Feasibility of CBCT‐based dose with a patient‐specific stepwise HU‐to‐density curve to determine time of replanning

Abstract Purpose (a) To investigate the accuracy of cone‐beam computed tomography (CBCT)–derived dose distributions relative to fanbeam–based simulation CT‐derived dose distributions; and (b) to study the feasibility of CBCT dosimetry for guiding the appropriateness of replanning. Methods and materials Image data corresponding to 40 patients (10 head and neck [HN], 10 lung, 10 pancreas, 10 pelvis) who underwent radiation therapy were randomly selected. Each patient had both intensity‐modulated radiation therapy and volumetric‐modulated arc therapy plans; these 80 plans were subsequently recomputed on the CBCT images using a patient‐specific stepwise curve (Hounsfield units‐to‐density). Planning target volumes (PTVs; D98%, D95%, D2%), mean dose, and V95% were compared between simulation‐CT–derived treatment plans and CBCT‐based plans. Gamma analyses were performed using criterion of 3%/3 mm for three dose zones (>90%, 70%~90%, and 30%~70% of maximum dose). CBCT‐derived doses were then used to evaluate the appropriateness of replanning decisions in 12 additional HN patients whose plans were previously revised during radiation therapy because of anatomic changes; replanning in these cases was guided by the conventional observed source‐to‐skin‐distance change‐derived approach. Results For all disease sites, the difference in PTV mean dose was 0.1% ± 1.1%, D2% was 0.7% ± 0.1%, D95% was 0.2% ± 1.1%, D98% was 0.2% ± 1.0%, and V95% was 0.3% ± 0.8%; For 3D dose comparison, 99.0% ± 1.9%, 97.6% ± 4.4%, and 95.3% ± 6.0% of points passed the 3%/3 mm criterion of gamma analysis in high‐, medium‐, and low‐dose zones, respectively. The CBCT images achieved comparable dose distributions. In the 12 previously replanned 12 HN patients, CBCT‐based dose predicted well changes in PTV D2% (Pearson linear correlation coefficient = 0.93; P < 0.001). If 3% of change is used as the replanning criteria, 7/12 patients could avoid replanning. Conclusions CBCT‐based dose calculations produced accuracy comparable to that of simulation CT. CBCT‐based dosimetry can guide the decision to replan during the course of treatment.


| INTRODUCTION
The need for adaptive radiotherapy has been demonstrated by many investigators. [1][2][3] New plans are adapted throughout the weeks-long course of fractionated radiotherapy to account for patient geometry changes resulting from weight loss, organ deformation, tumor shrinkage, and other causes. New adaptive plans may also be needed if the immobilization device needs to be adjusted or remade for variety of reasons. For some patients receiving intensity-modulated radiotherapy (IMRT) or volumetric-modulated arc therapy (VMAT), the significant benefit of replanning has been demonstrated. 4 The frequency of replanning in patients with head and neck cancer was reported to be 32%-70%, depending on criteria. 5 It is challenging, however, to decide on the appropriate time for replanning. Several investigators have looked for indicators to predict substantial dosimetric change. Although correlations between several parameters (such as weight loss, skin separation, and others) and dose change to target or organ at risk (OAR) were observed, [4][5][6] no single parameter can be reliably used to decide the time of replanning for patients with head and neck cancer. 4 Therefore, decisions on replanning are frequently based on the practical experience of clinicians.
The main challenge in initiating the replanning process is a lack of tools for estimation of dosimetric changes for targets and OARs.
Onboard kV cone-beam CT (CBCT) is now widely available, and CBCT-based dose calculation makes it possible to evaluate dosimetric change during the course of treatment. Although kV CBCT technology is mainly used to set up patients and localize anatomy, its potential for use in dose calculation has been recognized and reported. [7][8][9][10] Dose calculation accuracy using CBCT images has been evaluated by investigators. 8,9,[11][12][13] The main source of dosimetric error stemming from CBCT-based dose calculations (relative to fan-beam-based CT simulation) comes from the uncertainty in Hounsfield unit (HU)-toelectron density conversion of CBCT images. As a more direct approach, phantom-based calibration of the HU-to-electron density curve was investigated. [7][8][9]13 Unlike fan-beam CT, kV CBCT suffers from scatter, which results in greater HU uncertainty. 13 A recently developed treatment planning system RayStation V5.0 (RaySearch Laboratories; Stockholm, Sweden) provides CBCT-based dose calculation using a patient-specific stepwise HU-to-density curve (i.e., patient CBCT HUs were converted to only six classes of materials: air, lung, adipose, tissue, cartilage/bone, and other high-density material). The method is similar to density override in its assignment of six classes of materials. Most modern treatment planning systems provide a density override function, so that this method could be used widely in clinical practice. The purpose of this study was to (a) investigate the accuracy of CBCT-based dose calculations in the RayStation treatment planning system, and (b) study the feasibility of using CBCT-based dose to select the appropriate treatment replanning time. In this study, dose calculation accuracy was assessed using 80 IMRT/VMAT plans for four anatomic sites: head and neck (HN), lung, pancreas, and prostate. The appropriateness of replanning decisions was evaluated with data from 12 rescanned patients with head and neck cancer.

2.A | Patient data
Image data from 40 patients who underwent radiation therapy at our institution were randomly selected for this institutional review board approved retrospective study; patients with large geometry change (external body contour change >1 cm between planning CT and CBCT) were excluded. All patients received step-and-shoot IMRT or VMAT treatments in our clinic for four anatomic sites: HN, lung, pancreas, and prostate. If the patient received IMRT (or VMAT) treatment, a complementary VMAT (or IMRT) plan was retrospectively made, with dose distributions comparable to the original clinical plan. In this way, a total of 80 plans were included in this study.

2.B | Patient-specific stepwise HU-to-density curve
Unlike the CT-based planning that uses only a single CT-to-electron density calibration curve in the treatment planning system, a patient-specific stepwise HU-to-density curve was created for each patient, assigning each voxel of CBCT images to one of the following categories of material (mass density): air (0.00121 g/cm 3 ), lung (0.26 g/ cm 3 ), adipose (0.95 g/cm 3 ), tissue (1.05 g/cm 3 ), cartilage/bone (1.6 g/cm 3 ), and other (3 g/cm 3 ). The treatment planning system is RayStation V5.0 (RaySearch Laboratories; Stockholm, Sweden), which provides the tool to adjust the HU threshold for each material via best match with the known material type (Fig. 1). The HU threshold was adjusted for each patient. The optimal thresholds were attained by identifying the range of HUs for each material category based on the CBCT. The corresponding mass densities were used in dose calculation.

2.C | CBCT-based dose calculation accuracy
For each patient, CBCT scans were transferred to the RayStation treatment planning system and then registered to the planning CT based on the bony anatomy. Contours, such as planning target volume (PTV) and OARs, were copied from the planning CT to the CBCT image via rigid registration. The dose was recalculated using the original dose calculation algorithm, but using the CBCT image for both IMRT and VMAT plans. The dose calculated based on CBCT was compared to that based on planning CT (Fig. 2). Differences between the two plans were documented for the following dose-volume variables: dose received by 98% of the PTV (D98%, near-minimum dose), PTV D95%, PTV D2% (near-maximum dose), PTV mean dose, and PTV volume receiving ≥95% of prescription dose (V95%). Gamma analysis was performed using a criterion of 3%/3 mm for three dose zones: the zone receiving ≥90% of maximum dose (high-dose region), the zone receiving 70%-90% of maximum dose (medium-dose region), and the zone receiving 20%-70% of maximum dose (low-dose region). These three zones represented three types of regions of interest: PTV, OARs adjacent to PTV, and OARs/normal tissue at some distance from the PTV and receiving low dose.

3.A | CBCT-based dose calculation accuracy
Dose differences between planning CT-based plans and CBCT-based plans for PTV are summarized in Table 1

3.B | Feasibility for evaluating dosimetry
The International Commission on Radiation Units & Measurements (ICRU) Report 83 22 recommends that 85% of target points should meet the criteria of absorbed dose difference within 5% if points are F I G . 1. Example showing mass density on CBCT images (left) and HU thresholds to define material type (right). Mass density was assigned to each voxel via mapping CBCT HUs to six classes of materials (patient-specific HU-to-density table; right). HU threshold to define different materials can be adjusted via best match with known tissue on CBCT. Black = air; pink = adipose; light blue = tissue; gold = cartilage/bone (lung and other material not shown). located at low-gradient areas (dose change <20%/cm) or that distance-to-agreement should be within 5 mm if points are located at high-gradient areas (dose change >20%/cm). CBCT-based dose accuracy was determined to be above the ICRU recommendation; therefore, the PTV dose-volume parameters (mean dose, D2%, D95%, D98%, and V95%) were used to decide the replanning time.
Gamma analysis using the criterion of 3%/3 mm can serve the same purpose.

4.C | Limitation of the study
Although 40 patients were selected to minimize geometric differences between planning CT and CBCT, small differences may still exist. The dosimetry difference between the CBCT plan and planning CT mainly results from the patient-specific stepwise HU-to-density F I G . 3. Change in PTV D2% relative to the prescription dose based on CBCT vs change based on CT2 for 12 patients with head and neck cancer who were rescanned because of weight loss.
curve, but geometric change, leading to dosimetric differences, cannot be excluded. However, without geometric change, dosimetric agreement between the two dose calculations would be expected to improve in this study.
This work investigates the feasibility of performing accurate dose calculations on CBCT images. We consider this calculation a necessary step toward implementing adaptive planning in our clinic. We do not address differences in anatomy observed between planning CT and CBCT as this involves clinical decision making. Actual implementation of CBCT-based replanning into the routine clinical workflow is beyond the scope of this paper.

| CONCLUSIONS
CBCT-based dose calculations produced accuracy comparable to that of simulation CT. CBCT-based dosimetry can guide the decision to replan during the course of treatment.