Estimation of patient-size dependent imaging dose for stereoscopic/monoscopic real-time kV image guidance in lung and prostate SBRT

Purpose. The purpose of this work is to quantify the dependence of patient-specific imaging dose on patient-size from ExacTrac stereoscopic/monoscopic real-time tumor monitoring during lung and prostate stereotactic body radiotherapy (SBRT). Approach. Thirty lung and 30 prostate SBRT patients that were treated with volumetric modulated arc therapy (VMAT) were selected and divided into three patient size categories. Imaging doses from all SBRT fractions were calculated retrospectively assuming patients went through real-time tumor monitoring during their actual VMAT treatment times. Treatment times were divided into periods of stereoscopic and monoscopic real-time imaging depending on the imaging view with linac gantry blockage. The computed tomography (CT) images and contours of the planning target volume (PTV) and organs at risk (OARs) were exported from the treatment planning system. Based on the CT data, patient-specific 3D imaging dose distributions were calculated in a validated Monte Carlo model using DOSEXYZnrc. Vendor-recommended imaging protocols (lung: 120–140 kV, 16–25 mAs; prostate: 110–130 kV, 25 mAs) were used for each patient size category. Patient-specific imaging doses received by PTV and OARs were evaluated using dose volume histograms, dose delivered to 50% of organ volume (D50), and 2% of organ volume (D2). Results. Bone and skin received the highest imaging dose. For the lung patients, the highest D2 for bone and skin were 4.30% and 1.98% of the prescription dose respectively. For prostate patients, the highest D2 were 2.53% and 1.35% of the prescription for bone and skin. Additional imaging dose to PTV as a percentage of the prescribed dose was at most 2.42% for lung and 0.29% for prostate patients. T-test results showed statistically significant difference in D2 and D50 between at least two patient size categories for PTVs and all the OARs. Larger patients received more skin dose in both lung and prostate patients. For the internal OARs, larger patients received more dose in lung treatment while the trend was opposite in prostate treatment. Conclusion. Patient-specific imaging dose was quantified for monoscopic/stereoscopic real-time kV image guidance in lung and prostate patients with respect to patient size. Additional skin dose was 1.98% (in lung patients) and 1.35% (in prostate patients) of the prescription which is within 5% recommended value by the AAPM Task Group 180. For internal OARs, larger patients received more dose in lung patients while the trend was the opposite for prostate patients. Patient size was an important factor to determine additional imaging dose.


Introduction
Tumor motion results in considerable geometric uncertainty during external beam radiation therapy, (EBRT) This is certainly the case for those tumors located in the thoracic (Hanley et al 1999, or pelvic (Langen et al 2008, Su et al 2011 regions. The geometric uncertainty of tumors causes dosimetric uncertainty. Thus, large planning target volume (PTV) margins are required for prostate (Tanyi et al 2010) and lung tumors (Ekberg et al 1998), which can result in increased dose to nearby organs at risk (OARs). In order to ensure accurate dose delivery and to reduce the PTV margins which is essential for stereotactic body radiation therapy (SBRT) of lung and prostate, a motion management technique is utilized such as breath-hold (Hanley et al 1999), gating (Giraud et al 2011, Colvill et al 2014, tracking , Keall et al 2020, or motion encompassing technique based on internal target volume (ITV) concept (Mori et al 2007). All these techniques rely on imaging techniques during planning and/or delivery to guide the radiotherapy.
Image-guided radiation therapy (IGRT) is now a clinical standard for patient positioning for EBRT and is an essential tool to achieve geometric accuracy during SBRT of lung (Keall et al 2006, Caillet et al 2017 and prostate , Keall et al 2011. IGRT can significantly reduce PTV margins and optimize the delivery of the prescribed dose (Grills et al 2008, Pawlowski et al 2010. Various IGRT systems are used for patient setup immediately prior to a treatment fraction or to correct for real-time intra-fraction tumor motion (Bertholet et al 2019). Linac-mounted cone-beam computed tomography (CBCT) provides three-dimensional (3D) images and are commonly used for patient setup prior to treatment. Room-mounted stereoscopic systems, such as Cyberknife (Accuray, Inc. Sunnyvale, CA), SyncTracX (Shimadzu Co., Kyoto, Japan), and ExacTrac (Brainlab AG, Germany) infer 3D information from two projection x-ray images. These systems have the advantages of fast image acquisition, fine resolution, and low imaging dose (Ding et al 2018). By acquiring continuous stereoscopic projections with two views during treatment, real-time tumor monitoring can be achieved with Cyberknife (Hamamoto et al 2015, Jung et al 2015, Nakayama et al 2018 or SyncTracX (Shiinoki et al 2016) systems. Similarly, real-time tumor monitoring in ExacTrac system has been reported for prostate (Stevens et al 2016(Stevens et al , 2017 and lung  sites. Even though image guidance improves the accuracy of tumor localization and therapeutic radiation dose delivery, imaging dose cannot be neglected as it adds an appreciable dose to the patient especially if it is performed in real-time. The imaging dose may result in reaching the dose thresholds of skin or other OAR particularly if they are adjacent to PTV and already receiving undesirable therapeutic dose. Quantification of imaging dose was recommended by the AAPM TG-180 (Ding et al 2018). A number of studies have quantified OAR 3D imaging dose distribution for thorax and pelvis sites for various image guidance techniques such as CBCT (Spezi et (Shiinoki et al 2018). For the ExacTrac image guidance system, published data on imaging dose is scarce and limited to surface dose (Darvish-Molla et al 2020) and selective dose points in pelvis region (Stevens et al 2016) without quantification of 3D dose distribution for various OARs. To the best of authors' knowledge, no previous study has quantified patient-specific imaging doses from the ExacTrac imaging system. Moreover, for kV imaging, dose depends on patient size since the imaging protocol (kV and mAs) is patient size dependent. However, the effect of patient size on imaging dose was largely ignored in previous studies.
The purpose of this study was to quantify the patient-specific imaging dose received by PTV and OARs, using a validated gold standard Monte Carlo (MC) technique, over the course of lung and prostate SBRT with ExacTrac imaging system. Here we present data from 60 patients and quantify 3D dose distribution from realtime image guidance for various patient sizes.

Methods and materials
2.1. Patients Thirty lung and 30 prostate SBRT patients were selected from Nova Scotia Health Cancer Center clinical database following a Research Ethics Board (REB) approval. Both lung and prostate patients were grouped into three categories of large, medium, and small based on their Body Mass Index (BMI) with 10 patients in each group. Patients with BMI lower than 23 were categorized as small patients. The medium patients had BMI in the range [23-33] and the large patients had BMI more than 33 (Darvish-Molla 2019). We verified that patients' cross-sectional effective diameter (Boone et al 2011) at the slice corresponding to the center of PTV correlates strongly with patients' BMI (Pearson's correlation coefficient 0.92 and 0.86 for lung and prostate patients respectively). Clinical characteristics of the lung and prostate patients are summarized in tables 1 and 2 respectively. All the patients were treated with Volumetric Modulated Arc Therapy (VMAT) without real-time tumor monitoring. Imaging doses from all treatment fractions were calculated retrospectively assuming they went through real-time tumor monitoring using ExacTrac image guidance system version 6.2 during their actual VMAT treatment times.

Treatment planning
Patient's treatment plans were created using Eclipse treatment planning system version 15 (Varian Medical Systems, Palo Alto, CA). All treatment plans were generated by qualified dosimetrists and medical physicists following the institutional clinical protocols and approved by certified radiation oncologists. PTV margins were isotropic and were 4 mm and 5 mm for prostate and lung patients respectively. All patients received two-arc VMAT technique, one clockwise (CW) and another one counterclockwise (CCW). The range of the gantry angles for the arcs are given in

Real-time tumor monitoring
For actual patients positioning, pre-treatment initial CBCT and verification CBCT were required as per imaging protocol for the prostate patients. For lung patients, the imaging protocol was initial and verification ExacTrac images which were then verified by two pre-treatment initial and verification CBCTs. Real-time tumormonitoring was assumed using a clinical ExacTrac system through duration of VMAT treatment. The imaging frequency for the continuous stereoscopic or monoscopic imaging was 1.67 Hz. Vendor recommended image acquisition techniques (kVp and mAs) were used depending on patient sizes as per tables 1 and 2. During a realistic treatment fraction, one of the x-ray tubes would be periodically blocked by the rotating linac gantry (Stevens et al 2016(Stevens et al , 2017. Therefore, real-time imaging were divided in to stereoscopic and monoscopic periods based on the gantry angle (figure 1(a)) to calculate realistic imaging dose by assuming that the dose rate and gantry speed is constant over the VMAT arcs. Since the treatment arcs are different for patients with PTV in left lung (figure 1(b)), right lung (figure 1(c)), and prostate (figure 1(d)), the stereoscopic and monoscopic periods were calculated for each treatment arc. Table 3 shows the periods of stereoscopic and monoscopic imaging as a fraction of total treatment time.

Monte Carlo simulation
A MC model of the ExacTrac imaging system was utilized based on our previous experimentally validated MC model of the ExacTrac (Darvish-Molla et al 2020). Briefly, the geometry of ExacTrac imaging system was modeled using EGSnrc code (Rogers et al 1995, Kawrakow 2000. The x-ray tube consists of Varex RAD-21 model with 12°tungsten anode, and a 1.2 mm focal spot. The beam collimation is fixed to irradiate the (20.4 × 20.4 cm2) detector area which results in ∼(132 × 132 mm 2 ) field of view (FOV) at the isocenter 218.5 cm away from the source. BEAMnrs (Ma and Rogers 2023) was utilized to model the ExacTrac tube geometry and generate phase space files for different x-ray energies (Darvish-Molla et al 2020) which were then used in the current study. The tube voltages used were 110, 120, 130, and 140 kVp as per tables 1 and 2 for different patient sizes. Patient specific 3D imaging dose distributions were simulated with 10 10 histories for each energy using DOSEXYZnrc (Rogers et al 1995). The phase spaces files defined above were used as beam inputs to DOSEXYZnrc which calculates dose deposition of the simulated particles in a given volume. The patients' computed tomography (CT) volumetric images were exported from Eclipse treatment planning system as standard Digital Imaging and Communications in Medicine (DICOM) files ('DICOM,' 2023). Since DOSEXYZnrc only accepts material characteristics (material type and density) in the volumetric images in 'egsphant' format, the CT images in DICOM format were converted to egsphant format by developing an inhouse software in MATLAB (MathWorks, Natick, MA). The voxel sizes of the egsphant 3D image in x, y, z directions were kept the same as the original CT images (2.5 mm × 2.5 mm × 1.25 mm). However, the Matlab code removed the CT couch and replaced it with a Brainlab treatment couch before creating egsphant files. The Brainlab couch has a 2 mm carbon fiber shell with a density of 0.7 g cm −3 which is filled with a 4.6 cm foam core having a density of 0.1 g cm −3 (Mihaylov et al 2008, BrainLAB 2010. For each patient, three dose calculation simulations were performed as per figure 1: (i) stereoscopic imaging with both tubes, (ii) monoscopic imaging with right tube only, and (iii) monoscopic imaging with left tube only. The 3D dose output from the three simulations were combined later to calculate the total 3D dose distribution using an in-house MATLAB code (section 2.6) by considering the stereoscopic and monoscopic time fractions as per table 3. The output of the DOSEXYZnrc is 3D dose calculated volume in the unit of gray per history (Gy/Hist) which was saved as a '3d dose' file format.

Calibration of Monte Carlo simulation and imaging dose calculation
Calibration is required to convert the MC simulation 3D dose files from the units of Gy Hist −1 to Gy mAs −1 . Calibration was performed using a previously reported method . Briefly, the calibration factor for the MC dose unit conversion (figure 2) was calculated by directly comparing the MC results to experimental surface dose measurement using the AAPM TG-61 protocol (Ma 2000). The experimental surface dose (Gy mAs −1 ) for various kVp values were taken from our previously published data, i.e. figure 10 of (Darvish-Molla et al 2020). Identical TG-61experimental water phantom geometry was simulated in MC and the calibration factor (F cal ) of MC simulation for each tube potential was calculated using: where D TG61 is the experimental dose (μGy mAs −1 ) and D MC is the simulated dose (μGy Hist −1 ) for each energy (kVp) level. The 2nd order polynomial fitted curve in figure 2 indicates that dose is following kVp 2 as per theoretical expectation. The imaging dose D abs in (μGy) in every voxel for a given patient was calculated using: where D MC is the MC dose in a given voxel (the superscripts indicate the left or right x-ray beam), F cal is defined above for the relevant tube potential, mAs is the tube setting for a given patient size (tables 1 and 2), and a and b represent x-ray tube beam on status for left and right tubes respectively (figure 1). i.e. a = 1, b = 1 for stereoscopic imaging period; a = 1, b = 0 for left tube only monoscopic period; and a = 0, b = 1 for right tube only monoscopic period. This is needed since the same number of histories were used for both stereoscopic and monoscopic MC simulation in DOSEXYZnrc. The factor 1.67 is the imaging frequency of ExacTrac in (Hz) and T is the total patient treatment time in seconds (s) from all fractions (tables 1 and 2). The factor (1.67 * T) indicates the total number of images from all fractions for a given patient.

OAR and PTV dose calculation
The DOSEXYZnrc output files of the MC simulation in 3d dose format were imported to an in-house MATLAB code and calibrated as per section 2.5 to create patient specific 3D dose distributions. In order to calculate 3D imaging dose within PTV and OARs, these contours were exported from the Eclipse treatment planning system as DICOM-RT structures (Law and Liu 2009 . T-tests are caried out to determine whether the distribution of D50 and D2 is statistically significant between different patient size categories, i.e. (medium versus small), (large versus small), and (large versus medium). Holm-Bonferroni corrections (Holm 1979) were applied for removing type 1 (false positive) errors.

Result
3.1. Imaging dose for lung SBRT patients Figure 3 demonstrates 3D dose distribution calculated using MC simulation for a sample medium size lung SBRT patient. The first three images illustrate coronal (3a), sagittal (3b), and axial (3c) views through isocenter. Red and yellow arrows represent the direction of kV x-ray beams and the PTV is marked with a red contour. The plot (3d) represents DVHs for PTV and each OAR. Figure 4 presents the distributions of D50 (4a) and D2 (4b) imaging dose for PTV and OARs for all 30 lung SBRT patients. The distribution of D2 and D50 indicates that PTV and OARs of large patients in general received higher imaging dose while the small patients received lower imaging dose. Statistical test results confirmed that D50 and D2 of PTV and all OARs are statistically significant between large versus small patients. Bone have received the highest average D2, followed by skin. Mean D2 received by skin for large, medium, and small patients were 760 ± (142), 491 ± (94), 316 ± (87) mGy respectively, which was statistically significant between all three categories. The impact of the PTV parameters (tumor location (medial/peripheral, upper/middle/  lower lobe), and PTV volume) on D2 and D50 were examined,and the results showed no correlation between PTV parameters and imaging dose. Figure 5 demonstrates the calculated 3D dose distribution for a sample medium size prostate SBRT patient in coronal (5a), sagittal (5b), and axial (5c) views along with the corresponding DVHs of PTV and OARs (5d). PTV is indicated by the red contour and the kV beams by the red and yellow arrows. Similar to the lung patients, bone and skin received more dose with long DVH tails. Figure 6 illustrates the distribution of D50 (6a) and D2 (6b) values for all 30 prostate patients. Results of Ttest for D50 and D2 distributions are presented in the relevant graphs in figure 6. The dose distribution in figure 6(b) indicates that the D2 values for skin increase with increasing patient size. However, all the other OARs and PTV exhibits an opposite trend in D50 and D2 with respect to patient size (i.e. internal organs received lower dose in larger patients). The statistical test confirmed the skin D2 is statistically significant between all three patient size categories. Mean D2 values for skin of large, medium, and small patients were 465 ±(30), 388 ±(15), 342 ±(21) mGy, respectively.

Imaging dose for prostate SBRT patients
3.3. Imaging dose relative to prescription dose Patient-specific imaging dose received by PTV and each OAR as a percentage of the prescribed dose are calculated. Imaging dose received by internal organs during prostate SBRT is noticeably lower than the lung SBRT patients. For both patient groups, the results show that the D2 values for PTV and all OARs did not exceed 5% of the prescription dose. Among all 30 lung patients the maximum D2 value received by bone during tumor monitoring was 4.30% of the prescribed treatment dose. During prostate tumor monitoring, the corresponding  maximum bone D2 value was 2.53%. The maximum D2 for PTV among all 30 patients was 2.42% and 0.29% of the corresponding prescribed dose for lung and prostate patients respectively.

Discussion
This study quantifies patient-specific imaging dose from real-time tumor monitoring using a commercial stereoscopic IGRT system for lung and prostate patients. Results showed that the additional imaging dose to PTV was a small fraction of the prescription dose (at most 2.42% for lung and 0.29% for prostate). We employed 60 patients and estimated 3D dose distribution to various OARs in pelvis and thorax regions, and unlike previous studies, the results characterized the effect of patient size.
Unlike other real-time IGRT technologies such as MRI-guided without ionizing radiation, imaging dose for OAR should be considered for x-ray based IGRT. Imaging dose received by a particular OAR is mainly affected by the x-ray energy and fluence as well as beam attenuation. Energy and fluence are set by patient size dependant imaging protocol. Larger kVp and mAs are required for larger patients (i.e. higher BMI) as per tables 1, 2. This causes an increase in the skin dose with patient size. In contrast, increasing soft tissue thickness and thus more attenuation with increasing BMI in large patients (Darvish-Molla 2019) reduces imaging dose received by internal OARs. Since there are no beam attenuation effects Darvish-Molla (2019) at the skin (except attenuation from the treatment couch which is uniform for all patients), the skin dose increases with the patient size. The statistical test has confirmed that the D2 values for skin were statistically significant and higher in large patients compared to small patients in both cases of lung (figure 4(b)) and prostate (figure 6(b)). A similar trend can be observed in the imaging dose of PTV and internal OARs for lung patients ( figure 4) where the large patients received statistically significant higher dose compared to the small patients (except for aorta). On the contrary, for prostate patients, the imaging dose to PTV and internal OARs followed an opposite trend (figure 6) than lung. i.e. larger patients receive lower doses. For instance, in prostate case, the average bone D2 in small patients was 827 mGy versus 586 mGy in large patients, while in lung case the average bone D2 for the small and large patients were 926 and 1622 mGy respectively. This is mainly due to the increased attenuation in larger patients in pelvis versus thorax anatomy considering the lower density of the lung.
Another observation is the difference in the spread of imaging dose data between thorax and pelvis anatomy. Standard deviations of the imaging dose for the lung patients were higher compared to the prostate patients. For example, the coefficient of variance (standard deviation/mean) of PTV and skin D2 in lung patients were 0.71 and 0.41 respectively while for prostate patients they were 0.27 and 0.14 respectively. This is due to the fact that the isocenter was geometrically similar for all 30 prostate patients. However, for lung patients, the isocenter location greatly varied from one patient to another depending on tumor location. Furthermore, the highest skin D2 from the lung patients was 951 mGy, while for the prostate patients it was 541 mGy. Patient specific factors that would likely cause imaging dose approach 5% limit are patient size and the location of PTV. In our data the same two lung patients who received the highest skin D2 values (951 mGy, 948 mGy), also received the highest bone D2 (1942 mGy, 2062 mGy respectively). Both cases were large patients with PTV location close to posterior to allow imaging beam overlap. Thus, for a posterior lung tumor close to skin, stereoscopic beam overlap may cause approaching the 5% limit. Table 4 shows a comparison of results from previous studies on pelvis and thorax imaging dose using different real-time tumor monitoring technologies. The first two studies (Juneja et al 2016, Legge et al 2017 have reported the imaging dose for real-time tumor monitoring with Kilovoltage Intrafraction Monitoring (KIM) technology for lung IGRT and prostate IGRT respectively. They measured surface dose and absorbed dose to PTV following the Institution of Physics and Engineering in Medicine and Biology (IPEMB) protocol (IPEMB 1996). Depuydt et al (2013) have measured imaging dose using thermoluminescence dosimeter (TLD) delivered from Brainlab Vero system during lung SBRT. Spezi et al (2012) and Shiinoki et al (2018) have utilized MC -EGSnrc technique to calculate the patient specific imaging dose from Elekta X-ray Volumetric Imaging (XVI) system (for prostate and lung SBRT) and SyncTraX system (for lung SBRT) respectively. The Vero system in Depuydt et al (2013), uses a similar imaging technique to ExacTrac system in this study. However, they report dose per fraction and use a correlation model in tumortracking allowing to image less frequently. Therefore, it is difficult to directly compare their results with ours (and the SyncTraX system) where continuous imaging is used and the total dose from all fractions is reported. The SyncTraX system has presented the same parameters (D50 and D2) as the current study with different detector (image intensifier in SyncTraX versus flat panel in ExacTrac) and their reported doses are similar to our results. All the above studies had fewer number of patients (1-10), compared to this study with 30 patients per site.
Although in this study we focused on lung and prostate sites, the results here can also apply to other treatment sites in pelvis (e.g. bladder) , or thorax for liver and spine SBRT (Koo et al 2021). As expected, the results show that bones received the highest average D2 (due to photoelectric effect) followed by the skin in both  (Bistrović et al 1986).
For pre-treatment positioning of the patients in our study, two initial and verification CBCTs were used for prostate patients. For lung patients, two initial and verification ExacTrac images were required in addition to two CBCTs. These pre-treatment CBCTs also contribute addition imaging dose to the OARs (Aird 2004, Murphy et al 2007. Several studies have been carried out to quantify imaging dose from Varian on board CBCT (Ding and Coffey 2009, Ding and Munro 2013. For thorax CBCT, skin and bone D10 (dose delivered to 10% of organ volume) are reported to be 6.4-8.9 mGy and 14.7-22.5 mGy respectively while for pelvic CBCT 22.6-29.2 mGy and 46.1-57.2 mGy are reported (Nelson and Ding 2014). These D10 values from 2 CBCTs are at most 0.3% of the prescribed doses of the current study (0.09% for lung 48 Gy/4fx; 0.29% for prostate 40 Gy/ 5fx). Moreover, the maximum imaging dose from a single pre-treatment ExacTrac image is 0.74 mGy and 0.61 mGy for thorax and pelvis respectively (Darvish-Molla et al 2020) which can be safely ignored in comparison to CBCT and real-time imaging. Thus, with the addition of pre-treatment CBCTs, the total dose in the worst case scenario for the highest reported imaging dose (4.30% of prescribed dose to bone in lung patients) do not exceed the 5% limit. Furthermore, PTV margins used in this study were based on standard pre-treatment imaging protocol without real-time tumor monitoring. In the case of utilizing real-time tumor monitoring during the treatments, PTV margin calculation methods such as van Herk technique (van Herk et al 2000) can be applied to reduce PTV margins. Thus, the total treatment and imaging dose to OARs can be further reduced. Also, the fraction of imaging dose during stereoscopic/monoscopic periods assumed to be dependant on fractional gantry angle without including variable gantry speed or variable monitor unit (MU) due to beam modulation. Since these details are available in the treatment planning system, they could be modeled to scale the dose accordingly.
Image guidance and in particular real-time tumor monitoring can greatly contribute to treatment margin reduction and thus reducing treatment radiation exposure (Ding and Munro 2013). However, imaging adds additional radiation to OARs on top of the treatment dose (i.e. always positive uncertainty) which may be important for an OAR already close to its tolerance of radiation toxicity. Fortunately, with the available technologies (such as MC), the imaging dose can be readily accounted for in the treatment planning process, although this is not a standard clinical practice yet. MC is considered the gold standard technique for accurate dose calculations and its use is widely adopted for treatment planning purposes (Andreo 2018). However, its assumption and limitations should be considered when implementing it in treatment planning systems (Chetty et al 2007, p. 105). For instance, it could result in statistical noise in dose calculation for lower number of histories (e.g. <10 4 ). Thus, a large number of histories (>10 10 as in this study) is needed which could cause long computation time. This issue has been largely addressed with the advent of fast MC algorithms and the use of modern computer hardware (Jia et al 2011).

Conclusion
Patient-specific imaging dose for stereoscopic/monoscopic real-time kV ExacTrac image guidance system was investigated using Monte Carlo in 30 lung and 30 prostate SBRT patients and the effects of patient size was demonstrated. The highest skin dose D2 values were 1.98% and 1.35% of the prescribed dose for lung and prostate patients respectively. Additional dose to PTV as percentage of the prescribed dose was at most 2.42% for lung and 0.29% for prostate patient. which is within the 5% range recommended by AAPM TG-180. Dose variations were larger in lung compared to prostate patients due to geometry. Patient size was an important factor to determine additional imaging dose. Larger patients received more skin dose in both lung and prostate patients. For internal OARs, larger patients received more dose in lung patients while the trend was opposite for prostate patients.

Data availability statement
The data cannot be made publicly available upon publication because no suitable repository exists for hosting data in this field of study. The data that support the findings of this study are available upon reasonable request from the authors.