Left ventricle segment-specific motion assessment for cardiac-gated radiosurgery

Purpose. Cardiac radiosurgery is a non-invasive treatment modality for ventricular tachycardia, where a linear accelerator is used to irradiate the arrhythmogenic region within the heart. In this work, cardiac magnetic resonance (CMR) cine images were used to quantify left ventricle (LV) segment-specific motion during the cardiac cycle and to assess potential advantages of cardiac-gated radiosurgery. Methods. CMR breath-hold cine images and LV contour points were analyzed for 50 controls and 50 heart failure patients with reduced ejection fraction (HFrEF, EF < 40%). Contour points were divided into anatomic segments according to the 17-segment model, and each segment was treated as a hypothetical treatment target. The optimum treatment window (one fifth of the cardiac cycle) was determined where segment centroid motion was minimal, then the maximum centroid displacement and treatment area were determined for the full cardiac cycle and for the treatment window. Mean centroid displacement and treatment area reductions with cardiac gating were determined for each of the 17 segments. Results. Full motion segment centroid displacements ranged between 6–14 mm (controls) and 4–11 mm (HFrEF). Full motion treatment areas ranged between 129–715 mm2 (controls) and 149–766 mm2 (HFrEF). With gating, centroid displacements were reduced to 1 mm (controls and HFrEF), while treatment areas were reduced to 62–349 mm2 (controls) and 83–393 mm2 (HFrEF). Relative treatment area reduction ranged between 38%–53% (controls) and 26%–48% (HFrEF). Conclusion. This data demonstrates that cardiac cycle motion is an important component of overall target motion and varies depending on the anatomic cardiac segment. Accounting for cardiac cycle motion, through cardiac gating, has the potential to significantly reduce treatment volumes for cardiac radiosurgery.


Introduction
Ventricular tachycardia (VT) is a type of cardiac arrhythmia characterized by a rapid heart rhythm originating in the lower chambers of the heart.VT is most frequently caused by re-entrant electrical circuits that arise in scarred myocardium and can lead to sudden cardiac death [1,2].Implantable cardioverterdefibrillators (ICDs) are used to control and restore the heart rhythm when a VT episode occurs, but frequent ICD shocks significantly decrease quality of life.Drug therapy is limited by effectiveness and toxicities [3,4], while catheter ablation is invasive and may be unsuitable for deep or widespread myocardial scar [5][6][7][8].
Cardiac radiosurgery (also known as cardiac radioablation or stereotactic arrhythmia radioablation) is a non-invasive treatment modality for VT that uses a medical linear accelerator to irradiate the arrhythmogenic region within the heart [9][10][11][12][13][14][15].In planning for treatment of the heart, target motion is generally more complex than other radiation therapy treatments as both respiratory-and cardiac-induced motion must be taken into account.The majority of cardiac radiosurgery treatments using C-arm linear accelerators have used a combined internal target volume (ITV) approach, where the ITV treatment margin encompasses the overall motion [13,[16][17][18][19][20][21][22].While this ensures the target is treated, it leads to significant radiation to surrounding non-target tissues.It is also important to note that the ITV is most often generated from respiratory 4DCT [15,23], where the cardiac motion simply presents as a blurring in the respiratory-binned images.This is generally acceptable when cardiac-induced motion has minimal dosimetric impact due to the smaller amplitude compared to respiratory motion.However, for some scenarios with treatment targets that have significant cardiacinduced motion or are proximal to critical organs at risk (OARs), the motion-encompassing ITV method may be inadequate without the use of cardiac-specific motion margins.Other motion management techniques such as breath holding [24], gating [25,26], and direct tracking [27] have been used to account for respiratory motion, but there are limited investigations of techniques to address cardiac-induced motion [28][29][30].
Stevens et al published a review of target motion in VT patients and found that detailed cardiorespiratory motion data for cardiac radioablation is still limited, despite growing numbers of patients being treated [23].Cardiac-specific ITVs were only reported in one study by Bellec et al, where the increase in target volume due to cardiac contraction ranged from 39% to 75% in 4 patients [31].Stevens et al also concluded that cardiorespiratory motion is highly patient-specific and individual motion management strategies are recommended.Investigation of motion management for cardiac-induced motion is warranted.
Cardiac gating of a radiosurgery treatment, where the radiation beam is only turned on during the quiescent interval (diastole) of the cardiac cycle when heart motion is minimal, would help improve dose conformity and minimize the dose delivered to surrounding normal tissue [28,29,32].The higher complexity, higher frequency, and lower amplitude of cardiac motion compared to respiratory motion makes cardiac-gated radiosurgery a more challenging task, but one that would be worth pursuing if the target area has a large cardiac motion component or is close to an OAR.
In this work, cardiac magnetic resonance (CMR) cine images were used to assess the potential advantages of a cardiac-gated radiosurgery treatment, in terms of treatment target volume reduction.Left ventricle (LV) segment-specific motion and treatment areas were compared for non-gated and cardiac-gated scenarios.The aim of the study was to outline the varying levels of potential benefit from cardiac gating in different regions of the heart, providing an overview of areas where a gating technique may or may not be necessary in cardiac radiosurgery.This study focuses on target motion-related effects and the potential for reduced margins, but does not include an evaluation of the potential dosimetric advantages of cardiacgated radiosurgery.

Methods
CMR images and LV epicardial and endocardial tracked contour points were analyzed for 50 controls (25 healthy volunteers and 25 at-risk for heart failure, all with normal cardiac systolic function) and 50 heart failure patients with reduced ejection fraction (HFrEF, EF < 40% with mixed etiology).The at-risk patients had at least one of hypertension, diabetes, obesity, atrial fibrillation, chronic coronary artery disease, or chronic obstructive pulmonary disease with no prior diagnosis of HF.CMR analysis was focused on the HFrEF group as this population is the most representative of VT patients who undergo cardiac radiosurgery [13,16,22].Subjects were chosen from the Alberta HEART study [33] (originally recruited between 2010 and 2014, approved by the Health Research Ethics Boards at the University of Alberta, Covenant Health and the University of Calgary, ClinicalTrials.govNCT02052804) by selecting the first 25 males and first 25 females for the control group, and the 18 available females and first 32 males for the HFrEF group.
Balanced steady state free precession (bSSFP) cine CMR images were acquired at 1.5 T on Siemens Sonata or Avanto scanners (Siemens Healthcare, Erlangen, Germany).Scanning parameters were typically: repetition time/echo time (TR/TE) 2.8/1.4 ms, 50-70-degree flip angle, 8 mm slice thickness, ECG gating with an 8-12 s breath-hold per slice, and 25 or 30 reconstructed cardiac phases.Epicardial and endocardial LV border points were manually traced on the end-diastolic image frame by an experienced interpreter.The border points were then propagated through the cardiac cycle using in-plane displacement fields, calculated from non-rigid registration of all image frames to the reference end-diastolic frame.A feature tracking approach based on b-spline deformable registration was employed, using the elastix image registration toolbox [34].Further details on CMR data acquisition are provided by Xu et al [35].
Two, three, and four chamber view long-axis slices were used for this analysis.For each slice, 500 equally spaced border points were generated from the original points using spline interpolation, then divided into segments according to the 17-segment model for the LV [36].Points were automatically assigned to segments based on the slice view and position of each point relative to the LV central axis.A detailed description of the segmentation process using in-house written Python code is included in the supplementary materials.
To assess the effectiveness of a cardiac-gated radiosurgery technique, each slice segment (defined by the divided contour points) was treated as a hypothetical treatment target.The centroid of each segment was tracked to determine the optimum treatment window where cardiac-induced motion was minimal during a fifth of the cardiac cycle.Once the optimum window was defined, the maximum centroid displacement was determined for the full cardiac cycle and for the treatment window.The treatment area, defined by the range of motion for the segment border points, was also compared for the full cardiac cycle and the treatment window.A concave hull (alpha shape) method was used to create the bounding polygon defining the treatment area, using the collection of border point positions over the selected cardiac phases (either the full cardiac cycle or the treatment window) as input points [37].Figure 1 shows an example of this analysis for a basal segment.For each patient, centroid displacements and treatment areas were determined for each of the 17 LV segments [36], and the reductions in displacement and area due to gating were quantified.The starting cardiac phase for each optimal treatment window was also recorded to determine the most common gating windows.Mean centroid displacement and treatment area values were then determined for the control and HFrEF groups.
Results are presented as mean ± standard deviation.Comparisons between gated and non-gated values and between subject groups were done using the two-sample t-test with Holm-Bonferroni correction where appropriate.A corrected P-value less than 0.05 was considered to be statistically significant.Sample size was chosen to maximize the number of HFrEF patients from the available data, which was acquired from a previous study, along with an equal control group size.Power analysis (two-tail t-test, difference between 2 independent means) for a large effect size of 0.8, alpha = 0.05, power = 0.95, and equal group size determines a sample size of 42 per group, and this study includes 50 subjects per group.
There was no difference between the groups for the gating window centroid displacements.The gating window treatment areas for the HFrEF group were larger than the controls for segments 1 (basal anterior), 6 (basal anterolateral), 7 (mid anterior), 9 (mid inferoseptal), 12 (mid anterolateral), and 14 to 16 (apical septal/inferior/lateral).The mean gated treatment areas for the rest of the segments were also larger for the HFrEF group, although not statistically significant.
Of note, segments 2 and 3 (basal septal) overall centroid displacements were significantly smaller than the other basal segments (anterior, inferior, and lateral) for both the control and HFrEF groups (p < 0.001 for all).Likewise, the basal septal centroid displacement reduction due to gating was also smaller than in other basal segments for control and HFrEF groups (p < 0.001 for all).The basal septal segment non-gated treatment areas and treatment area reductions were also significantly smaller than other basal segments for both groups (p < 0.001 for all).
The minimal motion treatment window for the controls most commonly began in the 55% to 70% cardiac phases, corresponding to the quiescent interval during diastole (figure 5).A small spike is also seen near the 20% to 30% phases during systole.The mean beam-on cardiac phase was 61 ± 14% for the controls.For the HFrEF group, the 55% to 70% diastolic phases were also the predominant starting phases for the optimal window, but a smaller proportion compared to the controls.The optimal beam-on phases for the HFrEF group were more spread out overall with a mean of 55 ± 21%.

Discussion
A cardiac gating method would be effective for sparing adjacent healthy cardiac tissue and nearby OARs in cardiac radiosurgery treatments, but its suitability will depend on patient characteristics and the area of the heart to be treated.The present study investigates the potential advantage of cardiac gating for specific LV segments, outlining the typical motions that are seen in VT patient hearts.The dosimetric impacts of cardiac gating in radiosurgery have not yet been evaluated-this will be investigated in future treatment planning studies, potentially using target and OAR volumes from VT patients.A recent editorial stated that there is no simple answer regarding how precise cardiac radiosurgery must be, as the required precision and safety margins depend on clinical goals, target characteristics, and other unknowns [38].Targets that have large cardiac-induced motion and are located close to critical OARs-such as the stomach, esophagus, and proximal bronchial tree-would likely benefit from cardiac gating, but further investigation is required to determine the potential dosimetric benefit and the scenarios for which such a technique is truly necessary.
The feasibility of a cardiac-gated beam delivery on a Varian TrueBeam linear accelerator was first investigated by Poon et al, where a volumetric modulated arc therapy (VMAT) plan was synchronized to an ECG signal and delivery accuracy was verified by using film to compare gated and non-gated dose distributions [28].Beam timing was controlled using multi-leaf collimator (MLC) leaf motion, and the synchronized plan was created pre-delivery due to the requirements of the clinical system.In a study by Reis and Robar, simulated cardiac gating using real-time ECG signal acquisition was demonstrated using a microcontroller interface to directly control the beam delivery of a Varian Clinac iX linear accelerator [29].Dose linearity, beam flatness, and symmetry were not strongly affected by cardiac gating, and acceptable agreement between gated and non-gated dose distributions was shown.
In another study by Akdag et al, cardiac gating, in addition to respiratory motion tracking, has also been demonstrated on an MR-linac [32].An IMRT plan was created for a Quasar motion phantom, and a film insert was moved with combined artificial (10 mm peak-to-peak) or subject-derived (15.3 mm peak-topeak) cardiac motion and respiratory motion (20 mm peak-to-peak).Real-time cardiorespiratory motion was estimated from MRI using rigid registration-the cardiac component of the motion was then used for beam gating, while the respiratory component was used for MLC-tracking.A reasonable duty cycle of 57% was achieved, and film analysis showed excellent agreement with the planned dose for a static reference delivery.The dose area histograms also showed the effect of cardiac motion on target coverage: without cardiac gating, the near minimum dose (D 98% ) decreased by 9.8% (artificial motion) or 14.3% (subject-derived motion).With cardiac gating, minimum dose was decreased by 1.1% without cardiac motion prediction and 2.5% with cardiac motion prediction.Monte Carlo simulations conducted in the same study also confirmed that a higher dose conformality can be achieved through cardiac gating-a dose loss of up to 3.7% (GTV D 98% ) and 3.8% (PTV D 95% ) due to cardiac motion could be prevented by limiting radiation to a pre-defined cardiac phase.However, the present study has shown that most segments for HFrEF patients exhibit less than 10 mm of cardiac-induced motion (figure 2(a)).The dosimetric effect of cardiacinduced motion may be minimal in many cases, and the usefulness of cardiac gating should be evaluated on a patient-by-patient basis.
Stevens et al [39] also developed a framework using a digital phantom to simulate cardiorespiratory motion in combination with different motion management strategies to evaluate the effects of cardiac target motion in radiosurgery.Using a 4-dimensional (4D) extended cardiac-torso (XCAT) phantom [40] expanded with the 17-segment LV model [36], target displacement was evaluated using motion-encompassing envelopes.Similar to the present study, each segment was regarded as an individual target volume.The simulated cardiac-induced motion resulted in maximum segment centroid motion of 4 mm superior-inferior, 6 mm posterior-anterior, and 3.5 mm left-right.The cardiac-induced motion resulted in a motion envelope increase of 49% (24%-79%), while inclusion of respiratory motion resulted in a motion envelope increase of 126% (79%-167%).The study showed that applying respiratory motion management (breath-hold and respiratory-gating) led to a decrease in the motion envelope volumes, but states that the possible gain from motion management depends on the target location and amount of motion within a patient.
The present study also emphasizes that the level of benefit provided by cardiac-gated radiosurgery varies depending on the treated area of the LV.Irradiating during an optimal treatment window of the cardiac cycle would be more beneficial when the target is closer to the basal LV, where cardiac-induced motion is greatest, compared to apical segments.The decreased motion of the basal septal segments is understandable as this is the region where the heart is anatomically fixed to other structures in the thorax/mediastinum and serves as a fulcrum.Cardiac gating for targets in these segments would be less useful compared to the other basal segments.Cardiac gating would be particularly advantageous in certain LV segments that are close to organs at risk (OARs), such as the esophagus, stomach, chest wall, great vessels, liver, spinal cord, and lungs.Gastropericardial and esophageal fistulas are a couple examples of late complications that have occurred after cardiac radiosurgery [41,42].The stomach-heart distance in arrhythmia patients was measured to be 7.5 ± 4.2 mm during end-systole and 6.1 ± 3.6 mm during end-diastole, where LV segment 10 (mid inferior) was the closest [43].The data in the present study suggests cardiac gating reduces HFrEF patient segment 10 centroid displacement from 7.1 ± 2.0 mm to 1.0 ± 0.4 mm and the treatment area is reduced by 35%.In this example, cardiac gating would maximize the distance between the irradiated target and the stomach and reduce the risk of gastropericardial fistula.However, a notable limitation of this study is that the positions of OARs relative to LV segments were not tracked during the cardiac cycle.Overall motion of the heart may change the distance between certain LV segments and OARs, and could increase or decrease the benefit of cardiac gating.
As seen in figure 5, the optimal treatment window most commonly began in the 55% to 70% cardiac phases during diastole, and a smaller proportion began during the quiescent interval associated with systole.Compared to the controls, the HFrEF group optimal phases saw the same general trend but were more spread out over the cardiac cycle.This could be due to the nature of impaired heart motion, where reduced contraction and slowed relaxation leads to a less well-defined quiescent interval [44].Some of the motion trends seen may also be explained by impaired heart motionalthough overall centroid displacement was smaller for the HFrEF group in all segments (figure 2(a)), the gated treatment areas were larger in each segment compared to the controls (figure 3(a)).Overall motion may be decreased in HFrEF patient hearts, but the slower contraction and filling of the LV results in a larger gated treatment area and a slightly reduced relative reduction in target size (figure 4).However, cardiac gating would still lead to significant improvement in dose conformity for HFrEF patients, becoming more important in more dynamic hearts.
A limitation of this study is the accuracy of the non-rigid registration used to propagate the LV contours over the cardiac cycle.Although the initial tracings on the reference frame were done by an experienced interpreter, the image registration process introduces additional uncertainty.This analysis was also limited to in-plane measurements using the longaxis slices, treating the LV segments as hypothetical targets and using treatment area as a surrogate for volume.Future work will look at similar measurements using full volume data from VT patients and will include contouring of surrounding OARs to assess their relative positions.However, motion assessment for patients with ICDs may be challenging in some cases due to susceptibility artifacts, and will depend on the proximity of the ICD [45].The anterior and apical LV segments are particularly affected by artifacts for patients with left-sided ICDs [46].
This work was based on breath-hold CMR scans to investigate heart motion for the purpose of cardiac gating, and respiratory motion is assumed to be accounted for separately during treatment.Deep inspiration breath hold (DIBH) has been used for radiosurgery treatment of the inferior LV to reduce gastrointestinal dose [24], but a dual-gated approach or direct tracking of respiratory motion may be more feasible in most cases.For these free-breathing treatments, the effect of respiration on heart deformation and contraction motion should be considered Histogram showing the optimal beam on phase for controls (blue) and HFrEF patients (orange).The 55% to 70% phases during the diastolic quiescent interval were the most common, more so in the control group compared to the HFrEF group.[47][48][49][50][51][52].The inclusion of respiratory motion in the analysis, especially for a future planning study, is important to understand the true value of the potential reduced margins from cardiac gating.Ultimately, the goal of a gating technique is to reduce the dose delivered to surrounding OARs to minimize the risk of adverse effects.The interplay between cardiacinduced and respiratory-induced motion needs to be considered for the different potential combinations of motion compensation to accurately assess the dosimetric effects of cardiac gating.

Conclusion
This data demonstrates that cardiac cycle motion is an important component of overall target motion and varies depending on the anatomic cardiac segment.Accounting for cardiac cycle motion, through cardiac gating, has the potential to significantly reduce treatment volumes for cardiac radiosurgery.
The results of this study provide an estimate of the reductions in target size for the 17 LV segments due to cardiac gating, giving an overview of the potential benefits in different heart regions.In general, cardiac gating would be more beneficial in the basal segments, where cardiac-induced motion is larger.There would be less benefit for the basal septal segments, however, due to the suppressed cardiac-induced motion in these locations.Ultimately, the potential use of a cardiac gating technique will need to be evaluated on a patient-specific basis, as cardiac-induced motion can vary greatly between patients.
Evaluation of the dosimetric effect of cardiac gating is still required and is a priority for future work.Detailed analysis is required involving the relative locations of radiosurgery targets and OARs, including the effect of both cardiac and respiratory motion, for a thorough evaluation of cardiac gating in different areas of the heart.

Figure 1 .
Figure 1.Example of the minimum motion treatment window for segment 6 (basal anterolateral) of a HFrEF patient.The target outline and centroid are shown on the left image.The green shaded area shows the gated treatment area, while the red shaded area shows the treatment area for the full cardiac cycle.The plot on the right shows the centroid displacement in the left-right, posterioranterior, and superior-inferior directions, where the shaded area shows the optimal treatment window.

Figure 2 .
Figure 2. Comparison between full cardiac cycle and treatment window centroid displacements for the 17 LV segments (a).Centroid displacement reduction due to treating during the optimal window is shown in (b).

Figure 3 .
Figure 3.Comparison between full cardiac cycle and gating window treatment areas for the 17 LV segments (a).The treatment area reduction due to treating during the optimal window is shown in (b).

Figure 4 .
Figure 4. Relative treatment area reduction as a result of using the minimal motion treatment window for the 17 LV segments.

Figure 5 .
Figure 5.Histogram showing the optimal beam on phase for controls (blue) and HFrEF patients (orange).The 55% to 70% phases during the diastolic quiescent interval were the most common, more so in the control group compared to the HFrEF group.