Schema describing the daily dose estimation using DIR from the planning CT to either the daily CBCT or re CT study (ReCT). Two different dose distributions computed on the PCT or ReCT are transferred to the moving image using a 6 degree of freedom rigid registration. The gold standard method is highlighted in yellow using the ReCT and recomputed dose. Day X is when ReCT was ordered due to observed significant anatomical changes. The average relative dose differences RD j for each organ presented in Table 1 show that for both parotids and spinal cord the RD j from the CBCT R method (which is a result of DIR error alone) is similar to the results from the ReCT P method, which is the error from using the planned instead of the recomputed dose. The CBCT P RD j includes both sources of error but is less than the sum of errors in ReCT P and CBCT R methods.
It has been shown that DIR error is specific to the algorithm used 27,28 and image quality. 22  F I G . 4. The difference between predicted D95 and the prescribed dose for the PTV for using a) CBCT P and b) CBCT R methods compared to ReCT R (gold standard). Values are presented as the difference from the prescribed dose. Dashed line represents conservative criteria (within 1 Gy of threshold). ReCT R is the DIR to ReCT using the recalculated dose. CBCT P is the DIR to daily CBCT using the planned dose. CBCT R is the DIR to daily CBCT using the recalculated dose.
18 of these cases was not necessary, with 11 of the original patient plans still within clinical tolerances. Clinical decisions of plan adaptation were made before the re-scan using personal experience, which explains the discrepancy in adaptation rates between our GS and that decided clinically. Our results have shown that both methods using daily CBCT studies (CBCT R and CBCT P ) yielded very conservative results and missed no required adaptations. If the simplest prediction method (CBCT P ) was used, only four patients would have been unnecessarily re-scanned and adapted. This demonstrates that using the DIR to the CBCT of the day without a dose calculation in CBCT P method can determine when to adapt a treatment plan better than that done clinically avoiding a number of unnecessary CT simulations and re-planning efforts. Performing an additional dose calculation in CBCT R caught two additional unnecessary plan adaptations at the cost of additional computation time, while without a dose computation the procedure can be completed within one minute allowing for an efficient "adapt or not" decision online.

| CONCLUSION
Improvements in IGRT and conformal radiation delivery have made adaptive radiation therapy a reality, but steps need to be taken to ensure its efficiency. Practical implementation requires an efficient method of daily evaluation and decision-making to determine when plan adaptation is truly necessary. The method of dose evaluation using on-board CBCT imaging alone is limited by the necessity for dose calculation, contouring and image registration. We have shown that the daily CBCT image mapped back to the planning CT without a dose calculation can provide sufficient information for the important decision of when to re-plan. The goal is to prevent the use of unnecessary additional CT simulations and dose computations with a quick online evaluation. Further research needs to be performed with more patients and other treatment sites including abdomen and thorax and for treatment techniques that will produce a different landscape of dose gradients.

ACKNOWLEDGMENT
The authors thank Dr. Bryan Schaly for supplying CT data and discussion.

CONF LICT OF I NTEREST
The authors declare no conflict of interest.