A Motion Analysis of Cardiac Substructures for Guiding Stereotactic Arrhythmia Radiotherapy Motion Management

Abstract

Background:

Stereotactic arrhythmia radiotherapy (STAR) has recently emerged as a novel noninvasive treatment option for critically ill and drug-refractory VT patients who cannot be treated or re-treated by catheter ablation. However, STAR requires precise management of cardiorespiratory motion to minimize radiation toxicity to the healthy heart and nearby organs-at-risk (OARs) and to ensure the radiation precision to deliver the prescribed dose to the VT target.

Purpose:

To investigate the characteristics of cardiac motion of VT patients to facilitate cardiac motion management for STAR treatments.

Methods:

Breath-hold cardiac 4DCTs (c4DCT) of 18 patients previously treated with STAR were analyzed retrospectively. Each c4DCT contained ten 3DCTs corresponding to ten phases of an entire cardiac motion cycle. For each c4DCT, a group-wise deformable image registration (DIR) method was used to register all ten 3DCTs, resulting in ten 3D deformation vector fields (DVFs) and an average position 3DCT. The DVFs were computed from each phase to the average position 3DCT instead of between pairs of phases. Metal artifacts caused by the Implantable Cardioverter-Defibrillator (ICD) leads were reduced using a diffusion procedure before DIR. The heart chambers were segmented on the average position 3DCT using an AI segmentation tool followed by manual evaluation and correction. Cardiac motion characteristics (magnitude and direction) were investigated temporally over the ten cardiac phases and spatially for the outer myocardium walls of the heart and per chamber, AV (atrium-ventricle) valves, septa (the muscular walls dividing the chambers), and STAR targets.

Results:

The motion magnitude maximum over cardiac phases was computed by taking each voxel’s maximum motion magnitude in 10 phases, referring to the average position. The cardiac motion magnitudes were found to be strongly patient-specific. The maximum, 99^th percentile, and 95^th percentile of the motion maximum for the myocardium wall among patients ranged from 8.7 to 17.8 mm, 4.9 to 11.9 mm, and 2.9 to 8.9 mm, respectively. The same metrics for STAR target motion ranged from 2.9 mm to 11.6 mm, 2.6 to 6.4 mm, and 2.4 to 6.0 mm, respectively. Z-tests showed a significant difference between the maximum motion magnitude of the myocardium and the STAR target, while there was no significant difference between the mean motion magnitude of the myocardium and the STAR target. The 95^th percentiles of the myocardium motion magnitudes were < 5 mm for 16 out of 18 patients. The largest motion appeared near the ventricle-atrium (AV) valves. All cardiac structures have their largest displacements from the average position at the end of systole and diastole.

Conclusions:

The cardiac motion characteristics (magnitude, direction, and spatial distribution) of STAR patients provided in this study would further facilitate cardiac motion management.

1. Introduction

Multiple clinical studies^1–6 conducted recently indicated that Stereotactic Arrhythmia Radiotherapy (STAR)⁷ is a promising modality for treating drug-refractory and recurrent VT effectively. STAR is noninvasive, fast (<25 minutes radiation delivery time), and can treat arrhythmogenic tissues that cannot be reached by catheter ablation^7–9.

However, STAR also has limitations. The treatment target is mobile due to the complex cardiorespiratory motion of the heart, and the radiation beam cannot follow the target motion. Hence, to ensure the VT target receives the prescribed dose, the target volume (TV), ~20 cc on average, is commonly expanded to form the internal target volume (ITV), ~35 cc on average, to cover the cardiorespiratory motion. ITV is then expanded by a 3 to 5 mm margin to form the planning target volume (PTV), ~80 cc on average, to account for patient setup and radiation delivery uncertainties¹⁰. Therefore, healthy cardiac tissues around the VT target will also be irradiated during the treatment, thus increasing the risk of cardiac radiotoxicity^11–15.

One key issue for delivering the prescribed dose to the VT target while minimizing radiotoxicity is to quantify the patient’s cardiac motion accurately so that the TV to ITV expansion can be managed precisely and the healthy tissue included in the ITV and PTV can be minimized. To achieve that, the cardiac motion characteristics of patients should be quantitatively analyzed.

A previous study¹⁶ evaluated the cardiac motion of 10 STAR patients using a pair-wise image registration method to compute the displacements of CTV, left ventricle (LV), and ICD lead. It concluded that the study could achieve patient-specific cardiac ITV margin estimation. The study computed tissue motion using the mean squared error (MSE) and Pearson correlation coefficient (PCC) between the moving/registered and target images. However, MSE and PCC values are normally only good for qualitative evaluation, not good enough for a quantitative evaluation for which the target-registration-error (TRE) should be computed against ground-truth motion. Additionally, the motion of other heart chambers was not computed and analyzed, and the number of patients was relatively small.

In this study, we developed an image-processing procedure to investigate the cardiac motion characteristics of the heart and applied the method to 18 STAR patient cases. To compute the cardiac motion for heart chambers and VT targets on the patient’s c4DCTs, a group-wise deformable image registration (DIR) method¹⁷ was applied to improve the computation accuracy. A metal-artifacts reduction method was developed to reduce the metal artifacts caused by patients’ ICD (Implantable Cardioverter-Defibrillator) leads. The cardiac motion was quantitatively analyzed spatially for the outer myocardium walls of the heart and per chamber, AV (atrium-ventricle) valves, septa (the muscular walls dividing the chambers), and STAR targets, as well as temporally between the 10 cardiac phases. The goal was to provide quantitative and patient-specific motion information of the heart and treatment target to support motion management for STAR treatments and mitigate cardiac radiotoxicity.

2. Methods

2.1. Dataset

The breath-hold cardiac 4DCTs (c4DCT) of 18 STAR patients were analyzed in this study. All 18 patients were previously treated at Washington University in Saint Louis before 2022 to 25 Gy in a single fraction under whole-body immobilization and abdominal compression^3,7. Patients’ DICOM files were retrospectively acquired with IRB approval and de-identified before being studied. The original image in-plane pixel sizes were from 0.47 to 0.68 mm, and the slice thicknesses were 1.5 mm. We resampled all images to 1.5 mm isotropic voxel sizes before processing them in the following steps. More details about the STAR treatment simulation, 4DCT reconstruction, target definition, treatment planning, motion management, and radiation delivery can be found in published papers^3,7,18. Figure 1 shows the overall data processing workflow. Each step will be discussed in detail in the later sections.

Figure 1.

The data processing workflow. A group-wise deformable image registration was performed on the 3DCTs of all 10 c4DCT phases to compute 4D cardiac motion after the ICD lead metal artifacts were reduced. The results were ten 3D deformation vector fields (DVF) and an average position 3DCT. The whole heart and heart chambers were segmented on the average position 3DCT.

2.2. ICD lead metal artifact reduction

The c4DCT of each patient had strong metal artifacts caused by the ICD leads. The artifacts would significantly affect the cardiac motion computation using DIR methods because the image registration methods operated under the assumption of consistent image appearance and matching voxel intensities. The artifacts would also interfere with the consistency of heart and chamber segmentation. We developed an ICD lead artifact reduction method. The overall idea was that, after the artifacts were manually segmented, the segmented areas were filled automatically by smoothly diffusing image voxel values from the area edges into the areas.

2.3. Group-wise deformable image registration

The ICD artifact-reduced images were registered using the group-wise DIR method of pTVreg (parametric Total Variation registration), which was previously shown to be accurate for lung 4DCTs¹⁷. The results of the group-wise DIR were the average position 3DCT image and ten DVFs corresponding to each of the ten phases of the c4DCT. Each DVF contained information on how the corresponding 3DCT deformed to the average position image.

Although pTVreg has been proven accurate for lung 4DCTs, it was necessary to assess its accuracy for computing cardiac motion in this study due to different anatomical structures and the effective image contrast. The 4D extended cardiac-torso (XCAT) digital phantom¹⁹ was used to assess the general registration accuracy. DVFs were computed on 10 cardiac phases using pTVreg under group- and pair-wise settings. The target registration errors (TRE) were calculated by 1) translating the XCAT phantom control points of the myocardia of each phase by the DVF of the phase and 2) measuring the absolute error between the ten translated control points and the ground truth control point positions.

2.4. Quantitative analysis of the cardiac motion for STAR patient cases

To investigate the cardiac motion characteristics of each patient, the computed DVFs, i.e., the motion relative to the average position image, were analyzed across the 10 c4DCT phases for myocardial walls and STAR treatment targets. The LV, left atrium (LA), right ventricle (RV), and right atrium (RA) were already segmented on the average position image by an AI segmentation tool “TotalSegmentator”²⁰. The treatment target contour was transferred from the planning CT to the average position image. This was achieved by first registering the plan CT to the average position CT using pTVreg¹⁷ and then using the resulting DVF to transfer the target contour in plan CT to the average position CT.

The DVF magnitudes’ mean, maximum, 99^th percentile, 95^th percentile values, and the spatial distribution were calculated for all voxels of all phases in the heart myocardium wall and the target regions. The motion directions of the STAR target were further investigated. The purpose was to understand the spatial distributions of cardiac motion and determine the location-dependent motion margins for treatment planning. The cardiac characteristics drawn from the patients could be useful for determining the motion margin for STAR patients and the procedure to manage STAR target motion depending on the target location.

3. Results

The overall DIR accuracy was evaluated using the XCAT digital phantom. The mean TREs computed by group-wise pTVreg were 0.96 ± 0.30 mm and 1.08 ± 0.36 mm for control points on the myocardium within the heart and on the heart surface, respectively. The mean TREs computed by pair-wise pTVreg were 1.66 ± 0.96 mm and 2.23 ± 1.31 mm for phase 3 (the closest to the average position image) and phase 6 (the farthest from the average position image) as the target image, respectively. These results confirmed that group-wise pTVreg improved the computation accuracy for cardiac motion registration.

Table 1 shows the maximum, 99^th-percentile, 95^th-percentile, and mean values of the heart myocardium and treatment target motion magnitudes in mm, target locations, and volumes for all STAR patient cases evaluated in this study. All motion information in Table 1 was related to the average position CT. The “Max of Max”, “99% and 95% of Max”, and “Mean and Std of Max” values were computed by first calculating the maximum values across ten phases for each voxel then taking the maximum, 99^th and 95^th percentile, average, and standard deviation values in the myocardium wall or target volumes. The “95% of All” was computed by acquiring the motion magnitudes of all voxels in the myocardium of all phases and taking the 95^th percentile of them. The “Mean, Std of Mean” values were computed by calculating the mean values across ten phases for each voxel and then taking the mean value in the total volume (myocardium wall or target).

Table 1.

List of the maximum, 99^th percentile, 95^th percentile, and average mean values of the heart myocardium and treatment target motion magnitudes (in mm), and the target volumes and locations, for all patients studied. All motion information in the table was relative to the average position CT.

#	Myocardium (mm)					STAR Target (mm)
	Max of Max	95% of All	99% and 95% of Max	Mean, Std of Max	Mean, Std of Mean	Max of Max	99% and 95% of Max	Mean, Std of Max	Mean, Std of Mean	Vol (cm3)	Location in AHA 17 segments model
1	9.9	4.3	7.8, 6.4	3.1±1.5	2.0±0.8	4.0	3.2, 2.7	2.0±0.4	1.4±0.2	43.7	8,9,14,15
2	12.4	3.9	9.5, 5.8	3.0±1.6	1.9±0.8	5.1	4.3, 3.9	2.7±0.7	1.8±0.3	43.5	1,2,3,7,8,9
3	15.6	4.8	10.4, 7.2	3.7±1.9	2.2±0.9	6.5	5.6, 5.2	3.1±1.1	1.8±0.5	73.5	4,5,10,11,15,16
4	13.5	2.5	7.4, 4.1	2.1±1.2	1.4±0.6	7.0	4.9, 3.7	2.1±0.8	1.4±0.3	72.9	4,5,10,11,12
5	12.6	2.9	6.3, 4.5	2.3±1.1	1.5±0.4	11.6	4.8, 3.2	2.0±0.7	1.4±0.2	58.6	4,9,10
6	12.7	3.0	7.9, 4.6	2.3±1.3	1.6±0.6	5.2	4.3, 3.7	2.3±0.8	1.6±0.3	38.3	9,10,15
7	10.1	3.1	6.9, 4.9	2.1±1.2	1.5±0.7	3.0	2.6, 2.4	1.9±0.3	1.5±0.2	15.2	2,3,8
8	11.7	2.1	6.9, 2.9	1.8±1.0	1.4±0.5	3.9	2.7, 2.4	1.8±0.3	1.5±0.2	44.9	1,2,3,8,9
9	9.0	3.0	6.4, 4.2	2.3±1.0	1.6±0.6	4.2	3.9, 3.6	2.8±0.5	2.0±0.3	15.1	1,2,6
10	17.7	5.2	11.2, 8.9	3.4±2.2	2.1±1.2	7.9	5.9, 4.4	3.2±0.7	2.1±0.3	11.8	2,3
11	17.8	4.2	11.9, 7.1	2.9±2.2	1.8±0.9	5.0	4.0, 3.6	2.6±0.5	1.6±0.2	11.7	5,6
12	9.9	2.7	4.9, 3.8	1.9±0.9	1.4±0.4	4.2	3.8, 2.8	1.5±0.5	1.2±0.3	16.3	2,3
13	15.8	6.0	11.1, 8.4	4.6±2.0	2.8±1.1	7.6	6.3, 5.6	4.6±0.5	3.1±0.3	5.0	2
14	8.7	3.3	6.4, 5.0	2.1±1.2	1.5±0.6	7.5	6.4, 6.0	3.9±1.3	2.4±0.8	17.8	1,6,7,12
15	10.0	3.1	5.4, 4.2	2.3±1.0	1.6±0.5	4.2	3.6, 3.2	2.1±0.7	1.4±0.3	18.5	3,4,9
16	14.9	3.5	8.6, 5.4	2.6±1.5	1.6±0.6	2.9	2.8, 2.6	2.0±0.5	1.3±0.2	13.5	15,16,17
17	12.3	2.5	6.1, 3.9	1.9±1.0	1.3±0.5	3.5	3.1, 2.7	1.9±0.4	1.4±0.2	16.4	1,6
18	13.4	4.3	8.4, 6.5	3.1±1.6	1.9±0.8	8.9	6.3, 5.5	4.1±0.9	2.3±0.3	5.9	1,6

Z-tests showed that there was a significant difference between the myocardium and STAR target data for “Max of Max” and “99% and 95% of Max”, while there was no significant difference between the myocardium and STAR target data for “Mean of Max” and “Mean of Mean”. It is important to note that the reported values in this study were the motion to the average position instead of the peak-to-peak motion reported in other studies^18,21, and the values reported in this study were closer to half of the peak-to-peak motion values.

STAR target volumes, computed as the shared volume between the myocardium and the STAR target, ranged from 5.0 cm³ to 73.5 cm³ for all patients. The maximum motion magnitudes ranged from 8.7 mm to 17.8 mm for the myocardium and 2.9 mm to 11.6 mm for the STAR targets. Table 1 indicated that the motion magnitude ranges of the hearts and targets were very patient-specific, partially due to the target position and size differences.

Table 2 presents the 99^th percentile of maximum motion magnitudes (relative to the average position CT) for the ventricular and atrial septa, AV valves, and outer surface myocardium. These values were determined by calculating the maximum motion across ten cardiac phases for each myocardial voxel, followed by deriving the 99th percentile of the maximum motion within each substructure. The results demonstrated that the right AV valves (RA-RV) exhibited a range of 26.9% to 240.7% greater motion compared to the left AV valves (LV-LA). Similarly, the right ventricular surface myocardium (RV-surf) displayed 4.8% to 214.3% larger motion than the left ventricular surface myocardium (LV-surf), except for Patient 12. The right atrial surface myocardium (RA-surf) also showed 28.6% to 307.7% higher motion magnitudes relative to the left atrial surface myocardium (LA-surf), excluding Patient 12. Statistical analysis using a Z-test confirmed that the right side of the myocardium surface generally had significantly larger motion than the left side.

Table 2.

List of the 99^th percentile of the maximum motion magnitudes (in mm) of the ventricular and atrial septa (LV-RV, RA-LA), AV valves (LV-LA, RA-RV), and outer surface myocardium (LV, RV, RA, LA-surf) for all patients.

#	LV-RV septa	LV-LA AV valve	RA-LA septa	RA-RV AV valve	LV-surf	RV-surf	RA-surf	LA-surf
1	4.5	6.7	4.1	8.5	7.2	8.7	8.5	5.7
2	4.3	5.8	4.0	10.0	7.0	11.5	10.1	4.3
3	5.7	8.6	6.2	11.8	8.1	10.8	13.5	7.4
4	3.3	4.9	3.2	6.6	4.0	9.4	11.2	3.0
5	4.6	5.6	5.5	10.2	5.3	6.1	6.6	2.7
6	4.0	5.2	3.9	7.4	4.7	7.9	11.8	3.7
7	3.8	2.9	2.7	6.7	2.9	7.0	9.1	2.7
8	3.4	2.7	2.6	9.2	2.8	8.8	10.6	2.6
9	4.2	4.9	6.2	6.4	3.7	6.7	7.9	4.8
10	4.8	5.3	6.3	10.6	4.4	11.3	13.4	4.9
11	6.4	5.5	7.6	10.0	4.3	10.8	16.1	6.5
12	3.7	3.9	2.7	5.4	5.3	5.3	3.5	3.7
13	6.7	8.9	8.1	11.3	7.2	12.2	13.6	9.0
14	3.0	4.9	3.1	8.0	6.2	6.5	6.7	4.9
15	3.9	3.8	3.1	6.3	5.1	6.3	6.4	3.5
16	4.7	5.0	7.1	12.8	4.9	10.6	9.1	4.7
17	4.6	2.9	3.3	5.8	4.5	8.5	4.5	3.5
18	4.9	5.6	6.0	10.8	4.6	8.6	10.2	6.9

Figure 2(a) shows the normalized mean motion magnitude of the myocardium substructures, myocardium, and STAR targets of each phase, averaged over all patients. The values of data points in this figure were computed by the following equation:

$$v(s,ph)=\frac{1}{N}\sum _{p=1:N}\frac{M_{s,ph,p}}{{MM}_p}$$

where $M_{s,ph,p}$ represents the mean motion magnitude (relative to the average position) within the structure $s$ of the phase $ph$ of Patient $p,{MM}_p$ was the mean motion magnitude (relative to the average position) within the myocardium of ten phases for Patient $p$ and then taking the maximum value for the patient, $N$ = the number of all patients.

Figure 2.

(a) Normalized mean motion magnitude of the ventricular and atrial septa, AV valves, outer surface myocardium, the whole myocardium, and the target of each phase of all patients. (b) The normalized maximum motion magnitude of myocardium voxels versus distance to AV valves for all patients. The red curve is the Gaussian fitted 95^th percentile curve.

Figure 2(a) demonstrates that the right AV valve (RA-RV) had the largest motion among all the cardiac structures studied. RA-RV, right ventricular surface myocardium (RV-surf), and right atrial surface myocardium (RA-surf) had larger motion than their left side counterparts. Figure 2(a) also shows that all cardiac structures have their largest displacements from the average position at phases 3 to 4 (at the end of systole), and phases 9 to 10 (at the end of diastole).

Figure 2(b) shows the relationship between the normalized maximum motion magnitudes in the myocardium per voxel and the distance from the voxel to the AV valves. Each data point was computed by normalizing the maximum motion magnitude of the voxel to the maximum motion of the AV valves of the corresponding patient. Figure 2(b) indicates that voxels with large motion were more likely to be located near (within 20 mm) the AV valves.

Figure 3 shows the motion magnitude on the 3D surfaces of cardiac chambers and targets for three randomly selected patients. The color assigned to the surfaces was the maximum motion magnitude of the 10 cardiac phases. The color bars were set to have the maximum value of 6 mm to highlight the area with large motions on the heart surface. The targets are mostly in LV, which was common for STAR patients. Patients 7 and 9 were chosen because most patients in this study shared similar motion spatial distribution features. Patient 1 was chosen because it is relatively different from most other patients. The VT target for Patient 1 was farther away from the AV valves and thus had a smaller but more complex motion.

Figure 3.

The motion of cardiac chambers and targets in the respective average position images for the three randomly selected patients. Top row: 3D surfaces of cardiac chambers with the surface motion magnitude displayed in a colormap. Middle row: 3D surfaces of cardiac chambers with treatment targets. The red pentagrams are the volumetric centroids of the targets. Bottom row: The target motion trajectories were shown in axial view to be consistent with the top and middle rows and present the target motion directions relative to the ventricle-atrium interface. The black curves are average target 3D trajectories. Each dashed curve represents a voxel’s trajectory. Green triangles represent the centroid positions of each phase.

The TV target centroid trajectory’s ranges (furthest distance between any two points on the 3D trajectory) were 2.5 mm, 3.7 mm, and 5.1 mm for these three patients, respectively. The bottom row also shows the smoothed motion trajectories (gray dashed lines) of the voxels in the target, besides the average target trajectories (black lines). The target motion ranges (the largest range of the dashed curves) are 5.6 mm, 5.1 mm, and 6.9 mm for the three patients.

Figure 2(a) and Figure 3 both show that large motion always appeared near the left and right AV valves, while the motion magnitudes of the LV-RV and the LA-RA septa were comparatively small. The observation above was true after inspecting all patients in this study.

For Patients 7 and 9, the target centroid motion magnitudes between adjacent phases were different. As seen in the green triangles in the centroid curve in the bottom row of Figure 3, the centroid positions were close for some phases and further apart for other phases. This suggests that the target motion could be non-uniform among different cardiac phases in a cardiac cycle.

4. Discussion

This study used a group-wise deformable registration method to quantitatively investigate the cardiac motion of the myocardium and treatment targets in the breath-hold cardiac 4DCT images. The AV valves exhibited larger motion for all patients due to their active role in regulating blood flow, while septal regions (LV-RV, LA-RA) showed smaller motion due to their passive structural roles²². All patients in this study showed a larger motion in the right AV valve than in the left AV valve. RV operates under lower-pressure conditions compared to LV, allowing for greater compliance and mobility of the tricuspid valve apparatus^22,23. Cardiac structures on the right side normally had larger motion (displacement magnitude from the average position). The motion magnitude generally diminished away from the AV valves. Such general information could be useful for STAR target motion management.

For the myocardium, Table 1 shows the Max of Max, 99% of Max, and 95% of Max ranged from 8.7 mm to 17.8 mm, 4.9 mm to 11.9 mm, and 2.9 mm to 8.9 mm. For the STAR target, the Max of Max, 99% of Max, and 95% of Max ranged from 2.9 mm to 11.6 mm, 2.6 mm to 6.4 mm, and 2.4 mm to 6.0 mm. Figure 3 also indicates that the motion of different portions (voxels) of the heart myocardium or the STAR targets is different. Thus, a non-uniform motion margin for TV to ITV expansion might be more appropriate than a uniform margin.

For common characteristics, Table 1 shows that the myocardium motion magnitudes relative to the computed average position were below 5 mm (see “95% of All”) for 16 out of 18 patients, and the average target motion (see “Mean, Std of Mean”) was within 2 mm for 13 out of 18 patients. For the myocardium, the Max of Max is 26.9% to 102.0% larger than the 99% of Max and 54.7% to 303.4% larger than the 95% of Max. For the STAR target, the Max of Max is 3.6% to 141.7% larger than the 99% of Max and 11.5% to 262.5% larger than the 95% of Max. The maximum motion magnitudes of the myocardium and target were much larger than the 99th and 95th percentiles of the motion magnitudes for some patients. Hence, the 99^th or 95^th percentile would be more appropriate metrics to determine the target margin instead of the maximum motion if a uniform margin is required to limit unnecessary extra dose to healthy tissue in proximity to the target. This was the reason why we presented the 99^th percentile of the maximum instead of the maximum values in Table 2.

The results of this study are considerably more comprehensive than the cardiac motion part of previous studies^{16,18,24–27}, and would be useful for motion management for the clinical practice of STAR. For example, a previous study²⁷ concluded that a 5 mm margin was needed to account for cardiac motion. However, with the generic characteristics drawn from this study, physicists should increase the margins for target volumes near the AV valves and reduce the margins for target volumes far from the AV valves. Hence, a STAR plan sparing more normal tissues becomes achievable with the procedure presented by this study.

The limitations of this study include: 1) Only 18 patients’ data were used in this study; 2) This study did not yet cover respiratory motion, another important factor for managing cardiac motion for STAR treatments. 3) When comparing the motion between the myocardium and the STAR target, the myocardium included the STAR target. In our future work, we will analyze the cardiac motion in both free-breathing respiratory 4DCTs and breath-hold cardiac 4DCTs to provide a complete solution for managing both respiratory and cardiac motions for STAR treatments.

5. Conclusion

The myocardium motion (“95% of All”) relative to the average position was below 5 mm for 16 out of 18 patients, and the average target motion (“Mean of Mean”) relative to the average position was within 2 mm for 13 out of 18 patients. The highest motion recorded in the myocardium among all patients was 17.8 mm, and the corresponding voxel was near the right AV valve. A few common characteristics were found among patients, such as large motion appearing near the left and right AV valves, which would be useful information to improve generic motion management guidance and STAR target margin design.

Data availability:

Raw and processed data are available upon request.