Technical Note: Comparison of the internal target volume (ITV) contours and dose calculations on 4DCT, average CBCT, and 4DCBCT imaging for lung stereotactic body radiation therapy (SBRT)

Abstract Purpose To investigate the differences between internal target volumes (ITVs) contoured on the simulation 4DCT and daily 4DCBCT images for lung cancer patients treated with stereotactic body radiotherapy (SBRT) and determine the dose delivered on 4D planning technique. Methods For nine patients, 4DCBCTs were acquired before each fraction to assess tumor motion. An ITV was contoured on each phase of the 4DCBCT and a union of the 10 ITVs was used to create a composite ITV. Another ITV was drawn on the average 3DCBCT (avgCBCT) to compare with current clinical practice. The Dice coefficient, Hausdorff distance, and center of mass (COM) were averaged over four fractions to compare the ITVs contoured on the 4DCT, avgCBCT, and 4DCBCT for each patient. Planning was done on the average CT, and using the online registration, plans were calculated on each phase of the 4DCBCT and on the avgCBCT. Plan dose calculations were tested by measuring ion chamber dose in the CIRS lung phantom. Results The Dice coefficients were similar for all three comparisons: avgCBCT‐to‐4DCBCT (0.7 ± 0.1), 4DCT‐to‐avgCBCT (0.7 ± 0.1), and 4DCT‐to‐4DCBCT (0.7 ± 0.1); while the mean COM differences were also comparable (2.6 ± 2.2mm, 2.3 ± 1.4mm, and 3.1 ± 1.1mm, respectively). The Hausdorff distances for the comparisons with 4DCBCT (8.2 ± 2.9mm and 8.1 ± 3.2mm) were larger than the comparison without (6.5 ± 2.5mm). The differences in ITV D95% between the treatment plan and avgCBCT calculations were 4.3 ± 3.0% and −0.5 ± 4.6%, between treatment plan and 4DCBCT plans, respectively, while the ITV V100% coverages were 99.0 ± 1.9% and 93.1 ± 8.0% for avgCBCT and 4DCBCT, respectively. Conclusion There is great potential for 4DCBCT to evaluate the extent of tumor motion before treatment, but image quality challenges the clinician to consistently delineate lung target volumes.


| INTRODUCTION
For early-stage lung cancer patients, stereotactic body radiotherapy (SBRT) has become one of the primary treatment options. 1 Improvements in patient localization using on-board imaging and cone-beam computed tomography (CBCT) and the ability to account for tumor motion have led to the increase in lung SBRT treatments. 2 One strategy for tumor motion management is the use of four-dimensional CT (4DCT) simulations to determine the total extent of motion for planning purposes. 3,4 Then a 3D CBCT is used for online patient setup and the patient is treated free breathing. This strategy, however, does not take into account potential variations in a patient's breathing pattern from simulation to treatment.
The four-dimensional CBCT is now becoming a clinically feasible tool in the treatment room. 5,6 The advantage is the ability to visualize the motion of the tumor at the time of treatment. While a phantom study suggested image guidance with 3DCBCT has similar accuracy to 4DCBCT image guidance, 7 a clinical study showed improvement in target localization with 4DCBCT as tumor motion increased. 6 Several investigators have looked at 3DCBCT for dose accumulation and adaptive planning of SBRT lung treatment. 8,9 Even so, the issue of calculating dose on a moving target exists. 10 11 such that at least 95% of the PTV is covered by the prescription dose. Key plan metrics evaluated include V100% (percent of target covered by the prescription dose) and D95% (dose covering 95% of the target).

| ME TH ODS
For each fraction, a free-breathing 4DCBCT was acquired. Varian Real-time Position Management (RPM) was used to track patient breathing. The RPM block was placed on the patient between the xiphoid and umbilicus for diaphragmatic displacement tracking. A total of 1800 projections were acquired over a full gantry rotation at a speed of 3 degrees per second. For patient localization a 3DCBCT (avgCBCT) was reconstructed (filtered back-projection) online using every projection. The same projections used for localization were binned into 10 phases based on the RPM breathing trace and reconstructed offline. A radiation oncologist contoured the target volume on the avgCBCT (avgCBCT ITV) and each 4DCBCT phase (Fig. 1).
The 4DCBCT ITV was formed by the union of the 10 contours. The Dice coefficient, Hausdorff distance, and center of mass difference between the 4DCT ITV, avgCBCT ITV, and 4DCBCT ITV were computed with in-house software. The Dice coefficient quantifies the overlap between two contours by the ratio of twice the intersection divided by the sum of the contour volumes. The Hausdorff distance is the greatest minimum distance between all vertices of the two contours.
Mean values for each patient (over four fractions) as well as mean of the means over nine patients (group systematic error) and standard deviation of means (SD of the systematic error) were calculated.
Using the online registration, the treatment plans were calculated on each phase of the 4DCBCT and the avgCBCT for each fraction of nine patients. The average treatment metrics over four fractions, D95%, and V100%, were compared with the coverages achieved in the original treatment plans. It was assumed that the length of each of the 10 phases was the same, thus each treatment field contributed 1/10 of its MU to each phase. A plan sum of all the phases yielded the single fraction dose delivered to the target.
Patient-specific calculations and measurements were acquired for each plan using a CIRS lung phantom. A 10-phase 4DCBCT of the phantom was acquired (Fig. 2), and patient plans were mapped to and calculate on each phase of the phantom's scan. The interphase average coefficient of variation (CV) was determined using all 76 treatment fields to quantify how uncertainty created from CBCT imaging noise affects the dose calculation accuracy. The interphase average CV was found by computing the standard deviation in calculated dose over the 10 phases of the 4DCBCT in each treatment field and averaging all treatment fields together. Point dose measurements were acquired with an IBA CC01 ion chamber at isocenter to test the calculation algorithm's precision for 4DCBCT based dose calculations. The ion chamber was calibrated in solid water against the treatment-planning system the day of the plan measurements.  Dice coefficients were equivalent between each comparison (0.7 ± 0.1 with p = 0.8). Similarly, the center of mass differences were consistent between the comparisons: 2.3 ± 1.2, 3.1 ± 1.6, and 2.6 ± 2.1 mm, respectively (p = 0.6). The Hausdorff distances showed no statistical differences (p = 0.4), however comparisons in section 2 and 3 (8.2 ± 2.9 mm and 8.1 ± 3.2 mm) were larger than section 1 (6.5 ± 2.5 mm).

| DISCUSSION
This study sought to quantify the capability to contour on 4DCBCTs and determine the dose delivered to patients receiving lung SBRT.
The 4DCBCT image quality affected the physician's ability to identify the tumor volume. Streak artifacts, as well as lesions adjacent to high intensity anatomy (blood vessels, the diaphragm, and the chest wall), reduced tumor contrast. Additionally, tumor size and composition may change over the course of treatment. 12 Contrast to noise ratio for CBCT has been shown to decrease linearly with faster gantry speed and fewer projections. 13  T A B L E 1 Comparison of 4DCT, average CBCT, and 4DCBCT ITV contours using the following evaluation metrics: Hausdorff distance, center of mass difference, and dice coefficient. dice coefficient mean SD mean SD mean SD mean SD mean SD mean SD mean SD mean SD mean SD   accuracy. This is backed up by the Table 1 comparison of average CBCT to 4DCBT. The COM and Dice coefficients for the two patients (7,9) are much worse than the other seven patients. If the avgCBCT is used for online patient setup, an appropriately sized ITV margin should be applied to encompass the diaphragmatic tumor motion uncertainty not captured in the average motion envelope. Average and standard deviation values were calculated over the four-fraction treatment course. The largest differences were found in patients 6 and 9. This discrepancy is due to the poor image quality in the AvgCBCT and 4DCBCT, caused by pleural effusion and poor chest wall-tumor contrast.

(a) (b)
F I G . 3. Single phase ITV (red) from patient 4 where tumor/chest wall border is indistinguishable due to significant streak artifacts in the 4DCBCT (A). Resulting 4DCBCT ITV (blue) expands posteriorly and laterally into the chest wall compared to the avgCBCT ITV (orange) and yields lower dose coverage reported to 4DCBCT ITV. This can be seen by the low Dice coefficient between avgCBCT and 4DCBCT in Table 1.
in sizably larger contours on the daily CBCTs, which did not match well with the planning CT. Compared to contouring island lesions (where the low-density background of the lung provides high con-

CONFLI CTS OF INTEREST
None.