Definition of internal target volumes based on planar X‐ray fluoroscopic images for lung and hepatic stereotactic body radiation therapy. Comparison to inhale/exhale CT technique

Abstract Purpose To compare tumor motion amplitudes measured with 2D fluoroscopic images (FI) and with an inhale/exhale CT (IECT) technique Materials and methods Tumor motion of 52 patients (39 lung patients and 13 liver patients) was obtained with both FI and IECT. For FI, tumor detection and tracking was performed by means of a software developed by the authors. Motion amplitude and, thus, internal target volume (ITV), were defined to cover the positions where the tumor spends 95% of the time. The algorithm was validated against two different respiratory motion phantoms. Motion amplitude in IECT was defined as the difference in the position of the centroid of the gross tumor volume in the image sets of both treatments. Results Important differences exist when defining ITVs with FI and IECT. Overall, differences larger than 5 mm were obtained for 49%, 31%, and 9.6% of the patients in Superior‐Inferior (SI), Anterior‐Posterior (AP), and Lateral (LAT) directions, respectively. For tumor location, larger differences were found for tumors in the liver (73.6% SI, 27.3% AP, and 6.7% in LAT had differences larger than 5 mm), while tumors in the upper lobe benefitted less using FI (differences larger than 5 mm were only present in 27.6% (SI), 36.7% (AP), and 0% (LAT) of the patients). Conclusions Use of FI with the linac built‐in CBCT system is feasible for ITV definition. Large differences between motion amplitudes detected with FI and IECT methods were found. The method presented in this work based on FI could represent an improvement in ITV definition compared to the method based on IECT due to FI permits tumor motion acquisition in a more realistic situation than IECT.

in some CT scanners [4][5][6] ; planar fluoroscopy images can be produced with a digital flat panel forming part of an X-ray system or with a CBCT 7 ; and 4DCT and 4DCBCT are actually considered the state of the art in the detection of tumor movement by providing a complete set of images of the tumor throughout the breathing cycle.
Due to some drawbacks, the use of IECT may not be suited to measure tumor movement. On one side, the patient is not imaged in a normal breathing situation, which means that the inhalation and exhalation images are not representative of the actual movement of the tumor. On the other hand, it is dependent on the capacity of the patient to follow instructions correctly. Finally, it lacks information about the tumor's itinerary between the inhalation and exhalation phases. These limitations could be partially overcome with the slowscan procedure in which each slice is reconstructed over many respiratory cycles. This procedure allows to obtain an "average" image of the tumor. Nevertheless, it is not always easy to find the edges of the volume that encompass all tumor positions during a respiratory cycle due to the blurring associated with the tumor motion.
With 4DCT it is possible to define more realistic ITVs. In this case, the tumor is scanned along its trajectory throughout the breathing cycle by synchronizing the CT scanner to the patient's breathing with a special device capable of measuring the breathing phase. Then, the projections obtained can be combined in different datasets depending on the phase or the breathing amplitude. 8,9 4DCT also allows to obtain the Maximum Intensity Projection (MIP) 10,11 that shows all tumor positions in only one image. 4DCT technique presents some shortcomings when compared to other methods like FI 12 or 4DMRI. [13][14][15] For example, 4DCT cannot detect inter-and intrafractional variations in the breathing pattern, which produces an insufficient representation of tumor movement. In addition, its high cost makes few departments have installed this type of equipment and its implementation is still limited even in developed countries. 16 Thus, it would still be of interest to have some alternative methods to increase the accuracy in the measurement of tumor trajectories avoiding the acquisition of new and costly hardware.
In this work, we present a method based on planar fluoroscopic x-ray images (FI) that permits realistic ITV using only a standard CBCT system. The results obtained are compared with those of IECT. Although many studies investigated the tumor movement with different devices and methods, [17][18][19][20][21][22][23] as far as we know, there are no previous works comparing IECT with other methods for the definition of ITV. In addition, FI may be useful even when a 4DCT system is available since FI allows studying multiple breathing cycles and obtaining curves of position vs. time. Therefore, the shape of the breathing movement can be studied and decide the margins to be applied accordingly.

2.A | Patients
A total of 52 patients were imaged with both FI and IECT. Treatment localizations were lung (39 patients, 12 in the upper lobe, 11 in the medial lobe, and 16 in the lower lobe) and liver (13 patients

2.B | Acquisition of planning CT scan
Each patient undergone three CT scans with a Philips Brilliance (Eindhoven, The Netherlands) system. One of them was a free breathing scan with 3 mm slice width for planning purposes; the other two were exhale and inhale scans where the patient was asked to stay in an exhale or inhale state during the image acquisition.
These scans were focused only on the tumor volume with slice widths of 1.5 mm. Since the coordinate system is common for the three datasets, no registration was necessary. The tumor was contoured on each of the scans, and then transferred to the primary CT, where the ITV was defined as the sum of the contours of the tumor delineated on each of the three scans.

2.C | Acquisition of fluoroscopic images
FI were acquired with the XVI CBCT system (Elekta, Crawley, UK) of the treatment unit. The patients were placed in the unit with the system isocenter on the tumor position and identical setup to that of treatment. By placing the tumor at the isocenter is possible to measure displacements without scaling because the mm-to-pixel ratio (0.52 mm/pixel) is known.
Two projections, anteroposterior (AP), and lateral (LAT) were acquired. From the AP projection we could obtain information of Superior-Inferior (SI) and LAT motion, while the LAT projection provides information of AP and SI motions.
We acquire a set of fluoroscopic images composed by 150 frames taken each 180 ms (total length of 27 s) that allowed the gathering of many breathing cycles.  shape that the user must select around the tumor in the frame (AP or LAT) where it is best visualized. This ROI is considered as the reference ROI. The ROI size is therefore dependent on the tumor size of each patient. The reference ROI is selected around the markers if fiducial marks are used and it is up to the user to select all fiducial markers or only a part of them. Next, the software performs a matching procedure to locate the tumor in the rest of the frames which is based on the calculation of the Normalized Cross-Correlation (NCC) index. NCC is a widely used standard tool designed to detect features or similarities in intensity between two images of the same kind. [24][25][26][27] The NCC is calculated according to the expression 20 :

2.D | Detection of tumor motion
where: f(x, y) is the value of the pixel intensity at the (x, y) coordinates of a frame.
(u, v) represents the displacement of the reference ROI in the x and y directions, respectively. t(x, y) is the reference ROI and t is its mean value.

2.E | Margin definition
In order to define an ITV with FI, we have considered that tumor motion curves were obtained in a reference system different from together to obtain a single dataset and a single position histogram associated with SI motion. Margins for SI directions were calculated from this histogram. Thus, we might expect margins in SI direction to be highly influenced by the projection in which detected SI motion is larger.
It is worth noting that this method implies applying asymmetric margins to the gross tumor volume (GTV) in the CT due to the breathing motion might not be symmetrical with respect to the mean position.

2.F | Test with breathing simulator phantoms
Accuracy of the employed algorithm was tested by measuring in FI the motion of two respiratory motion phantoms: Quasar phantom (Modus QA, USA) and a Synchrony® phantom (Accuray, Sunnyvale, The Quasar phantom is designed as a motion 3.B | Test with breathing simulator phantoms  | 59 The differences between the nominal and measured amplitudes for the Quasar phantom (Fig. 1a) come mainly from a loss of data in the inhale phase, where the tumor moves at a higher speed. We investigated the influence of the sampling frequency of fluoroscopic images in the underestimation of motion amplitude. The motion curve used by the software to control the tumor movement samples the tumor position each 10 ms. We randomly resampled the motion each 180 ms in a curve with an amplitude of 10 mm and a period of 3 s. We found that the maximum difference in amplitude was 0.03 mm. (AP), and 6.7% (LAT) of the patients.

3.C | Comparison between ITV margins obtained from fluoroscopic images and IECT
Bar plots of the differences between amplitudes obtained with IECT and FI for each treatment location are shown in Fig. 2. As can be seen, for lung tumors we found important differences in all directions except for the upper lobe. These differences were negatives (smaller amplitudes in FI than in IECT) in all directions except in SI direction. For lung, the proportion of patients having differences of more than +5 mm and less than -5 mm was similar (17.9% and 20.5%, respectively). For liver tumors, the percentage of differences larger than +5 mm (13%) in SI was smaller than those with negative sign (60%).
The quotient between PTV volumes obtained with FI and IECT is shown in Figure 3. As can be seen, for most patients, the PTV values obtained with FI images were higher than the IECT volumes despite the lower mean values of the movement amplitudes measured in FI.
The mean value of the quotient of PTV volumes was 1.14 which implies an increasing around 15% in PTV volumes defined with FI. reason for this underestimation. However, the amplitude underestimation has a weak impact when breathing curves from patient are considered due to their high degree in variability.
The workflow presented in this work for ITV definition was successfully included in our clinical practice. As explained in the Results These differences also justify the distributions shown in Fig. 2.
There were a large proportion of patients with differences between FI and IECT larger than 5 mm with a maximum percentage of a 73% for liver tumors in the SI direction. On the contrary, we found that these differences are smaller for patients treated in the upper lobe, even though we still find differences larger than 5 mm in the AP direction in 36% of patients.
We would have expected to find smaller amplitudes when mea-  22 Mampuya et al. 23 reports longitudinal mean amplitudes of 20 mm for free breathing and 12.4 mm when applying abdominal compression. This last result is greater than that found in this work (mean value of 8.5 mm) in F I G . 3. Bar plot of the values for the quotient of PTV volume obtained from fluoroscopic images (FI) and that obtained with inhale/exhale CT (IECT) for the sample of patients. The mean value of PTV volume with FI is 1.14 times higher than the one from IECT.
which only tumors with motion amplitudes greater than 8 mm were considered.
The comparisons with the studies from other authors suggest that an inefficient abdominal compression can explain the larger values obtained in our study.
The fact that differences found in SI motion from LAT and AP projections were the same for lung and liver suggests that our method succeeded to detect the same tumor structures in both projections for lung patients. Fiducial markers used in liver patients avoid confusion in tumor detection when they are visible. Also, the value of the standard deviation of the differences between projections obtained in this work (2.2 mm) is very similar to that of 1.5 mm reported by Suh et al. 29 for Cyberknife patients. Thus, these differences could be explained by intrafraction variations in tumor motion and not by inaccuracies in the tracking algorithm.

| CONCLUSIONS
The

ACKNOWLEDGMENTS
The authors would like to thank C. Vallejo, M. Martin, and C. de la Pinta for their help in GTV definition and J. D García and R. Colmenares for their support in data collection.

CONF LICT OF I NTERESTS
The author have no other relevant conflict of interests to disclose.