Electromagnetic-Guided DMLC Tracking Enables Motion Management for Intensity Modulated Arc Therapy

Abstract

Purpose

Intensity modulated arc therapy (IMAT) is attractive due to high dose conformality and efficient delivery. However, managing intrafraction motion is challenging for IMAT. The purpose of this research was to develop and investigate electromagnetic-guided DMLC tracking as an enabling technology to treat moving targets during IMAT.

Materials/Methods

A real-time 3D DMLC-based target tracking system was developed and integrated with a linac. The DMLC tracking software inputs a real-time electromagnetically measured target position and the IMAT plan and dynamically creates new leaf positions directed at the moving target. Low and high modulation IMAT plans were created for lung and prostate cancer cases. The IMAT plans were delivered to a 3-axis motion platform programmed with measured patient motion. Dosimetric measurements were acquired by placing an ion chamber array on the moving platform. Measurements were acquired with tracking, without tracking (current clinical practice), and with the phantom in a static position (reference). Analysis of dose distribution differences from the static reference used a γ-test.

Results

On average, 1.6% of dose points for the lung plans and 1.2% of points for the prostate plans failed the 3mm/3% γ-test with tracking; without tracking 34% and 14% (respectively) of points failed the γ-test. The delivery time was the same with and without tracking.

Conclusions

For the first time electromagnetic-guided DMLC target tracking with IMAT has been investigated. Dose distributions to moving targets with DMLC tracking were significantly superior to those without tracking. There was no loss of treatment efficiency with DMLC tracking.

Introduction

The purpose of this research was to develop and investigate electromagnetic-guided dynamic multileaf collimator (DMLC) tracking as an enabling technology to treat moving targets during intensity modulated arc therapy (IMAT). IMAT is attractive due to the high dose conformality and efficient delivery (1-3). However, managing intrafraction motion is challenging for rotational therapy techniques. Gating is not suitable as a heavy rotating gantry needs to be quickly and repeatedly started and stopped, and gated delivery is inefficient. An attractive solution that reduces the mechanical stress and increases efficiency is to use the dynamic multileaf collimator (DMLC) to track the moving target when real-time target motion data is available. Other potential motion management solutions for IMAT (not discussed here) are couch tracking, breath hold and abdominal compression techniques.

DMLC tracking has been experimentally investigated on Accuknife,(1) Siemens (2), Tomotherapy, (3) and Varian (4-11) systems. Based on publication dates, the number of research groups and companies interested in DMLC tracking is growing rapidly, though to our knowledge DMLC tracking has yet to be implemented for patient treatments.

A method to monitor real-time target motion during radiotherapy delivery is the Calypso 4D Localization system (Calypso Medical, Seattle, WA) that uses electromagnetic radiation to localize implanted transponders with respect to a planar array, and optical tracking to determine the array position. Balter et al. (12) demonstrated sub-millimeter system accuracy for static and dynamic motion up to 3cm s⁻¹. Sawant et al. (13), using an earlier version of the integrated DMLC tracking system used for the current experiments, showed the system geometric accuracy to be less than 2 mm for lung and 1 mm for prostate motion. With the same system used for the Sawant geometric accuracy measurements, Smith et al. (14) quantified the dosimetric accuracy and found that tracking dose measurements were similar to the gating measurements for intensity modulated radiation therapy (IMRT), with a 2-5 fold increase in delivery efficiency for tracking compared with gating.

In this study we describe the implementation and dosimetric investigation of electromagnetic-guided DMLC tracking during IMAT using clinically realistic plans and patient-derived motion traces.

Methods and Materials

The clinical focus of this study concentrated on lung and prostate cancer. Lung tumors were chosen due to the motion magnitude and complexity, and the canine (15) and human (16) lung investigations with the Calypso system. Prostate cancer was chosen because significant complex motion is seen in some (17) cases and the Calypso system is FDA cleared for clinical radiotherapy treatments in the prostate.

Motion traces and IMAT plans

The goal of this study was to span the complexity of motion and IMAT modulation that may be expected in clinical practice. Though it is clearly difficult, if not impossible, to span the entire spectrum of clinical variability, to achieve varied motion, motion files of higher than average motion magnitudes and complexity were selected from large abdominal/thoracic (18) and prostate (17) motion databases. The selected abdominal/thoracic traces were all from lung tumors as used by Poulsen et al. (9) representing ‘typical’ motion, high frequency breathing, predominantly left-right motion, and baseline variations. The selected prostate cases represent continuous drift, persistent excursion, transient excursion, and high frequency excursions as used by Poulsen et al. (10). The motion files used for the current study are shown in Figure 1. Consistent with the Poulsen studies, and to allow the prediction algorithm 20 seconds of learning time, the beam was started 20 s after the start of the data files (the pre-beam-on data is not shown in Figure 1, but the phantom center was at the isocenter (0,0,0) 20 seconds prior to the beam initiation). The phantom motion and treatment initiation was approximately the same for the tracking and without tracking measurements.

Figure 1.

(a) The lung and (b) prostate tumor motion traces used for this study representing higher than average motion magnitude and complexity for both sites. The lung tumor sites were right hilum (typical case), right lower lobe (high frequency), right lower lobe (predominantly left-right) and left upper lobe (baseline shifts).

The IMAT plans were created using the RapidArc software in the Eclipse Treatment Planning System (version 8.6) from Varian. To include different patient motion, lung (larger motion with a periodic motion component) and prostate (smaller motion with a random motion component) plans were created based on patients treated with RapidArc at Stanford. Plans with low modulation (by optimizing only the PTV dose), and high modulation (by optimizing the PTV and critical structures with difficult constraints) were created to span the range of clinical complexity. The lung plan monitor units (MUs) were 342 and 596 for the low and high modulation cases respectively, and the prostate plan MUs were 432 and 737. All arcs spanned 358° starting at 1° from the right posterior direction of a supine patient. The collimator was set to 90° to have the population-averaged major axis of respiratory motion (superior-inferior direction) (18) aligned with the MLC leaf travel direction. For all plans 2 Gy was given to 95% of the PTV. The energy and dose rate for all plans was 6 MV and 600 MU/minute respectively, with treatments lasting approximately 1 minute.

Electromagnetic (EM)-guided DMLC tracking system

The experimental set-up for the measurements is shown in Figure 2. A real-time 3D DMLC-based target tracking system was developed and integrated with a 21EX linear accelerator (Varian Medical Systems). The real-time 3D position input was provided by a (1) Calypso 4D Localization system monitoring the position of embedded transponders and sent to the (4) DMLC tracking system. (2) For the lung motion traces a kernel density estimation algorithm (19) was used to account for the 150 ms system latency. No prediction algorithm was used for the prostate cases. (3) The DMLC tracking software integrated the real-time position data stream with the planned IMAT leaf sequence to dynamically create new leaf positions directed at the moving target. The modified leaf sequence accounting for the target translation was sent to the (5) DMLC controller that actuated the leaf motion. Dosimetric measurements were acquired by placing a Seven29 2D ion chamber array (PTW, Freiburg) with 2cm of solid water buildup material on the moving platform. The PTW has 729 vented plane-parallel 5×5 mm² ion chambers in a 27×27 array with a center-to-center spacing of 10 mm. The Seven29 array has a small angular dependence. However, for the current study where the difference between delivery with and without tracking was studied, the angular variation is common to all of the delivery types and therefore is expected to have a minimal impact on the findings. An electromagnetic shielding ‘Faraday cage’ (two layers of aluminum foil) was placed around the detector and cable to reduce the current induced in the ion chamber array by the Calypso system. Without this magnetic shielding, a dark field measurement over one minute yielded ~0.2 Gy. With the magnetic shielding the dark field readings were reduced to 0.002Gy/minute and below.

Figure 2.

A schematic (top) and picture (bottom) of the experimental set-up used for the electromagnetic-guided DMLC tracking IMAT experiments.

For each plan and motion trace combination three delivery scenarios were considered:

1. Tracking, in which EM guidance was used with DMLC tracking to deliver IMAT to the moving target representing the investigational arm of the study;

2. No tracking, in which the target was moving, but no modification to the planned delivery was made representing the current clinical practice of IMAT treatments; and

3. Static delivery, in which the target was not moving, used as the reference for the dynamic studies, and represents the plan that the treatment planning team evaluates on the patient anatomy.

The tracking and no tracking dose distributions were compared to the static delivery dose using 2 mm/2% and 3 mm/3% dose difference γ-tests (20), with a 5% low dose threshold below which comparisons were ignored. The percentage dose difference was computed relative to the maximum dose in the reference static plan. The 3 mm/3% threshold results are focused on here as the values are commonly used clinically for IMRT QA (21), and also 3 mm corresponds to a little over half the inner leaf width of the Millennium MLC (2.5 mm) below which, on an individual leaf basis, explicit motion compensation cannot be performed unless the leaves could be moved laterally and/or couch compensation was used. The fraction of points failing the γ-test was recorded for each plan and each tumor motion trace. The comparison method is shown in Figure 3.

Figure 3.

Comparison method for this study. The isodose curves obtain from the moving target with and without tracking scenarios were compared with the isodose curves from the delivery to a static target. The percentage of values greater than 5% of the maximum dose that failed the 3% dose difference and 3 mm distance to agreement γ criteria were computed.

Note that the delivery of the treatment plans did not modulate the dose rate or gantry speed during the delivery, as this capability (Rapidarc for the Varian linacs) was not available on the accelerator used for these experiments. This limitation is unlikely to significantly affect the results or conclusions as the dose rate does variation in RapidArc does not affect the leaf position, and any modulation of the gantry speed would slow the delivery and put less stress on the leaf motion. DMLC tracking results for Rapidarc delivery including dose rate or gantry speed modulation with DMLC tracking (for sinusoidal motion monitored by an optical input) have been shown by Zimmerman et al. (11).

All measurements were performed three times to quantify the reproducibility of the results. Statistical tests to quantify the significance of the motion tracking vs. no motion tracking γ-test results used a one-sided paired Student’s t-test.

Results

Example isodose curves of comparing motion tracking and not motion tracking to the static delivery are shown in Figure 4. The plans delivered without motion tracking tended to have misshaped and shifted dose distributions compared with the static distributions, as the short term intrafraction motion and longer term baseline drifts were not accounted for.

Figure 4.

Example isodose curves focused on the high dose region. The measurements were from the low modulation lung plan with the ‘typical’ case motion pattern. The red squares indicate points failing the γ-test. For scale purposes, the centers of adjacent red squares are 0.5 cm apart.

The 2mm/2% and 3mm/3% γ-test results over all measurements are shown in Figure 5/Table 1 and Figure 6/Table 2 respectively. In all cases, individually and collectively, the motion tracking results showed significantly fewer points failing the γ-test than the without motion tracking results. The three exceptions to statistical significance differences were for the 3mm/3% γ-test low and high modulation lung plans for the baseline shifts traces (p=0.15 and 0.07 respectively) and low modulation prostate plans for the transient excursion traces (p=0.15). For two of these cases, the tracking results showed zero failure and the without tracking failures were not significantly different from zero. In some cases the without motion tracking results showed up to 74% 3mm/3% γ-test failure results (low modulation lung plan, typical tumor trace). The maximum discrepancy observed for the 3mm/3% γ-test with motion tracking was 8% (high modulation lung plan, baseline shift tumor trace).

Figure 5.

The percentage of dose values that failed the 2% dose difference and 2 mm distance to agreement γ-test with and without tracking for the IMAT plan types and motion traces studied. P-values comparing with motion tracking to without motion tracking are given for each motion trace and also for the plan type combining all of the motion traces. The error bars are ± 1 standard deviation. Note the y-scale is different for all four plots.

Table 1.

Summary of the average number of values failing the 2% dose difference 2 mm distance to agreement γ-test for all experiments

Plan type and motion trace	With tracking	Without tracking	% reduction	p-value
Low modulation lung	4.1%	59%	93%	<0.001
High modulation lung	9.1%	56%	84%	<0.001
Combined lung	6.6%	58%	89%	<0.001
Low modulation prostate	0.6%	31%	98%	<0.001
High modulation prostate	11%	36%	70%	<0.001
Combined prostate	5.7%	34%	83%	<0.001
All combined	6.1%	46%	86%	p<0.001

Figure 6.

The percentage of dose values that failed the 3% dose difference and 3 mm distance to agreement γ-test with and without tracking for the IMAT plan types and motion traces studied. P-values comparing with motion tracking to without motion tracking are given for each motion trace and also for the plan type combining all of the motion traces. The error bars are ± 1 standard deviation. Note the y-scale is different for all four plots and different from Figure 5.

Table 2.

Summary of the average number of values failing the 3% dose difference 3 mm distance to agreement γ-test for all experiments

Plan type and motion trace	With tracking	Without tracking	% reduction	p-value
Low modulation lung	0%	34%	100%	<0.001
High modulation lung	3.2%	35%	91%	<0.001
Combined lung	1.6%	34%	95%	<0.001
Low modulation prostate	0%	12%	100%	0.009
High modulation prostate	2.4%	16%	85%	<0.001
Combined prostate	1.2%	14%	93%	<0.001
All combined	1.4%	24%	94%	p<0.001

For the lung plans combined (Table 2) 1.6% of the dose points failed the 3mm/3% γ-test with tracking compared to 34% without tracking (p<0.001). For the prostate plans 1.2% and 14% of the dose points failed for the tracking and without tracking cases, respectively (p<0.001). Overall, comparing tracking to without tracking, a 95% 3mm/3% γ-test improvement was observed for the lung plans and 93% for the prostate plans.

IMAT plans with DMLC tracking were delivered without beam holds and completed in the same time as the delivery of the static and no-tracking plans.

Discussion

EM-guided DMLC target tracking with IMAT has been investigated. Dose distributions to moving targets with DMLC tracking were significantly superior to those without tracking, though in some cases discrepancies from static delivery were observed with motion tracking. The tracking results can be further improved by faster and/or thinner leaves, lower latency, improved prediction, and planning changes. There was no loss of treatment efficiency with DMLC tracking.

The only previous publication describing dosimetric results from electromagnetic-guided DMLC tracking studied dynamic and step and shoot IMRT delivery but did not include arc therapy (14). Since the experiments for the previous study, significant improvements have been made to the DMLC tracking system, including general bug fixes and software improvements, a different motion prediction algorithm (kernel density estimation (19) rather than modified linear adaptive filter), and a reduction in the system latency from 220 ms (13) to 150 ms (22) due to engineering improvements and hardware changes performed by Calypso. Another important change is that the number virtual sub-leaves (a parameter in the leaf sequencing code (5)) was set to one, i.e. equal to one leaf. We have found that a sub-leaf parameter value greater than one can be limiting for apertures in adjacent leaves spaced widely from each other that can occur in IMRT and IMAT fields, potentially placing dose in between these apertures. Improvements to the limitations of the current leaf sequencing process are currently under development (23). As expected, the benefit of motion tracking is highest when the motion is largest – without tracking more points failed the γ-test for the lung than the prostate, however improvements with tracking were seen for all of the plans and motion traces. There was a tendency for the tracking results to be worse as the degree of modulation increased: no points failed the γ-test with tracking for any of the motion traces for the low modulation lung and prostate plans indicating a clear plan modulation dependence. The higher failures also appeared to occur for traces where the motion perpendicular to the leaf motion direction was significant. The left-right and anterior-posterior components of motion in Figure 1 will be perpendicular to the leaf motion direction for gantry angles near 0/180 and 90/270 respectively, as seen for the baseline shifts (lung) and persistent excursion (prostate) traces. This limitation of accounting for target motion perpendicular to the leaf travel direction when apertures formed by adjacent leaf pairs are far apart, as can occur for IMRT and IMAT plans, is known (see Fig. 1 of George et al. (24) for a schematic diagram of this feature). Also, motion along the leaf direction can be corrected for continuously; the finite leaf thickness limits the resolution with which motion parallel to the leaf direction can be corrected. To alleviate this problem, during planning, in principle, the placement of these apertures could be adjusted to avoid this problem, however this would increase the constraints on the treatment plan and degrade the plan quality as calculated by the treatment planning system, though neither the implementation nor magnitude of this degradation at planning – and subsequent improvement in plan delivery – has been studied here. There were some high frequency components in the traces used in this study, particularly for the prostate cases as shown in Figure 1. These high frequency components may be due to measurement noise rather than actual tumor motion, and will have the effect of increasing the requested MLC leaf motion. Having a smoother input may further reduce some of the difference between the tracking and static measurements. Additionally, gating the beam when the leaves are outside tolerance would be expected to further improve the motion tracking results at the expense of increased treatment time. The current DMLC IMAT tracking approach can be considered the maximum efficiency solution, and increased accuracy could be traded off by increased treatment time and potentially gantry mechanical stress when gating arc delivery. Another method that may improve the dosimetric results for DMLC tracking is to align the collimator with the patient-observed major axis of motion. As seen in Figure 1 tumor motion and direction change from patient to patient and with time. Also for arc radiotherapy the angle of the major axis of tumor motion will change as the gantry rotates for all motion apart from pure superior-inferior motion.

The difference in the dose distributions without motion tracking is due to a combination of the shift in the target position from the initial set-up point prior to treatment, and shifts in the baseline position and motion during treatment causing the ‘interplay’ effect initially described by Yu et al. (25) with further theoretical (26-29) and experimental (30-33) studies by several groups. The estimated dosimetric impact of the interplay effect varies widely between these studies and is a combination of the target motion pattern, delivery complexity and delivery time. In the current study the worst result from a dosimetric perspective was without motion tracking for the lung ‘typical case’ trace and low modulation plan. This trace exhibits a shift between the initial set-up point (the target was at the isocenter 20 s prior to starting beam delivery in this study) as well as large motion during the treatment. The dosimetric effects are shown in Figure 4 where the without tracking results are shifted and misshaped with respect to the static distribution. If the target position were being monitored during treatment the shift could have been corrected for. The results for the other three lung traces (high frequency breathing, predominantly left-right motion and baseline shifts) do not exhibit this shift prior to treatment and the results are more indicative of the interplay effect only rather than the combined shift and interplay effect.

An important observation from the repeat measurements (Figures 5 and 6) is the large variability in the dosimetric results without motion tracking. For these experiments, both with and without tracking the phantom motion and beam delivery were started at approximately the same time (variability of seconds). This large dose variability without tracking for repeat experiments, even for the same motion trace, implies that the measurement results – and hence absorbed dose to patients – are not predictable and are sensitive to the motion magnitude and pattern which will change with time for each patient.

The lung and prostate motion traces here had higher than ‘normal’ patient motion, but they were patient-derived and therefore the measured results can be applied to estimate the difference in dose delivered to the patient compared to that from the plan for certain clinical situations. The prostate motion was considerably less than that of the lung, however in most of the prostate cases studied, over 5% of the points failed the γ-test without tracking. The dosimetric results without tracking for many of these cases would not pass normal pre-treatment QA guidelines. The AAPM Task Group 119 Report (21) showed for an IMRT prostate dosimetry study comparing planned and delivered doses to static targets, that on average, 2% of points failed the 3mm/3% γ-test. Comparing tracking to static target results showed on average less than 2% (1.2%) 3mm/3% γ-test disagreement. The without tracking to static delivery results showed on average 14% γ-test disagreement, seven times the disagreement of that found during the TG 119 study. This result indicates that unaccounted for motion may be a significantly larger source of error in treatment delivery than all of the contributing factors between planned and delivered doses to static targets combined.

The current study was performed on research systems that are not FDA-cleared for clinical use, however the dosimetric results on the phantom cases are compelling and clinical implementation will allow more accurate radiotherapy for improved patient care.

A limitation of the current study is that only target translation was accounted for, and further work, should such information be available, could include accounting for real-time tumor rotation and/or deformation. Rotation of over 30 degrees for prostate tumors (34) and over 45 degrees for lung tumors (35) have been observed, indicating an important future direction for real-time tumor tracking.

Conclusion

For the first time electromagnetic-guided DMLC target tracking with IMAT has been demonstrated to account for lung and prostate motion during treatment delivery. Dose distributions to moving targets with DMLC tracking were significantly superior to those without tracking. On average, 1.6% of points for the lung plans and 1.2% points for the prostate plans failed the 3mm/3% γ-test with tracking; without tracking, 34% and 14% (respectively) of points failed the γ-test. There was no loss of treatment efficiency with DMLC tracking. This promising method uses research software integrated with commercially available position monitoring (Calypso) and delivery (Varian) hardware and therefore has a clear path to clinical implementation.