Comparison of CT images with average intensity projection, free breathing, and mid‐ventilation for dose calculation in lung cancer

Abstract The purpose of this study was to compare three computed tomography (CT) images under different conditions—average intensity projection (AIP), free breathing (FB), mid‐ventilation (MidV)—used for radiotherapy contouring and planning in lung cancer patients. Two image sets derived from four‐dimensional CT (4DCT) acquisition (AIP and MidV) and three‐dimensional CT with FB were generated and used to plan for 29 lung cancer patients. Organs at risk (OARs) were delineated for each image. AIP images were calculated with 3D conformal radiotherapy (3DCRT) and intensity‐modulated radiation therapy (IMRT). Planning with the same target coverage was applied to the FB and MidV image sets. Plans with small and large tumors were compared regarding OAR volumes, geometrical center differences in OARs, and dosimetric indices. A gamma index analysis was also performed to compare dose distributions. There were no significant differences (P > 0.05) in OAR volumes, the geometrical center differences, maximum and mean doses of the OARs between both tumor sizes. For 3DCRT, the gamma analysis results indicated an acceptable dose distribution agreement of 95% with 2%/2 mm criteria. Although, the gamma index results show distinct contrast of dose distribution outside the planning target volume (PTV) in IMRT, but within the PTV, it was acceptable. All three images could be used for OAR delineation and dose calculation in lung cancer. AIP image sets seemed to be suitable for dose calculation while patient movement between series acquisition of FB images should be considered when defining target volumes on 4DCT images.

underdosage to tumor and overdosage to normal organs, and significantly diverted the planned and the delivered doses. 4 Several approaches have been developed to manage the effects of respiratory motion during radiation therapy. 5,6 One of them is a motionencompassing method, which addresses the entire range of tumor motion and adds a margin to the target volumes. For dose calculation in this method, Vinogradskiy et al. 7 and Guckenberger et al. 8 compared three-dimensional (3D) and four-dimensional (4D) dose calculations and revealed minimal dosimetric differences in the gross tumor volume and the internal target volume (ITV). Thus, 3D dose calculation is still required in radiation therapy. For 3D radiation treatment planning in lung cancer, the ITV was delineated on maximum intensity projection (MIP) images, whereas organ at risk (OAR) contouring and static 3D dose calculation were done with 3DCT images. [9][10][11][12] 3DCT images were also used to define OARs and targets for radiation therapy treatment, but the use of static images of a moving organ remains problematic. 13 A common approach is to acquire helical computed tomography (CT) scans during free breathing (FB). Respiration-induced target motion during acquisition, however, can cause motion artifacts. [14][15][16] Several groups have demonstrated that dose calculation on average intensity projection (AIP) images, with each pixel holding the average value of all equivalent pixels in the data set, can yield results comparable to those attained with 4D dose calculation. [17][18][19][20] Others have proposed that mid-ventilation (MidV) CT images represent the tumor in its time-averaged position over the respiratory cycle and can be used for dose calculation. This position is an appropriate representation of the mean geometry and density of the target volumes for respiration-induced anatomical variations. [21][22][23][24] This study was designed to compare three types of CT imaging (AIP, FB, MidV) for contouring and radiation treatment planning for lung cancer using 3D conformal radiation therapy (3DCRT) and intensity-modulated radiation therapy (IMRT). Their differences are compared in terms of the OAR volume, geometrical center of the OARs, dosimetric indices, and dose distribution with regard to tumor size.

2.A | Patient selection
A total of 31 patients with various stages of lung cancer were enrolled in this study between January 2012 and December 2013.
Our institutional review board approved the study protocol. Two patients were excluded because of their movement between the FB and 4DCT acquisitions.

2.B | CT scanning for treatment planning
The patients were positioned in an immobilization device with a wing board on the scanner table, in the supine position with arms raised above the head (the treatment position). CT images of the thorax were acquired as normal FB and 4DCT scans using a helical 16-slice Brilliance Big Bore CT scanner (Philips Medical Systems, Cleveland, OH, USA). The parameters for image acquisition were 120 kVp, 400 mAs, 512 9 512 matrix, 3.0-mm slice thickness, and 0.5 s per rotation, with pitch values according to the manufacturer's recommendation for the particular respiratory rate of each patient. During CT image acquisition, the respiratory signal was recorded using either a Philips bellows system or a real-time position management respiratory gating system (Varian Medical System, Palo Alto, CA, USA) synchronized with the CT data. For the 4DCT images, respiratory signal data were reconstructed and sorted into 10 equidistant time-percentage bins (0% at maximum inhalation to 90%) throughout a respiratory cycle, each reflecting 10% of the respiratory cycle. Thus, the 0% respiratory phase corresponded to peak inhalation and the 50% respiratory phase corresponded to peak exhalation. MIP and AIP images were reconstructed from 10-phase 4DCT data relating to the percentage of time. MidV images (representing the tumor in its time-averaged position over the respiratory cycle) were chosen from the 10 phases of the 4DCT image data set. To make them appropriate for clinical use, the 4DCT images during the exhalation phase were displayed. We then selected the data set image that was closest to the central position of the tumor. All CT images (AIP, FB, MidV, MIP, 10-phase 4DCT) were then transferred to a radiation treatment planning system (Eclipse, version 10.0.42; Varian Medical System) with a DICOM protocol connection.

2.C | Delineation of the target and OARs
An experienced radiation oncologist generally delineated the planning target volume (PTV) and OARs are created by a geometric expansion based on the setup methods and the institutional guidelines. For each patient, the ITV was defined on the MIP images.
The 10-phase 4DCT was used to verify that the target motion was contained within the ITV during all phases. Each patient's PTV was obtained by adding 5 mm of circumferential expansion from the ITV and then copied to the AIP, FB, and MidV images. For these images, the OARs were delineated according to the guidelines provided by our respective protocols. The OARs used for plan comparisons in this study included the lung, trachea, heart, esophagus, and spinal cord, which are also shown on the images. The lung and trachea (including the main bronchus) were delineated with lung windows. Both lungs were automatically segmented using a threshold algorithm in the treatment planning system. The rest were con-

2.D | OAR geometrical center
To identify an OAR's geometrical center, the OAR's volume was first determined from a CT image. The geometrical center of the OAR was then calculated in the Eclipse treatment planning system. The centroid of the volume coordinates for each OAR was measured from the radiation beam isosenter. The coordination was displayed as x, y, and z, which refer, respectively, to displacement in the mediolateral (ML), anteroposterior (AP), and superoinferior (SI) directions.
The 3D vector was calculated for comparing the centroid of the OAR's volume in each image. The difference of the geometrical center of FB and MIP image sets comparison with respect to AIP image sets was calculated and analyzed.

2.F | Plan comparison and statistical analyses
The volume of each OAR and the centroid of each OAR's volume were measured and recorded. The impact of target volume size was studied with separation of small and large volume using volume cut-off at 150 cm 3

| RESULTS
In all, 31 lung cancer patients underwent FB and 4DCT simulations.
The patients' PTV information regarding target location, PTV volume, and target motion amplitude in the SI, ML, and AP directions as well as the prescribed dose and number of fractions were recorded. They are shown in Table 1.
According to the findings, tumors were found in the upper lobes To evaluate the dosimetric differences between the calculated plans from the three types of CT images, the AIP image plan was copied to the FB and MidV images and recalculated using the same beam arrangement and PTV coverage. Figure 1 represents the comparison between the three plans to illustrate the dose distributions for 3DCRT and IMRT.
The volumes for OAR contouring in the remaining 29 patients are shown in Table 2  Results are reported as the average AE SD.
Further study is needed to evaluate the accuracy of dose delivery to patients. As radiation treatment planning represents the actual dose distribution, it is necessary to verify dose distribution in a moving target volume compared with dose distribution in treatment planning.

ACKNOWLEDGMEN TS
The authors would like to thank Dr. Danupon Nantajit and Ms.
Sunattee Kessung for their assistance in writing the manuscript. We would also like to thank Ms. Kamonwan Soonklang for her helpful statistical analysis.