“Dose of the day” based on cone beam computed tomography and deformable image registration for lung cancer radiotherapy

Abstract Purpose Adaptive radiotherapy (ART) has potential to reduce toxicity and facilitate safe dose escalation. Dose calculations with the planning CT deformed to cone beam CT (CBCT) have shown promise for estimating the “dose of the day”. The purpose of this study is to investigate the “dose of the day” calculation accuracy based on CBCT and deformable image registration (DIR) for lung cancer radiotherapy. Methods A total of 12 lung cancer patients were identified, for which daily CBCT imaging was performed for treatment positioning. A re‐planning CT (rCT) was acquired after 20 Gy for all patients. A virtual CT (vCT) was created by deforming initial planning CT (pCT) to the simulated CBCT that was generated from deforming CBCT to rCT acquired on the same day. Treatment beams from the initial plan were copied to the vCT and rCT for dose calculation. Dosimetric agreement between vCT‐based and rCT‐based accumulated doses was evaluated using the Bland‐Altman analysis. Results Mean differences in dose‐volume metrics between vCT and rCT were smaller than 1.5%, and most discrepancies fell within the range of ± 5% for the target volume, lung, esophagus, and heart. For spinal cord Dmax, a large mean difference of −5.55% was observed, which was largely attributed to very limited CBCT image quality (e.g., truncation artifacts). Conclusion This study demonstrated a reasonable agreement in dose‐volume metrics between dose accumulation based on vCT and rCT, with the exception for cases with poor CBCT image quality. These findings suggest potential utility of vCT for providing a reasonable estimate of the “dose of the day”, and thus facilitating the process of ART for lung cancer.


| INTRODUCTION
Radiotherapy is a widely used treatment option for unresectable or inoperable non-small cell lung cancer (NSCLC) patients.
Although significant progress has been made in radiotherapy for lung cancer in recent decades, improving clinical outcomes for NSCLC is still challenging. 1 Dose escalation is one of the potential strategies to improve outcomes, but it may increase normal tissues toxicities, 2,3 for example, pneumonitis, pulmonary fibrosis, and cardiac injury, among others. A phase III randomized trial failed to demonstrate a survival benefit to dose escalation, 1 with speculation that normal tissue toxicity may have increased deaths with high-dose RT. 4 It has been proposed that as tumor shrinks during the course of treatment, adaptive radiotherapy (ART) may be beneficial in reducing normal tissue toxicities and may allow safe dose escalation. 5 While ART is often done with repeat planning CT (rCT), utilizing cone beam computed tomography (CBCT) for estimating the "dose of the day" has been an attractive research topic since they are readily available along the treatment course. However, direct use of CBCT in dose calculation is limited by inferior image quality and thus inaccurate Hounsfield units (HU). 6,7 Methods to mitigate or resolve this issue include image correction [8][9][10][11] and image deformation. [12][13][14][15][16][17][18][19][20][21] The former has shown promises, yet dose calculation inaccuracy still varies by up to 5% in phantom 7 and patient studies of various sites. [22][23][24] The latter approach, which creates deformed CT images from the initial planning CT (pCT) to CBCT, has potential to maintain anatomical information from CBCT while mapping accurate HU information from the pCT. Previous studies have shown promising results for estimating the "dose of the day" based on deformed CT images. 15,[17][18][19]25 Marchant et al. 15 found that less than 0.5% mean dose errors can be achieved with the deformed CT images for lung cancer patients, when compared with pCT. Veiga et al. 18 also demonstrated that the dose difference between deformed CT and rCT images for head and neck patients treated with intensity modulated radiotherapy (IMRT) was generally less than 2%. Same group also reported 3.4 ± 2.7 mm and 12 ± 12% average difference between deformed CT and rCT for lung cancer patients receiving passive scattering proton therapy. 17 Cole et al. 21 also found that the dose distribution based on deformed CT matched closely the rCT dose distribution in lung cancer patients receiving conformal external beam radiotherapy. An open source deformation algorithm was used in this study, which may not be practical for clinical adoption.
Adapting from the literature, 18,19,21 our present study aimed to explore dosimetric accuracy of the "dose of the day" calculated on virtual CTs for lung patients receiving nine-field IMRT treatment using commercially available deformation image registration (DIR) algorithm from a treatment planning system. A virtual CT (vCT) was created by deforming the initial planning CT (pCT) to the simulated CBCT that was generated from deforming CBCT to the repeat planning CT (rCT) acquired on the same day. And the accuracy of the "dose of the day" calculation based on vCT images was evaluated by comparing the accumulated dose based on vCT to that of rCT acquired on the same day using Bland-Altman analysis with dosimetric parameters.

2.A | Patients and imaging
This study investigated twelve patients with stage III NSCLC receiving nine-field IMRT with a prescription dose of 60 Gy delivered in

2.B | Virtual CT with deformable image registration
The Demons DIR algorithm of the Pinnacle 3 treatment planning system (research version 9.7, Philips Radiation Oncology Systems, Fitchburg, WI) was used in this study. This algorithm performs image deformation through matching image intensity with the assumption that pixels representing the same anatomical point on each image have the same image intensity values. 26 The rCT and its corresponding CBCT were acquired on the same day, therefore, should have similar external contour and internal anatomy. Nevertheless, they may still present small organ deformation and volume variation, due to respiratory motion, positioning deviations, etc. In order to minimize such differences, a simulated CBCT image was first created by deforming CBCT to rCT (The workflow is shown in Fig. 1). Then the pCT was deformed to the simulated CBCT to create vCT images.

2.C | Treatment planning
The initial treatment plan based on pCT was created on the Pinnacle The PTV was obtained by adding 5 mm margin to CTV in all directions. Organs at risk (OAR), including bilateral lungs, spinal cord, esophagus, and heart were also delineated.

2.D | "Dose of the day" calculations based on rCT and vCT
The initial treatment plan was performed with coplanar or non-coplanar 6 MV photon beams on the pCT image sets. The treatment beams from the initial plan were copied and applied to the isocenters of the rCT and vCT for subsequent dosimetric study (denoted as rCT dose and vCT dose, respectively) using the collapsed cone algorithm in Pinnacle with a 3 × 3 × 3 mm 3 dose grid (Fig. 1). Note that all beam parameters including isocenter, control points, and monitor units were kept identical to the initial treatment plan in this dose recalculation step.

2.E | Dose accumulation
The vCT-based accumulated dose was also performed with the Demons DIR algorithm to warp the vCT dose distributions as well as vCT to pCT image sets. It was presumed that the pCT plan was delivered in the first 10 fractions, followed by delivery of the vCT plan or rCT for the rest 20 fractions, therefore, accumulated dose was achieved by summing 10 times of pCT dose with 20 times of vCT or rCT dose. And rCT-based accumulated dose performed by the same approach was used as reference to access the accuracy of vCT-based accumulated dose. A set of DVH metrics was evaluated in this study, including PTV-D95% (minimum dose delivered to 95% of the PTV), lung-V20Gy (volume receiving at least 20Gy to the lung), heart-V45Gy (volume receiving at least 45Gy to the heart), the maximum dose (D max ) to the spinal cord, as well as the mean doses (D mean ) to PTV, lung-CTV and heart.

2.F | Statistical analysis
Bland-Altman analysis was used for analyzing dosimetric difference between vCT and rCT based accumulated dose for all DVH metrics.
The mean differences ± 1.96 times the standard deviation (SD) were defined as the limit of agreement (LOA). Statistical analyses were performed using GraphPad Prism software (v5.0, GraphPad Software, LaJoIIA, USA).  It should be noted that the poor case shown in Fig. 2 also corresponds to the red arrows in Fig. 4, which show large differences with −11.13% [ Fig. 4(c)] and −4.67% [ Fig. 4

(b)] in spinal cord D max
and lung-CTV D mean , respectively. Dose difference in esophagus D mean and heart D mean are −3.03% and −3.69%, respectively (as pointed out by red arrows in [Fig. 4(d)-(e)]. While the good case shown in Fig. 3 which indicated by the green arrows in Fig. 4 show less than 2% dose difference for head and neck patients, 18,19 and close match for lung cancer patients 21   The Pinnacle's Demons algorithm used in this study has been validated by other groups for CT-CBCT image deformation. [29][30][31] Visual inspection was used for validating DIRs in this study. We found from this study that the critical limiting factor to overall dose accumulation accuracy is CBCT image quality, which would subsequently affect DIR results' accuracy. As shown in the first example, Previous studies have shown that truncated artifacts, patient size, and imaging parameters can affect CBCT image quality. 32  image signal-to-noise ratio drops with increased phantom sizes.
Veiga et al. 19 also pointed out large body size for CBCT imaging may be the main source of error in recalculation proton dose on CBCT.

(f)]
can be observed for superior CBCT image (Fig. 3(D)). Dose difference of less than 2% can be observed for all evaluated metrics ( Fig. 4, green arrows), DVH and isodose distribution comparison [ Fig. 3(a)-(c)]. In such a retrospective setting, we are limited to previously acquired images. We aim to improve our clinical protocol in terms of CBCT scan setting in our future prospective studies.
Despite the promising results of the present study, there are inherent limitations. First, due to the limited scanning range and truncation in the CBCT, lung D mean between vCT and rCT dose was not evaluated without the whole lung contour. Second, considering the slow gantry motion while acquiring CBCT image, to the study may be improved if rCT was created as AveCT from 4D CT for the comparison with CBCT. Third, statistical analysis cannot be achieved in this study due to limited number of patients included. A follow-up study with large patient sample and standardized imaging parameters is warranted to further identify uncertainties in using CBCT and DIR for "dose of the day" calculation.

| CONCLUSION
The accuracy of "dose of the day" calculation based on vCT was evaluated by comparing vCT with rCT based accumulated dosimetric parameters. The vCT created in this study can be used to reasonably estimate the "dose of the day" calculation for lung cancer patients with mean difference smaller than 1.5% for most accumulated dosevolume metrics. The "dose of the day" also has the potential to become a very useful tool for ART. Critical to this calculation approach is CBCT image quality, which was found to be the main contributing factor to less ideal vCT creation, and thus less accurate dose accumulation.

ACKNOWLEDG MENTS
This study was supported in part by grants from the National Key

CONFLI CTS OF INTEREST
The authors declare no conflicts of interest.