Correlation between intrafractional motion and dosimetric changes for prostate IMRT: Comparison of different adaptive strategies

Abstract Purpose To retrospectively analyze and estimate the dosimetric benefit of online and offline motion mitigation strategies for prostate IMRT. Methods Intrafractional motion data of 21 prostate patients receiving intensity‐modulated radiotherapy was acquired with an electromagnetic tracking system. Target trajectories of 734 fractions were analyzed per delivered multileaf‐collimator segment in five motion metrics: three‐dimensional displacement, distance from beam axis (DistToBeam), and three orthogonal components. Time‐resolved dose calculations have been performed by shifting the target according to the sampled motion for the following scenarios: without adaptation, online‐repositioning with a minimum threshold of 3 mm, and an offline approach using a modified field order applying horizontal before vertical beams. Change of D95 (targets) or V65 (organs at risk) relative to the static case, that is, ΔD95 or ΔV65, was extracted per fraction in percent. Correlation coefficients (CC) between the motion metrics and the dose metrics were extracted. Mean of patient‐wise CC was used to evaluate the correlation of motion metric and dosimetric changes. Mean and standard deviation of the patient‐wise correlation slopes (in %/mm) were extracted. Results For ΔD95 of the prostate, mean DistToBeam per fraction showed the highest correlation for all scenarios with a relative change of −0.6 ± 0.7%/mm without adaptation and −0.4 ± 0.5%/mm for the repositioning and field order strategies. For ΔV65 of the bladder and the rectum, superior–inferior and posterior–anterior motion components per fraction showed the highest correlation, respectively. The slope of bladder (rectum) was 14.6 ± 5.8 (15.1 ± 6.9) %/mm without adaptation, 14.0 ± 4.9 (14.5 ± 7.4) %/mm for repositioning with 3 mm, and 10.6 ± 2.5 (8.1 ± 4.6) %/mm for the field order approach. Conclusions The correlation slope is a valuable concept to estimate dosimetric deviations from static plan quality directly based on the observed motion. For the prostate, both mitigation strategies showed comparable benefit. For organs at risk, the field order approach showed less sensitive response regarding motion and reduced interpatient variation.

On the one side, the resulting motion information facilitated the development of real-time mitigation techniques using a couch, [15][16][17] multileaf-collimator (MLC)-tracking, 18,19 or gated delivery. 20,21 On the other side, it is possible to reconstruct the delivered dose, that is, a four-dimensional (4D) dose distribution, based on the actually observed motion during the treatment. [22][23][24] Although motion of the prostate during radiotherapy has been detected 25,26 and adaptive strategies have been applied for prostate patients, 2,15,27 a comprehensive comparison of motion management strategies is still pending. 28 Here, we present a coherent analysis of the correlation between intra-fractional prostate motion and dosimetric changes induced by the motion considering both target coverage and organ at risk (OAR) sparing metrics. Real-time motion data 29,30 stemming from 21 patients were used, and different motion mitigation approaches were compared. Our study provides valuable guidance to implement clinical decision support regarding the adequate motion mitigation strategy for prostate IMRT patients. Furthermore, we introduce two innovations: (a) a novel motion metric which quantifies the target motion relative to the actual delivered beam (DistToBeam) and (b) the strategy of changing the field order (AngHV) that can be applied for many clinical scenarios in IMRT.

2.A | Patient data
Data for 21 prostate patients who received step-and-shoot IMRT with a Siemens Artiste treatment device equipped with a 160-leaf MLC (Siemens AG, Munich, Germany) in 35 fractions at our institution was acquired. Informed consent was obtained from all included individual participants. The study was approved by the ethics committee of the medical faculty at our university. The prostate was defined as an integrated boost volume and the treatment plan was optimized with the goal of a prescribed median dose of 76 Gy with an enclosing isodose of at least 95% (72.2 Gy). The prostate was expanded by a margin of 7 mm including the base of the seminal vesicles to construct the planning target volume (PTV) with a prescribed dose of 70 Gy with an enclosing isodose of at least 95% (66.5 Gy). Treatment planning was performed with our in-house clinical treatment planning systems VOXELPLAN and KonRad which facilitate a singular value-decomposed pencil beam algorithm. 31 The immanent inaccuracies of pencil beam algorithms compared with more sophisticated approaches in heterogeneous media were assumed negligible for prostate treatments. 32 Nine coplanar equiangular fields with a 6-MV photon beam with a flattening filter were used (Gantry angles at 200°, 240°, 280°, 320°, 0°, 40°, 80°, 120°, 160°in a clockwise order according to the IEC 61217 standard). The patient was positioned in head-first supine position using a head rest, a knee rest, and a foot rest. There was no dietary protocol, but the patients were asked for having an empty rectum and a full bladder during the planning CT and treatment fractions. For each patient, the clinically applied treatment plan was used as a reference to simulate different motion mitigation strategies. The average number of MLC segments per plan was 81, and the total number of fractions was 734 in our cohort.

2.B | Motion metrics
Target motion was monitored with the Calypso 4D localization system (Varian Medical Systems Inc., Palo Alto, CA) 33 To extract intrafractional motion for this study, the initial displacement has been set to zero with an offset correction based on average motion data considering 30 s before the onset of treatment for each fraction. This approach corresponds to perfect patient positioning. Missing data (0.4% in the recorded data) were imputed by linear interpolation if the gap was during the irradiation and lasted less than a few seconds. The actual treatment was delivered without intrafractional motion adaption (NoCorr). Based on the time correlated beam delivery information, the observed target motion was sampled per delivered beam segment and used for the reconstruction of the dose distribution that was actually delivered during the treatment.

2.D.2 | Repositioning scenario
Real-time adaptation scenarios with a couch-shift technique have been simulated with repositioning thresholds of 3, 5, and 7 mm (Rep3, Rep5, and Rep7, respectively) by using the sampled target motion of the patients. In this scenario, the repositioning has been applied if the sampled displacement exceeds the threshold at any orthogonal direction. Therefore, the displacement of all components is immediately set to zero simulating infinitely fast patient repositioning. After repositioning, the sampled motion continues relative to the applied centering. Since the duration of repositioning in this simulation is neglected, the beam delivery timing and ending of the sampled motion coincides with the NoCorr case.

2.D.3 | Field order modification scenario
For this offline motion management scenario, we simulate a treatment that starts with horizontal fields and finishes with vertical fields (AngHV), that is, we apply a gantry angle order of 280°, 80°, 120°, 240°, 320°, 40°, 160°, 200°, and 0°, instead of the clockwise order. This approach is motivated by the following considerations: (a) the mean displacement of the target is larger along PA and SI than along LR direction, 25

2.E | Correlation analysis
The Pearson product-moment correlation coefficient (CC) between the dose metric (DD95 or DV65) and the motion metric (LR, PA, and SI, DistToOrigin, DistToBeam) was computed for every patient individually. The number of data points underlying the CC calculation was up to 35 (number of fractions). Subsequently, the mean values of all 21 patients were extracted. The mean value of the CC in the NoCorr scenario was used to identify the motion metric with the best correlation to the dosimetric endpoints.
Linear regression was performed to quantify the sensitivity of all dose metrics with a slope (i.e., dose metric per motion metric) in %/ mm. A motion management strategy with low motion sensitivity would result in less dose changes from the static case for a given motion (i.e., a smaller slope). Consequently, the strategy can be considered more robust against motion. The standard deviation of the slope in patient statistics was included as error estimate.

3.A | Motion metric
Mean, standard deviation (SD), minimum and maximum of the sampled target displacements are listed in Table 1

3.B | Dose quality metric
The changes in the dose metrics calculated for all scenarios are listed in Table 2

3.C | Correlation coefficients
Mean values of the patient-wise CCs are listed in   shows a more clear correlation than for the total data set averaged over all patients (compare Fig. 1).

3.D | Regression slopes
The patient-wise analysis of regression slope is shown in

3.F | Prostate
For the prostate DD95, the selected motion metric was DistToBeam with mean slope (AESD) of À0. 6   the AngHV showed reduction in both mean and SD.

3.H | Rectum
For the rectum DV65, the motion metric PA showed the highest correlation among all motion metrics. The target displacement toward   In agreement with a previous study, 24   the beam geometry in combination with a patient-wise specific analysis to extract clear correlations.
The scenarios without adaptation, NoCorr and AngCW, showed similar motion sensitivity as measured by the regression slopes, even though the sampled motion amplitudes and timings were different.
The real-time adaptation scenarios with repositioning, Rep3, Rep5, and Rep7, showed a gradual change in the sensitivity as Rep3 is the most motion insensitive among them. In this study, the Rep7 showed minor improvement in dose, see Table 2, and the motion sensitivity in mean and SD was similar to the free motion cases.
F I G . 3. Correlation plot between the prostate DD95 and the motion metric DistToBeam, plotted for each patient. The linear regression curves of patient-wise analysis is shown with a solid line; the corresponding correlation coefficients (CC) and slopes are given for each subplot.
The offline strategy, AngHV, reduced the motion sensitivity to the level of Rep3 for both PTV and prostate. As shown in Table 2, the dose quality for the AngHV was comparable to the Rep3 or The motion in SI direction was not mitigated in the AngHV approach. That needs to be investigated in further studies possibly applying noncoplanar optimized beams. 36 The extracted correlations are based on the assumption of rigid patient motion. We understand the possibility of shape changes In this study, the dosimetric changes on rectum were most beneficial in the free motion scenario ( Table 2). The motion adaptation strategies reduced deviations from the planned dose which was higher than in the free motion case (NoCorr). This means, on average, all investigated adaptation techniques increased the dose in the rectum compared with the free motion scenario. This can be explained by the prostate drift toward the posterior direction in the free motion case which moves the rectum into a lower dose region, too.
This study was restricted to step and shoot (SNS) IMRT. The applicability of the field order modification is not straightforward for some modern irradiation techniques (e.g., VMAT, tomotherapy) where irradiation follows specific irradiation trajectories on (helical) arcs. However, in the context of noncoplanar arc therapy, 37 similar considerations as exercised here could inform the design of optimized irradiation trajectories.
The correlation and motion sensitivity analysis presented in this study offers the possibility to estimate dose on moving organs before the onset of treatment. This may support decision making regarding an adequate motion management strategy before the treatment or allow for a simple estimate of the actually delivered dose after the treatment. While our finding may not easily generalize for alternative margin concepts, the presented methodology is applicable for various margin recipes, adaptation strategies, fractionation schemes, and treatment sites. Our results regarding a modified field order may inform a general discussion of the quality of beam angles in the context of anatomical motion.

| CONCLUSIONS
Correlations between intra-fractional motion and dosimetric quality have been obtained based on measured intrafractional motion of prostate patients considering treatments with and without motion management. The prostate D95 of the cumulative dose was found to be within 5% from the static treatment plan for all 21 patients in all treatment scenarios. Both online and offline mitigation strategies showed comparable benefit in motion sensitivity regarding the individual fraction doses. For the OARs, the offline approach with field order modification resulted in reduced sensitivity to motion and showed less patient variations in the individual fraction doses.