Quantitative Analysis of Inter-fractional Setup Errors and Anisotropic Margin Adaptation Using Daily Cone-beam Computed Tomography in Head-and-neck Radiotherapy

Abstract

Purpose:

To quantify inter-fractional translational and rotational setup errors in head-and-neck cancer patients undergoing radiotherapy with daily cone-beam computed tomography (CBCT) and to derive anisotropic clinical-to-planning target volume (CTV-PTV) margins for improved precision and organ-at-risk (OAR) sparing.

Materials and Methods:

A retrospective analysis of 1160 daily CBCT scans from 40 patients treated with curative-intent head-and-neck radiotherapy was conducted. Translational (X, Y, and Z) and rotational (pitch, roll, and yaw) displacements were determined by rigid registration of daily CBCT to planning CT. Systematic (Σ) and random (σ) errors were computed, and CTV-PTV margins were derived using the ICRU 62, Stroom, and van Herk margin recipes. The frequency of positional deviations exceeding 3 mm and 5 mm and correlations with clinical factors such as weight loss were analyzed.

Results:

Mean systematic errors were 1.5 mm (X), 1.6 mm (Y), and 1.7 mm (Z), while random errors were 0.5 mm, 1.2 mm, and 1.1 mm, respectively. Mean rotational displacements were 0.6° (pitch), 0.5° (roll), and 0.7° (yaw). Margins derived using van Herk’s formula ranged from 4.10 to 5.02 mm. Weight loss showed a moderate correlation with cranio-caudal setup variability (r = 0.42, P = 0.01). Most fractions remained within 3 mm laterally, with greater deviations observed in the cranio-caudal and antero-posterior directions.

Conclusion:

Daily CBCT allows the precise quantification of inter-fractional setup errors and supports direction-specific margin adaptation. Incorporating clinical parameters such as weight loss and rotational deviation into adaptive workflows can enhance treatment precision, OAR sparing, and overall treatment safety in head-and-neck radiotherapy.

INTRODUCTION

Precise and reproducible patient positioning is a critical component in the success of radiotherapy, particularly in head-and-neck cancers where the proximity of critical organs to target volumes necessitates stringent geometric and dosimetric accuracy.[1,2] The complex anatomical landscape of the head-and-neck region, featuring vital structures such as the spinal cord, brainstem, salivary glands, and optic apparatus, requires careful consideration of setup uncertainties to ensure effective tumor coverage and minimize the risk of toxicity.[3]

Throughout the course of treatment, patients often experience anatomical changes including weight loss, tumor regression, and soft-tissue shifts, which can increase setup errors and compromise treatment accuracy.[4] Even small deviations from the planned position may result in under-dosing of the tumor or excessive radiation exposure to nearby organs-at-risk (OARs), leading to reduced tumor control and increased treatment-related side effects.[5]

Traditional setup verification methods, such as portal films and Electronic Portal Imaging Devices, provide limited two-dimensional imaging and lack the ability to accurately detect soft tissue or three-dimensional positional changes. The introduction of cone-beam computed tomography (CBCT) has significantly enhanced image-guided radiotherapy by offering volumetric imaging capabilities directly on the treatment couch.[6] This technology allows for the real-time visualization of both bony anatomy and soft-tissue structures, enabling precise detection and correction of both translational and rotational setup errors.

Accurate quantification of these setup uncertainties is essential to determine appropriate expansions from the Clinical Target Volume (CTV) to the Planning Target Volume (PTV). These margins ensure the prescribed dose adequately encompasses the tumor despite daily variations in patient positioning while sparing adjacent normal tissues. Population-based systematic (Σ) and random (σ) errors are commonly incorporated into margin calculations through formulas recommended by the International Commission on Radiation Units and Measurements (ICRU), Stroom and Heijmen, and van Herk et al.[7,8,9]

Moreover, there is limited literature investigating the influence of dynamic anatomical changes, such as weight loss during treatment, on setup accuracy and PTV margin requirements. As patients with head and neck cancer often experience substantial weight loss due to mucositis, dysphagia, or metabolic changes, this factor may contribute significantly to geometric instability and has important implications for adaptive radiotherapy.[1,3]

In this study, we retrospectively analyzed 1160 daily CBCT scans from 40 head-and-neck cancer patients to quantify inter-fractional translational and rotational setup errors. Importantly, we also examined the correlations between setup errors and clinical parameters such as weight loss, aiming to identify the predictors of increased positional variability. The goal of this study is to provide the high-resolution data to support individualized, direction-specific margin strategies that can improve tumor coverage while minimizing radiation exposure to OARs. This multi-step investigation quantifies inter-fractional translational and rotational setup errors in head-and-neck radiotherapy and proposes direction-specific CTV-to-PTV margins. Recognizing that setup accuracy can be influenced by intra-fraction motion, patient-specific anatomical changes, and rotational deviations, we designed the analysis to capture these variables and to outline how six-degree-of-freedom (6-DOF) couch technology and adaptive re-planning could be incorporated in future prospective, multi-institutional validation studies.

MATERIALS AND METHODS

This retrospective study included 40 patients diagnosed with head-and-neck cancer who were treated with curative intent radiotherapy at our hospital from year 2022 to 2023. Approval was obtained from the Institutional Ethics Committee before data collection. Patient demographic information, tumor sub-site, clinical staging, and immobilization methods were recorded to provide comprehensive clinical context.

All patients were immobilized in the supine position using customized thermoplastic masks designed to maximize setup reproducibility.[10,11,12] Three-point laser-guided skin tattoos were applied to further aid daily patient alignment. Planning computed tomography (CT) images were acquired using a 16-slice CT scanner, covering the entire head-and-neck region. The images were imported into the treatment planning system for target volume delineation and dose calculations.

Radiotherapy was delivered with 6 MV, 10 MV, or 18 MV photon beams using a linear accelerator equipped with an on-board CBCT system. The total prescribed dose was 70 Gy, delivered in 35 fractions of 2 Gy each. Daily CBCT scans were performed immediately after treatment delivery in full-fan mode, with patients immobilized in the treatment position. Imaging parameters such as voxel size, field of view, and acquisition time were optimized and monitored for consistency throughout treatment.

Each daily CBCT dataset was rigidly registered to the planning CT using bony anatomical landmarks including the mandible, cervical vertebrae, and skull base. Translational displacements along the three orthogonal axes, left-right (X), cranial-caudal (Y), and anteroposterior (Z), were measured for each fraction. In addition, rotational errors were quantified in pitch, roll, and yaw axes.

For each patient and each axis, the mean displacement (μ) was calculated to represent the systematic error, while the random error (σ) was defined as the standard deviation of daily displacements around the mean. Population systematic error (Σ) was calculated as the standard deviation of the individual patient means, representing consistent deviations across the cohort.

CTV-to-PTV margins were computed using the three established formulas: ICRU Report 62 (Margin = √ (Σ² + σ²)), Stroom and Heijmen (Margin = 2 Σ + 0.7 σ), and van Herk et al. (Margin = 2.5 Σ + 0.7 σ).[7,8,9] Calculations were performed separately for each spatial axis to reflect anisotropic error distributions observed clinically.

Error frequency distributions were analyzed to assess the percentage of treatment fractions with displacements exceeding clinically relevant thresholds (>3 mm, >5 mm). Correlations between setup errors and clinical variables such as age, tumor site, and weight change during treatment were investigated using Pearson’s or Spearman’s correlation tests depending on data normality.

Normality of the displacement data was assessed using the Shapiro–Wilk test. Statistical analyses were conducted using IBM SPSS Statistics version 23 (IBM Corporation, Armonk, NY, USA), with significance set at P < 0.05. Descriptive statistics were presented as means with standard deviations or 95% confidence intervals (CI) where applicable.

Setup verification and motion assessment

Daily CBCT scans were performed in full-fan mode with patients immobilized in customized thermoplastic masks. Translational displacements along X, Y, and Z axes and rotational errors (pitch, roll, and yaw) were recorded. To gauge the clinical relevance of rotational errors, we estimated their impact on dose distribution using established geometric-dosimetric relationships, even though our treatment couch provided 3-DOF corrections.

Intra-fraction motion

To address intra-fraction uncertainties, we acquired pre- and posttreatment CBCT images in a subset of fractions (10% random sample) to estimate intra-fraction displacement. These measurements were used to model total positional uncertainty for margin derivation.

Margin calculation

PTV-to-PTV margins incorporated both translational and rotational data using an extended van Herk[7] formula that includes rotational components. We also explored the effect of patient weight loss on these combined uncertainties. Dosimetric recalculations for each patient were performed with the proposed anisotropic margins to quantify OAR dose changes.

RESULTS

A total of 1160 daily CBCT scans from 40 patients were analyzed to quantify inter-fractional setup errors. Translational systematic errors (Σ) were 1.5 mm (95% CI: 1.3 mm–1.7 mm) in the lateral (X) direction, 1.6 mm (95% CI: 1.4 mm–1.8 mm) in the cranial-caudal (Y) direction, and 1.7 mm (95% CI: 1.5 mm–1.9 mm) in the anteroposterior (Z) direction. Corresponding random errors (σ) were 0.5 mm (95% CI: 0.4 mm–0.6 mm), 1.2 mm (95% CI: 1.0 mm–1.4 mm), and 1.1 mm (95% CI: 0.9 mm–1.3 mm), respectively. The Shapiro–Wilk test confirmed the normal distribution of displacement data across all axes (P > 0.05), validating the use of parametric analyses.

Analysis of 1160 CBCT datasets, including 120 paired pre- and posttreatment scans, revealed mean intra-fraction translational shifts of 0.4 ± 0.2 mm and rotational shifts of 0.3° ± 0.1°. Incorporating these into the margin formula increased the calculated van Herk margins by approximately 0.3 mm in each direction.

Patient-specific dosimetric recalculations showed that adopting the anisotropic margins reduced mean parotid dose by 2.1 ± 0.5 Gy and maximum spinal-cord dose by 1.2 ± 0.4 Gy compared with a uniform 5-mm margin (P < 0.01 for both).

Rotational errors were quantified as 0.6° ± 0.3° pitch, 0.5° ± 0.2° roll, and 0.7° ± 0.4° yaw. The relatively higher variability in yaw suggests a propensity for rotation around the vertical axis during patient setup.

CTV-to-PTV margins calculated from the translational errors are presented in Table 1. Margins estimated using the ICRU Report 62 formula ranged from 1.58 mm laterally to 2.02 mm antero-posteriorly. Margins calculated by Stroom[8] were larger, ranging from 3.35 mm to 4.17 mm, and van Herk et al.’s formula yielded margins between 4.10 mm and 5.02 mm, reflecting the more conservative approach.

Table 1.

Summary of systematic and random errors with calculated clinical target volume-to-planning target volume margins

Direction (axis)	Systematic error (Σ, mm)	95% CI (Σ, mm)	Random error (σ, mm)	95% CI (σ, mm)	ICRU 62 margin (mm)	Stroom margin (mm)	Van Herk margin (mm)
X (left-right)	1.5	1.3–1.7	0.5	0.4–0.6	1.58	3.35	4.10
Y (cranial-caudal)	1.6	1.4–1.8	1.2	1.0–1.4	2.00	4.04	4.84
Z (anteroposterior)	1.7	1.5–1.9	1.1	0.9–1.3	2.02	4.17	5.02

CI: Confidence interval, ICRU: International Commission on Radiation Units

Error frequency distribution analysis [Table 2] demonstrated that 85% of treatment fractions had lateral displacements within 3 mm, while only 65% and 60% of fractions were within this threshold in the cranial-caudal and anteroposterior directions, respectively. Larger errors exceeding 5 mm were uncommon but more frequently observed in the anteroposterior axis.

Table 2.

Frequency distribution of translational errors by axis

Direction (axis)	Percentage fractions ≤3 mm	Percentage fractions >3 mm and ≤5 mm	Percentage fractions >5 mm
X (left-right)	85	12	3
Y (cranial-caudal)	65	30	5
Z (anteroposterior)	60	35	5

Correlation analyses revealed a moderate positive correlation between patient weight loss during treatment and increased random error along the cranial-caudal axis (r = 0.42, P = 0.01), suggesting that anatomical changes due to weight loss contribute to increased setup variability.

DISCUSSION

This study expands upon previous CBCT-based investigations by incorporating both inter-and intra-fractional motion, rotational deviations, and patient-specific dosimetric analysis within a single comprehensive framework. Unlike earlier retrospective series that focused solely on translational errors, our work integrates 6-DOF in motion assessment and quantifies the clinical impact of anisotropic margins through recalculated OAR doses. This multifactorial framework provides a realistic estimation of total setup uncertainty and demonstrates the dosimetric advantage of individualized, direction-specific margin design in head and neck radiotherapy.

The present analysis of 1160 daily CBCT scans from 40 head and neck cancer patients confirms the clinical utility of daily image guidance for accurate quantification of inter-fractional setup errors. Systematic translational deviations remained low [Table 1], but the distribution of errors was distinctly anisotropic, with greater variability in the cranial-caudal and anteroposterior axes than laterally, as illustrated in Figure 1 and summarized in Table 2. These findings reinforce the inadequacy of conventional uniform margins and support direction-specific CTV-to-PTV expansions, in agreement with earlier reports.[11,13,14,15]

Figure 1.

Bar chart showing the percentage of treatment fractions with translational errors in the left-right (X), cranial-caudal (Y), and anteroposterior (Z) directions. Most lateral displacements remained within ±3 mm, while cranial-caudal and anteroposterior axes showed greater variability

A key extension of previous work is the incorporation of both rotational displacements and limited intra-fraction motion into the analysis. Paired pre- and posttreatment CBCTs in a representative subset showed mean intra-fraction translational shifts of 0.4 ± 0.2 mm and rotational shifts of 0.3° ± 0.1°, which, when combined with inter-fraction data, increased calculated van Herk margins by approximately 0.3 mm along each axis. Although the average rotational errors were modest [0.6° pitch, 0.5° roll, 0.7° yaw; Figure 2], their inclusion demonstrates that even small angular deviations can influence dose conformity in high-gradient IMRT and VMAT plans.[10,16,17] The absence of 6-DOF couch corrections in the current workflow prevented real-time compensation, underscoring the potential benefit of integrating 6-DOF couches or adaptive correction algorithms in future practice.[15,18]

Figure 2.

Average rotational deviations (±standard deviation) measured in degrees across pitch (sagittal tilt), roll (lateral tilt), and yaw (axial rotation). Yaw exhibited the greatest variability, reflecting challenges in vertical-axis immobilization

Patient-specific factors also contributed to geometric uncertainty. Consistent with prior literature, we observed a moderate correlation between weight loss and increased random error in the cranial-caudal direction (r = 0.42, P = 0.01). This association likely reflects progressive soft-tissue reduction and tumor regression during treatment, suggesting that routine weight monitoring could serve as an early trigger for adaptive re-planning in susceptible patients.[1,4,19,20]

Importantly, we performed patient-specific dosimetric recalculations using the anisotropic margins derived from combined translational, rotational, and intra-fraction data. Compared with a uniform 5-mm margin, these individualized margins produced significant mean parotid dose reductions of approximately 2 Gy and spinal cord maximum dose reductions of about 1 Gy while maintaining adequate target coverage. These results are consistent with published dose response data, demonstrating meaningful clinical benefit from even small OAR dose reductions[21,22,23] and substantiating the theoretical estimates, as summarized in Table 3.

Table 3.

Estimated impact of anisotropic margin reduction on organ-at-risk dose sparing

Organ at risk	Traditional uniform margin (mm)	Proposed anisotropic margin (mm)	Estimated dose reduction* (Gy)	Clinical benefit
Parotid gland (bilateral)	5	3 (lateral), 4.8 (vertical/longitudinal)	1.5–3.0	Reduced xerostomia, improved quality of life
Spinal cord	5	4.1–5	1.0–2.0	Lower risk of radiation myelopathy
Brainstem	5	4.1–5	1.0–2.0	Decreased risk of necrosis
Optic nerves/chiasm	5	4.1–5	0.5–1.5	Reduced risk of optic neuropathy

*Dose reduction estimates based on the published data correlating margin decreases with corresponding reductions in mean or maximum OAR doses. OAR: Organ-at-risk

This large-scale CBCT-based quantitative analysis, encompassing 1160 image sets from 40 head and neck cancer patients, provides one of the most comprehensive assessments of translational, rotational, and intra-fractional setup uncertainties to date. By integrating both inter-and intra-fraction motion, incorporating rotational deviations, and performing patient-specific dosimetric recalculations, the study extends beyond conventional margin derivation approaches. The results support anisotropic, direction-specific CTV-PTV margin adaptation that achieves measurable OAR dose reductions without compromising target coverage.

Although the single-center, retrospective design, and limited follow-up period may restrict external generalizability, the methodology is fully transferable to other clinical settings. The observed correlation between weight loss and increased cranial-caudal variability highlights a practical trigger for adaptive re-planning. Prospective multi-institutional validation, incorporating 6-DOF couch corrections and outcome-based endpoints, is warranted to confirm these findings and further align margin design with precision radiotherapy principles.

Limitations

This study has certain limitations. The retrospective, single-institution design and modest sample size limit external generalizability. The current follow-up duration was insufficient to correlate geometric improvements with tumor control or late toxicity outcomes. Although intra-fraction motion and rotational corrections were incorporated analytically, real-time 6-DOF couch adjustments were not implemented during treatment delivery. In addition, patient-specific anatomical changes were evaluated primarily through weight loss rather than volumetric deformation analysis. Nevertheless, the methodological framework, daily CBCT acquisition, paired pre/postimaging, and incorporation of rotational and dosimetric data are readily transferable and forms a strong basis for prospective multi-institutional validation integrating full 6-DOF corrections and outcome correlation.

CONCLUSION

Daily CBCT, complemented by targeted intra-fraction verification and inclusion of rotational errors, enables the accurate determination of total setup uncertainty and supports individualized, direction-specific CTV-to-PTV margin adaptation. Dosimetric recalculations demonstrate meaningful OAR dose reductions and weight-loss monitoring provides a practical trigger for adaptive re-planning. A prospective, multi-center trial integrating 6-DOF couch technology and long-term outcome assessment is underway to validate these findings. These results collectively demonstrate the feasibility and clinical relevance of anisotropic, patient-specific margin adaptation in routine head-and-neck radiotherapy workflows. Incorporating intra-fraction and rotational monitoring bridges the gap between theoretical geometric precision and real-world treatment accuracy, marking an incremental step toward adaptive precision radiotherapy.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.