Intrafractional relationship changes between an external breathing signal and fiducial marker positions in pancreatic cancer patients

Abstract Background and purpose The purpose of this study of pancreatic cancer patients treated with respiratory‐guided stereotactic body radiotherapy (SBRT) on a standard linac was to investigate (a) the intrafractional relationship change (IRC) between a breathing signal and the tumor position, (b) the impact of IRC on the delivered dose, and (c) potential IRC predictors. Materials and methods We retrospectively investigated 10 pancreatic cancer patients with 2–4 implanted fiducial markers in the tumor treated with SBRT. Fluoroscopic images were acquired before and after treatment delivery simultaneously with the abdominal breathing motion. We quantified the IRC as the change in fiducial location for a given breathing amplitude in the left–right (LR), anterior–posterior (AP), and superior–inferior (SI) directions from before to after treatment delivery. The treatment plans were re‐calculated after changing the isocenter coordinates according to the IRCs. Four treatment‐ or patient‐related factors were investigated as potential predictors for IRC using linear models. Results The average (±1 SD) absolute IRCs in the LR, AP, and SI directions were 1.2 ± 1.2 mm, 0.7 ± 0.7 mm, and 1.1 ± 0.8 mm, respectively. The average 3D IRC was 2.0 ± 1.3 mm (range: 0.4–5.3 mm) for a median treatment delivery time of 8.5 min (range: 5.7–19.9 min; n = 31 fractions). The dose coverage of the internal target volume (ITV) decreased by more than 3% points in three of 31 fractions. In those cases, the 3D IRC had been larger than 4.3 mm. The 3D IRC was found to correlate with changes in the minimum breathing amplitude during treatment delivery. Conclusion On average, 2 mm of treatment delivery accuracy was lost due to IRC. Periodical intrafractional imaging is needed to safely deliver respiratory‐guided SBRT.


| INTRODUCTION
Respiration-induced motion of pancreatic tumors can be substantial, and it affects all radiotherapy (RT) treatment steps from pretreatment imaging to dose delivery. 1 While motion is present in all directions, it is usually largest in the superior-inferior (SI) direction, where the pancreas moves superiorly during the exhalation phase of the breathing cycle and inferiorly during the inhalation phase. 2 The range of tumor motion in free-breathing patients has been reported to be up to 6, 10, and 20 mm in the left-right (LR), anterior-posterior (AP) and SI directions, respectively. 3 Creating and delivering treatment plans with dose distributions that cover the tumor would require large internal margins (IMs) to take the entire motion range into account. In stereotactic body radiotherapy (SBRT), this method would also result in irradiation of large volumes of the surrounding organ-at-risk (OARs) such as the stomach and the duodenum, to a high dose. 4,5 Therefore, to accurately treat the tumor while keeping the irradiation of the OARs within tolerance limits, a more complex motion management strategy is required.
One motion management strategy for pancreatic cancer SBRT is respiratory-gated RT. 6 By considering select phases of the breathing cycle (such as the end-of-exhale) during pretreatment imaging, treatment planning and delivery, there will be a decrease in total volume encompassed by the moving tumor. The patient's breathing motion can be monitored by, for instance, measuring the external position of the abdominal or chest surface. This approach will decrease the volume of the required dose distribution to cover the tumor and, consequently, normal tissue irradiation will be reduced accordingly. 4 During treatment setup, the tumor motion is observed (with, for instance, fluoroscopic imaging) simultaneously with the breathing motion, and a fraction-specific relationship between the breathing signal and the tumor position is established to ensure that the planned dose distribution can be accurately delivered. The treatment delivery then is typically done using the breathing signal alone, and its accuracy is therefore dependent on the assumption that the relationship between the breathing signal and the tumor position that was established at setup also holds up during treatment delivery.
This assumption may not hold, however, due to, for instance, muscle relaxation or movement of internal organs close to the tumor during treatment. The assumption has been extensively investigated for lung tumors, 7 10 They did not, however, report how large these changes were. Several aspects of how the treatment accuracy for respiratory-gated RT holds up during treatment delivery remains to be investigated. One of the most important properties is how the shorter treatment delivery times typical for SBRT on standard linacs influence the consistency of the external-internal relationship. If larger margins between the internal target volume (ITV) and the planning target volume (PTV) are to be used to compensate for increased geometrical uncertainty, it is of interest to know whether the relationship changes are similar in all directions. Moreover, if one were aware of observable signs suggesting that the intrafractional relationship had substantially changed, treatment delivery could be halted, and the treatment setup redone before accurate treatment delivery started again.
In this study of patients with pancreatic cancer treated with SBRT on standard linacs, we investigated intrafractional relationship changes (IRC) between an external breathing signal and the positions of internal fiducial markers, and estimated its impact on the dose distribution in the ITV and OARs. We also investigated patient-and treatment-related factors potentially predicting the occurrence of IRC.

2.B | Contouring and treatment planning
The ITV was contoured on the CT 3070AVE image series, and a PTV was created from the ITV using a 3-mm isotropic margin. The CT 3070AVE images were also used for contouring of OARs. The OARs that were considered for the ten patients included in this study were the stomach, the duodenum, the spinal cord, the bilateral kidneys, and the liver. The fiducials were contoured on the CT 3070MIP images.
The planning objective for the PTV was to cover at least 95% of its volume with the prescription isodose (V 100% > 95%), whereas no objective was used for the ITV. The objective for both the stomach and the duodenum was that the dose received by 1 cm 3 should be below the prescribed dose (D 1cm3 < 100%). The spinal cord objective was D 1cm3 < 8 Gy, the bilateral kidney objective D 75% < 12 Gy, and the liver objective D 700cm3 < 15 Gy. In cases with conflicting objectives, trade-offs could be made by the treating physician.
Volumetric-modulated arc therapy (VMAT) treatment plans using two 180-degree arcs, one with clockwise and one with anticlockwise rotation, were created using the treatment plan objectives above. The nominal photon beam energy was in most cases 6 MV using the flattening filter, but 6 and 10 MV flattening-filter free beams were also used. All dose distribution calculations were performed in Eclipse with the analytical anisotropic algorithm (AAA; version 13.6.23) using a dose grid size resolution of 2.5 mm x 2.5 mm x 2.5 mm.
The SBRT was prescribed to be delivered in five fractions with a total dose ranging from 25.0 (5 × 5.0) to 45.0 (5 × 9.0) Gy. Digitally reconstructed radiographs (DRRs), including overlays of contoured fiducials, were calculated to assist in patient setup.

2.C | Treatment setup
The following procedure was utilized to treat each fraction using

2.D | Data acquisition and analyses
When the patient setup was considered satisfactory, we made a simultaneous acquisition of the breathing motion signal and fluoroscopic image data in both AP and lateral (patient right-to-left) directions before treatment delivery. After completed treatment delivery, the same AP and lateral sequences were re-acquired. Using orthogonal imaging, we evaluated the relationship between breathing signal and fiducial position in three orthogonal directions (LR, AP, and SI).
Typically, data were acquired for 15-20 s per sequence. The goal was to acquire data for two to three full breathing cycles, but in some cases with irregular breathing, this was not feasible. Only sequences containing at least one full breathing cycle were used in the analysis. Fluoroscopic images were captured at 14.8 frames per second with a detector element size of 0.388 mm x 0.388 mm. The imaging source-detector-distance was 1500 mm and the treatment source-axis-distance distance 1000 mm.
The fluoroscopic images and the RPM data were exported to Matlab (version 2014b or higher, MathWorks, Natick, MA, USA) for analysis. Using an in-house template-matching algorithm based on the normalized cross-correlation, 11 we determined the center pixel position of each fiducial on every image. To make sure that fiducial positions were accurately tracked, we visually inspected all fiducial positions in all sequences.

2.E | Imaging geometry considerations
To accurately convert the projected fiducial locations on the x-ray detector into positions in the patient coordinate system, we need to take the divergence of the x-ray beam between the fiducials and the detector into account. As described in our previous study, we estimated one representative LR and one representative AP in-room fiducial position using an iterative approach. 12 We defined the LR axis as positive toward the left-hand side of the patient, the AP axis as positive in the anterior direction, and the SI axis as positive in the superior direction.

2.F | Intrafractional relationship changes
We evaluated the IRC between the breathing signal and the fiducial positions in the AP, LR, and SI directions separately, and, whenever all four sets of images (AP and lateral before and after treatment delivery) had been acquired, the total 3D length of the changes (denoted 3D change). For each image sequence, we fit one linear curve between the RPM signal and SI fiducial positions, and one for either LR or AP fiducial positions (Fig. 1) depending on the imaging direction. To counteract the effect of irregular breathing cycles as well as uneven sampling of the breathing cycle, we made the linear fits equally weighted over the RPM range by stratifying the breathing signal into 1-mm bins and fitting the curve to the average fiducial position within each bin. We calculated the IRC as the vertical distance between the two linear fits at 25% of the common breathing range (Fig. 1). We chose 25% because, in the case of 30-70% gating, it corresponds to fiducial positions close to the center of the gating window. Fig. 1 illustrates the quantification of the IRC between the breathing signal and the fiducial positions.
In some cases, the couch was used to reposition the patient between image acquisitions. For such fractions, we corrected for this before proceeding with the analysis. In cases with AP couch shifts, the RPM signal was changed accordingly.
The calculated relationship changes can be considered overall RPM-fiducial relationship changes in that they will quantify all changes that occur during treatment delivery, including those intro-

2.G | Dose distribution impact
To estimate the dose distribution impact in the ITV, the PTV, and the OARs, we moved the patient isocenter according to the relationship changes and recalculated the dose distribution. We evaluated changes to the treatment plan objectives, and for the ITV, we evaluated changes to V 100% .

2.H | Prediction of intrafractional relationship changes
To investigate whether it is possible to predict if an IRC occurs during treatment delivery without using additional imaging, we evaluated the absolute value of the baseline drift (BLD) of the breathing amplitude, 14 the treatment delivery duration, the pretreatment fiducial motion ranges, and the fiducial motion consistency as potential predictors (see Fig. 1). The BLD was defined as the difference between the RPM signal at maximum exhale before and after treatment delivery. A negative BLD means a more posterior RPM box position after treatment delivery. We calculated the fiducial motion consistency as two standard deviations (SD) for fiducial positions in a 2-mm window around the mid-range RPM position.
We modeled the IRC in each direction for the four potential predictors using univariable linear regression. For fractions with a complete set of four image acquisitions, we also modeled the 3D IRC. In those cases, predictor values for both the motion range and the fiducial motion consistency were the corresponding 3D lengths. We also considered the squared Pearson correlation coefficient (R 2 ) in our analysis. 11 A P-value below 0.05 was considered statistically significant.

3.A | Collected data
An overview of the fluoroscopic imaging and breathing signal sequences used in the analysis is shown in Table 1. A total of 38,147 images from 152 sequences were analyzed. For 31 fractions, data from all four image acquisitions (AP and lateral before and after treatment delivery) were available and could be used to calculate the 3D IRC and its estimated dosimetric impact.
For 70 pairs of LAT and AP sequences acquired immediately after one another (Table 1)

3.C | Dose distribution impact
The dose distribution impact in target volumes and OARs is shown in Fig. 3 AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  LAT  AP  F I G . 2. Intrafractional relationship change between the breathing signal and fiducial in-room coordinates (average value for all implanted fiducials). For patients with more than one fiducial marker, the standard deviation of the relationship change for all the fiducials is typically less than 0.5 mm. LR, left-right; AP, anterior-posterior; SI, superior-inferior.

3.D | Prediction of intrafractional relationship changes
For the LR relationship change (Fig. 4, first row), the pre-RT range and the fiducial motion consistency were statistically significant predictors in univariable analysis with P-values of 0.033 and 0.005, respectively. The corresponding R 2 -values were 0.12 for the range and 0.20 for motion consistency.
For the AP relationship change (Fig. 4, second row), the absolute BLD and the treatment delivery time were statistically significant predictors in univariable analysis with P < 0.001 and P = 0.049, respectively. The corresponding R 2 -values were 0.42 for the BLD and 0.12 for treatment delivery time.
For the SI relationship change (Fig. 4, third  Given the impact of the absolute BLD on the absolute relationship change, we also analyzed the directional intrafractional change versus the BLD (Fig. 5). The relationship between the BLD and the SI shift was statistically significant (P < 0.001), whereas there was no statistically significant linear relationship in the AP direction.

| DISCUSSION
In this study, we investigated the intrafractional consistency of the relationship between a breathing signal measured on the patient surface and the motion of fiducial markers in pancreatic cancer patients.
Patients were treated with respiratory-gated RT on a standard linac with a median treatment delivery time of 8.5 min. We quantified the relationship changes in the LR, AP, and SI directions separately and found that the IRC was somewhat larger in the LR and SI directions   independently from patients' movement, we compensated for changes in patient position in the analysis. 10,15 In three fractions with large IRC, the ITV coverage decreased by more than 3% points compared to the treatment plan. In the prediction analysis, IRC was found to be correlated with different metrics for different directions, indicating that the relationship between the breathing signal and the fiducial positions may be lost for various reasons. However, the metric most strongly associated with the 3D IRC was the baseline drift (BLD) of the breathing signal. For each millimeter of absolute BLD, the 3D IRC was about 1 mm (Fig. 4, the leftmost panel on the fourth row), which alone explained 39% of the IRC variance.
The relationships between the BLD and the relationship changes ( Fig. 5) can be understood from the slopes of the RPM-fiducial positions curves (Fig. 1). If the BLD is interpreted as a systematic offset of the RPM signal (not caused by a patient position shift) that occurred between pre-and post-RT imaging, the BLD will horizontally shift the post-RT linear fit, and the IRC (in each direction) will be -BLD × slope. This is closely related to the argument above regarding patient shifts in the AP direction. and 30-40-min blocks increased the average 3D errors to 2.0, 2.8 and 3.6 mm, respectively. In our study, we did not find a statistically significant relationship between treatment delivery time and 3D IRC (P = 0.074). There were, however, statistically significant relationships for the treatment delivery time for the AP IRC and the SI IRC ( Fig. 4). It should be kept in mind that we investigated the IRC after removing the effect of patient shifts. The likelihood of uncorrected patient shifts occurring during treatment delivery increases with time, 9 and such shifts will decrease treatment accuracy independently of IRC.
The risk that the relationship between an external breathing signal and the tumor location may change during treatment must be

ACKNOWLEDG MENT
We gratefully thank Varian Medical Systems for funding this study.

CONFLI CTS OF INTEREST
Varian Medical Systems funded this study.