Distance-adaptive geometric margins for residual rotational uncertainty in single-isocenter multitarget stereotactic radiosurgery Authors: Jiaxin Deng, Zun Piao, Xinyu Song, Danyang Li, Jiang Hu, Shouliang Ding, Yanfei Liu, Li Chen, Guanqun Zhou, Ying Sun, Xiaoyan Huang, Guangyu Wang Published: 2026 ABSTRACT Graphical abstractKeywords: Single-isocenter multitarget, Stereotactic radiosurgery, Error propagation, Distance-adaptive geometric margin Introduction Single-isocenter multitarget stereotactic radiosurgery (SIMT SRS) improves treatment efficiency by using one isocenter and volumetric modulated arc therapy to treat multiple spatially separated lesions, particularly in patients with multiple brain metastases [1], [2], [3]. However, the steep dose gradients that support SRS also make target coverage highly sensitive to geometric uncertainty. Even small translational or rotational deviations can reduce target coverage, especially for small lesions treated with narrow margins [4], [5]. A key uncertainty in SIMT is the distance-dependent amplification of residual rotational uncertainty. Although modern linear accelerators offer high mechanical stability and six degree-of-freedom (6-DoF) image guidance, residual rotational uncertainty remains unavoidable due to mechanical hysteresis, intrafraction motion, and immobilization relaxation [6], [7], [8]. For off-axis targets, small-angle rotations in pitch, roll, or yaw are amplified into larger spatial displacements with increasing distance from the isocenter, creating a leverage effect that preferentially compromises distal target coverage and may increase unintended dose to nearby organs of interest [5], [9], [10]. Despite this recognized risk, many clinical workflows still rely on uniform translational tolerances and fixed rotational thresholds that do not account for distance-dependent uncertainty amplification [11]. Prior experimental, simulation, and analytic studies have explored rotational uncertainty and distance-related margin concepts [4], [5], [12], [13], [14], but a practical framework linking institution-specific rotational performance to target location and margin selection remains needed. In this study, we derived and validated an analytic rigid-body model for rotational error propagation in SIMT. We then used this model to quantify distance-dependent geometric and coverage risk and to develop a Distance-Adaptive Geometric Margin (DAGM) strategy for risk stratification, planning guidance, and quality assurance. Materials and methods The study workflow is summarized in Fig. 1. The workflow consists of four progressive stages. We first derived an analytic rigid-body model for residual rotational uncertainty, then validated it using large-scale Monte Carlo simulations. We further quantified distance-dependent geometric and coverage risk under different rotational precision scenarios and derived a DAGM strategy. Methodological workflow. The study derived a rigid-body analytic model for rotation-induced TRE, validated it against exact matrix calculations using 1,000,000 simulated targets, quantified distance-dependent geometric and coverage risk, and derived the Rayleigh-based DAGM formula. Analytic modeling of rotational variations In SIMT treatments, geometric deviations arise from translational and rotational variations. While translational variations are spatially uniform, rotational variations increase with target-to-isocenter distance and dominate off-axis geometric uncertainty. An analytic formulation is therefore derived to model residual rotational error propagation. Let the nominal target position vector in the planning coordinate system be r. Residual rotation is represented by a small-angle vector θ=(θx,θy,θz), corresponding to pitch, roll, and yaw. Since post-IGRT residual rotations are typically small, a first-order approximation gives the rotation-induced displacement as This cross-product form shows that the displacement is perpendicular to the plane spanned by θ and r, and that its magnitude scales with distance from the isocenter, which explains the leverage-driven amplification of distal-target risk in SIMT. This expression aligns with prior SIMT rotational error modeling and maps residual rotation to distance-dependent equivalent displacement uncertainty [14]. The displacement magnitude satisfies where ϕ is the angle between θ and r. The target registration error (TRE) is defined as TRE=∣Δr∣. For a conservative upper bound, we consider the worst-case configuration with ϕ=90°, which yields with d=r and θ=θ. Because this linear form strictly requires θ to be expressed in radians and may be inaccurate for larger angles, a trigonometric expression is used in subsequent analyses, with θ expressed in radians. The full matrix derivation is provided in Supplementary Section 1. This analytic formulation forms the basis for the statistical modeling of random rotational variations and the derivation of distance-adaptive PTV margins in Section 2.3. Large-scale Monte Carlo simulation validation A virtual cohort of 1,000,000 targets was uniformly sampled within a 200 mm radius sphere to cover clinically relevant SIMT target-to-isocenter distances. Residual rotations about pitch, roll, and yaw were modeled as independent, zero-mean, isotropic Gaussian variables with a common standard deviation σrot, representing post-correction residual rotational uncertainty that can be estimated from institution-specific image-guided radiotherapy (IGRT) or quality-assurance (QA) data. For each target, TRE was calculated using both the exact rotation-matrix method and the analytic expression. Agreement between the two methods was assessed by sample-wise residuals. The simulated TRE distributions were then summarized by target-to-isocenter distance to support risk analysis and margin derivation. DAGM model To construct the DAGM, we used a geometric-containment approach [15], [16], [17] rather than a population dose-blurring approach [18], [19]. The objective was to ensure that, in 95 % of fractions, the target remains geometrically encompassed by the high-dose volume across different target-to-isocenter distances. In the SIMT geometric model, the rotation component parallel to the target radial direction produces no displacement. TRE is therefore determined by the two effective rotational components perpendicular to the radial direction. Under the isotropic Gaussian assumption, the magnitude of this tangential displacement follows a Rayleigh distribution, as shown in Supplementary Section 2. The 95 % Rayleigh coverage factor is k=2.45, as derived and validated in Supplementary Section 3. Based on the above, we propose a distance-adaptive geometric margin for SIMT, termed the DAGM. The total PTV margin is modeled as the sum of a distance-invariant translational component and a distance-dependent rotational component, Using the scalar relationship in Section 2.1 and the 95 % coverage rotation amplitude kσrot, the rotational margin becomes where d is the target-to-isocenter distance, and σrot is the measured residual rotational standard deviation. We used linear addition rather than root-sum-square combination as a conservative geometric-containment choice, because translational and rotation-induced displacements may project in the same direction relative to the local dose gradient. If institution-specific joint translational and rotational error data are available, less conservative root-sum-square or covariance-based formulations may also be considered. Sensitivity analysis scenarios To evaluate sensitivity to residual rotational precision, we used σrot as the control parameter and defined three scenarios: high precision 0.2°, standard clinical precision 0.5°, and low precision or failure 1.0°. These values were selected as sensitivity-analysis scenarios, rather than fixed clinical categories, based on SRS-related tolerances in AAPM TG-142 [20] and frameless intracranial radiosurgery workflows using 6-DoF image guidance or robotic couches [7], [8], [21]. They span strict rotational control to degraded immobilization or machine-performance conditions. The high-precision scenario represents low post-correction residual rotations achievable with strict image guidance, reproducible immobilization, strict correction thresholds, and, when available, a 6-DoF couch. The standard scenario served as the baseline for the main analyses. The low-precision scenario was used to assess distal-target risk when residual rotations exceed usual control levels. Geometric coverage loss (GCL) Geometric coverage loss (GCL) was used as a surrogate for target dose-coverage risk [10]. GCL was defined as the percentage of target volume outside the PTV after displacement, where VGTV is the target volume and VGTV∩VPTV′ is the overlap between the displaced PTV and the original target. A smaller GCL indicates better geometric coverage. A spherical target with a 10 mm radius and a fixed 1 mm PTV expansion was used to evaluate coverage loss across target-to-isocenter distances. Results Numerical validation of the analytic model The analytic approximation showed high agreement with the exact rotation-matrix method across 1,000,000 simulated targets. Absolute residuals were typically at the micrometer scale, and the maximum absolute residual remained below 0.2 mm (σrot=1.0°, Fig. 2C). Given that key clinical uncertainty sources, including treatment planning CT spatial resolution and linac mechanical isocenter tolerances, are generally on the millimeter scale, the linearization error is negligible for geometric uncertainty assessment. Thus, for small residual rotations, the analytic model provides a lower-cost alternative to matrix calculations for rapid and stable estimation of SIMT geometric uncertainty. Validation of the analytic TRE model. (A) TRE versus target-to-isocenter distance under the standard precision scenario σrot=0.5°. (B) Agreement between analytic and exact matrix calculations. (C) Distance-dependent residuals σrot=0.5° across precision scenarios. (D) Residual distribution in the standard scenario. Visualization and quantification of distance-dependent TRE The validated model was used to characterize maximum TRE as a function of target-to-isocenter distance and residual rotation magnitude (Fig. 3). Under the standard clinical scenario with σrot=0.5°, the 1.0 mm contour intersects the distance axis at about 12 cm, indicating that beyond this range the maximum TRE exceeds 1.0 mm. At 20 cm, a residual rotation of 0.5° corresponds to a maximum displacement of about 1.75 mm, suggesting that distance-invariant fixed PTV margins may introduce a structural risk of inadequate coverage for distal targets. The corresponding discrete lookup table is provided in Supplementary Table S1 for rapid clinical use. Clinical reference heat map of maximum TRE. Deterministic worst-case TRE is shown as a function of target-to-isocenter distance and residual rotation magnitude. Contours indicate 1.0 and 2.0 mm reference levels, and the horizontal line marks σrot=0.5°. Zonal statistical distribution of geometric uncertainty Because clinical residual rotations are stochastic, TRE is more appropriately described by a probability distribution rather than a single angle (Supplementary Section 4.2). To quantify distance-dependent risk, we performed a Monte Carlo simulation with σrot=0.5°N=106 and summarized TRE by radial bins using dispersion, quantiles, and exceedance probability. TRE dispersion and tail risk increased with distance. In the 0 to 50 mm bin, uniform volumetric sampling resulted in greater target density near the outer boundary, giving a mean TRE of 0.41 mm but a failure rate of 2.1 %. In the 50 to 100 mm bin, the mean TRE increased to 1.09 mm and the proportion with TRE>1 mm reached 35.2 %. In the 100 to 150 mm bin, the distribution widened further and the exceedance proportion increased to 66.4 %. In the 150 to 200 mm bin, the mean TRE increased to 2.46 ± 1.34 mm, the empirical P95 reached 5.03 mm, and the failure rate reached 80.9 % (Table S2). These zonal statistics are consistent with the scenario-specific zonal failure-rate comparison in Fig. S3, which provides the corresponding failure percentages across the three precision scenarios. Sensitivity analysis of geometric uncertainty across clinical precision scenarios Increasing σrot broadened the TRE distribution and shifted the failure-rate curve toward shorter distances. P95 increased from 1.40 mm at σrot=0.2° to 3.49 mm at σrot=0.5° and to 6.98 mm at σrot=1.0°, indicating that exceedances of a 1 mm reference threshold become much more frequent as precision degrades (Fig. 4A-C). This shift also compresses the safe distance range. In Fig. 4D, failure-rate curves, defined as the proportion with TRE>1.0 mm, increased nonlinearly with distance, and the inflection point moved to shorter distances with increasing σrot. Failure rates increased gradually in the high-precision scenario and were approximately 35 % at 200 mm. The standard-precision scenario showed an accelerated rise beyond about 75 mm. The low-precision scenario rose rapidly and exceeded 50 % at about 50 mm, consistent with unstable off-axis coverage. Overall, geometric risk in SIMT is jointly determined by distance and residual rotational precision, and lowering σrot delays the rise in failure with distance and expands the acceptable distance range. Bin-specific failure rates are provided in Supplementary Section 5.3. Sensitivity of TRE distributions and geometric failure rates. (A–C) TRE distributions under σrot=0.2°, 0.5°, and 1.0°. (D) Failure rate versus target-to-isocenter distance, defined as the proportion with TRE >1 mm. Dose-coverage impact of geometric displacement The clinical relevance of geometric error is determined by its effect on target dose coverage. Under σrot=0.5°, target dose coverage was evaluated using V100% with a pass criterion of V100%≥95% (Table 1). Results for the other precision scenarios are provided in Supplementary Table S3. Table 1Displacement quantiles and dose-coverage risk at different target-to-isocenter distances in the standard-precision scenario σ=0.5°.d(mm)P95(mm)Mean coverage (%)(V100%)P5(%)Coverage pass rate (%)V100%≥95%501.07100.098.8100801.7197.692.588.41002.1495.085.452.11202.5792.276.934.51503.2187.665.212.82004.2879.748.30.0Note: Mean coverage is the expected V100% obtained by statistical analysis of the error distribution under a fixed 1 mm PTV expansion. P5 is the 5th percentile of coverage. The pass rate is the proportion of fractions satisfying V100%≥95%. Displacement quantiles and dose-coverage risk at different target-to-isocenter distances in the standard-precision scenario σ=0.5°. Note: Mean coverage is the expected V100% obtained by statistical analysis of the error distribution under a fixed 1 mm PTV expansion. P5 is the 5th percentile of coverage. The pass rate is the proportion of fractions satisfying V100%≥95%. Geometric exceedance did not cause immediate coverage failure. At 80 mm, the required displacement quantile P95 was 1.71 mm, but mean coverage remained 97.6 % and the pass rate was 88.4 %, consistent with buffering by penumbra and volume-averaging effects. In contrast, distal instability was driven by dispersion and tail risk rather than mean coverage. At d=100 mm, mean V100% was 95.0 %, but pass probability was 52.1 % and P5 was 85.4 %, showing that mean metrics can mask substantial failure probability. Pass probability and low-quantile coverage therefore better capture both failure frequency and severity, as detailed in Supplementary Sections 6.2 and 6.3. The geometry-to-dose mapping also exhibited nonlinear sensitivity amplification. As displacement P95 increased from 1.07 mm to 4.28 mm, the decline in mean V100% steepened, consistent with a penumbra-dominated regime in which coverage becomes highly sensitive to displacement (Table 1). Risk therefore concentrates within a narrow transition band rather than accumulating linearly with distance, and different rotational precision levels mainly shift the location of this band, as shown in Supplementary Section 6.4. The divergence between mean coverage and pass probability is visualized in Fig. S4, supporting the use of distribution-based endpoints for distal-target risk assessment. Distance-adaptive PTV margin requirements The minimum expansion Mreq required to maintain a 95 % coverage pass probability increased monotonically with target-to-isocenter distance and was strongly dependent on σrot (Fig. 5). Under σrot=0.5°, the Mreq was 2.14 mm at 100 mm and 3.21 mm at 150 mm. With approximately isotropic expansion, irradiated volume scales cubically with margin size, which may increase the risk of complications such as radiation necrosis. In contrast, σrot=0.2° maintained the required margin at approximately 1.2–1.3 mm even at 150–200 mm. These results indicate that improving rotational precision can reduce the need for large distal margins and may better balance target coverage and normal-brain exposure. For clinical use, we extended the analytic relationship to a quick-reference heat map over a continuous precision range (Supplementary Section 7, Fig. S7) to enable direct lookup of Mreq from machine performance and target distance. Distance-based PTV margin requirements. The minimum PTV expansion required to maintain a 95 % coverage pass probability is shown for σrot=0.2°, 0.5°, and 1.0°. Curves are computed from Mreq=d·sin(2.45σrot). Background shading provides visual feasibility strata and is not intended as universal clinical standards. Discussion This study quantified distance-amplified residual rotational uncertainty in SIMT SRS and linked geometric error propagation to target dose-coverage failure risk. Using the small-angle rigid-body approximation Δr=θ×r, we verified accuracy over the SIMT-relevant range and achieved results comparable to matrix-based calculations at lower computational cost, enabling rapid geometric uncertainty assessment in routine workflows. Prior SIMT reports have shown distal coverage loss using plan recalculation studies, perturbation analyses, or measurements, but conclusions can be plan and case dependent, which limits cross-center transferability [3], [4], [5], [12], [13]. Statistical modeling work by Chang previously translated residual rotations into distance-dependent margin requirements [14]. Our framework builds on these efforts but defines the rotational margin from the two effective tangential components that produce target displacement, rather than from a three-dimensional resultant rotation vector. This distinction explains the different statistical factors (Chang’s model: k=2.795; present Rayleigh-based model: k=2.45) and reflects differences in the defined coverage event rather than a contradiction between models. By reducing key inputs to target distance d and an institution-measurable σrot, DAGM provides an interpretable route for risk stratification and distance-adaptive margin selection. The results show that geometric risk in SIMT is governed by both target-to-isocenter distance and residual rotational precision. Under σrot = 0.5° and a 1 mm margin, 100 mm emerged as a clear risk threshold, where mean V100% remained 95.0 % but the coverage pass rate dropped to 52.1 %, supporting distance as an independent risk factor in SIMT planning, consistent with Nakano et al. [10]. This finding indicates that mean coverage alone can mask distal-target instability. Pass probability and low-quantile coverage better describe both failure frequency and severity, especially when error distributions broaden at larger distances. Clinically, distal risk can be reduced by increasing PTV margins or improving rotational control. Our results clarify the trade-off. Margin-only compensation can restore coverage probability, but it increases normal brain exposure. Improving residual rotational precision shifts the failure transition outward and allows smaller margins for distal targets. These findings suggest several practical considerations for SIMT implementation. During setup, initial rotational offsets should be minimized because 6-DOF corrections do not remove distance-dependent amplification for distal targets [7], [21]. During planning, target-to-isocenter distance should be incorporated into risk review, especially when small lesions are located far from the isocenter, and distance-adaptive margins can mitigate distal under-coverage. Table S1, Fig. S7, or the SIMT error calculator (Supplementary Section 8) can provide rapid estimates of geometric uncertainty and distance-adaptive margin requirements from local σrot. For quality assurance, off-axis geometric verification, such as extended-distance Winston-Lutz testing, may help confirm whether local rotational performance is sufficient for distal targets [12], [13], [22]. In practice, σrot should be estimated from local post-correction IGRT logs or rotational QA records obtained under the same immobilization and correction workflow. Pitch, roll, and yaw should be evaluated separately using a sufficiently large local sample, with periodic updating as data accumulate, before using a pooled value or a conservative axis-specific/empirical percentile. The proposed DAGM should be interpreted mainly as a per-fraction geometric-containment framework. In fractionated SIMT treatments, independent random rotations may partially average out in accumulated dose, but this effect is limited in hypofractionated regimens and does not compensate for systematic bias or repeated workflow errors. Therefore, margin relaxation in fractionated settings should be supported by accumulated-dose or patient-specific robustness evaluation. This study has limitations. First, the model assumes rigid-body motion, which is appropriate for intracranial SRS [23] but does not capture deformation and intrafraction motion in extracranial SBRT. Second, the Rayleigh-based factor assumes independent, zero-mean, isotropic Gaussian residual rotations. This assumption applies when the two rotation components perpendicular to the target radial direction, which produce first-order displacement, are independent Gaussian variables. However, SGRT and open-face mask workflows may show skewed, heavy-tailed, anisotropic, or biased residual error distributions. In such cases, the rigid-body displacement model remains valid, but 2.45σrot should be replaced by the empirical 95th percentile of the locally measured perpendicular rotational component derived from IGRT, surface-guided radiotherapy (SGRT), or QA data. Third, the dose-coverage analysis used a simplified spherical target and idealized gradient representation rather than full patient-specific dose recalculation. Therefore, it does not capture irregular target shape, lesion-specific modulation, spatially varying gradients, nearby organs of interest, or plan-quality metrics such as the van’t Riet/Paddick conformity index and gradient index. Future work should combine local dose-gradient perturbation, joint translational and rotational uncertainty modeling, and patient-specific dose recalculation, including conformity, gradient, and normal tissue dose metrics, to validate the clinical value of distance-adaptive strategies. Overall, these findings show that residual rotational uncertainty and target-to-isocenter distance jointly determine rotation-induced displacement and target dose-coverage risk in SIMT. Distal-target instability is primarily driven by distribution tails and reduced coverage pass probability, whereas mean coverage can mask clinically relevant variation. Accordingly, SIMT evaluation should include pass probability and low-quantile coverage, and target-to-isocenter distance should be considered as an independent planning and quality-assurance factor. Although increasing margins can improve distal-target coverage, this approach increases normal tissue exposure. Improving rotational accuracy through system performance optimization and strict quality assurance may therefore be a more efficient strategy for extending treatable distances while limiting margin expansion. The proposed DAGM framework may support risk stratification and improve distal-target coverage robustness in SIMT. Funding statement This work was supported by the National Key Research and Development Program of China (2023YFC2413900), the National Natural Science Foundation of China (62401636 and 12405409). CRediT authorship contribution statement Jiaxin Deng: Writing – original draft, Visualization, Software, Methodology, Formal analysis, Conceptualization. Zun Piao: Writing – original draft, Validation, Software, Methodology, Investigation, Data curation. Xinyu Song: Writing – review & editing, Visualization, Validation, Investigation, Formal analysis. Danyang Li: Validation, Formal analysis, Data curation. Jiang Hu: Resources, Investigation. Shouliang Ding: Validation, Software, Funding acquisition. Yanfei Liu: Writing – review & editing, Resources. Li Chen: Investigation, Funding acquisition, Data curation. Guanqun Zhou: Visualization, Validation. Ying Sun: Resources, Project administration. Xiaoyan Huang: Writing – review & editing, Supervision, Project administration, Conceptualization. Guangyu Wang: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Guangyu Wang reports financial support was provided by the National Natural Science Foundation of China. Shouliang Ding reports financial support was provided by the National Natural Science Foundation of China. Li Chen reports financial support was provided by the National Key Research and Development Program of China. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements None. Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.phro.2026.101012. Contributor Information Xiaoyan Huang, Email: huangxiaoy@sysucc.org.cn. Guangyu Wang, Email: wanggy4@sysucc.org.cn. Supplementary data The following are the Supplementary data to this article: Supplementary Data 1 Supplementary material contains detailed mathematical derivations, Monte Carlo validation, additional tables and figures, dose-coverage analyses, lookup heatmaps, and information on the SIMT geometric error calculator supporting the DAGM framework. Data availability Research data and code supporting this study, including the Monte Carlo simulation program, analysis scripts used to generate the figures and tables, and the open-source “SIMT error calculator” (source code and compiled application), will be publicly available via GitHub (https://github.com/Xina359/SIMT-DAGM-Calculator). References Associated Data Supplementary Materials Supplementary Data 1 Supplementary material contains detailed mathematical derivations, Monte Carlo validation, additional tables and figures, dose-coverage analyses, lookup heatmaps, and information on the SIMT geometric error calculator supporting the DAGM framework. Data Availability Statement Research data and code supporting this study, including the Monte Carlo simulation program, analysis scripts used to generate the figures and tables, and the open-source “SIMT error calculator” (source code and compiled application), will be publicly available via GitHub (https://github.com/Xina359/SIMT-DAGM-Calculator).