Abstract
Objective
Cone-beam computed tomography (CBCT) has been widely adopted for patient setup evaluation and error correction in head and neck radiotherapy. Although different anatomical registration regions are routinely used in clinical practice, the characteristics of six-degree-of-freedom (6DoF) setup errors associated with different registration regions have not been systematically compared within the same CBCT acquisition. This study aimed to compare the distributional characteristics of 6DoF setup errors across different anatomical registration regions using a within-scan comparison framework and to describe their structural features from a multivariate perspective.
Methods
CBCT data acquired during the radiotherapy course of patients with head and neck cancer were retrospectively analyzed. For each CBCT scan, multiple anatomical registration regions were independently applied based on the same image dataset, and the corresponding 6DoF setup errors were recorded. Paired statistical methods were used to compare the distributions of setup errors among different registration regions. Principal component analysis (PCA) was further performed to characterize the overall structural patterns of the 6DoF error components.
Results
Within the same CBCT acquisition, the distributions of 6DoF setup errors differed among anatomical registration regions. The observed differences were mainly reflected in certain rotational components, whereas the differences among translational components were relatively less pronounced. PCA results demonstrated that the structure of 6DoF setup errors could be characterized by different variation patterns dominated by either rotational or translational components. Although partial overlap was observed among registration regions in the principal component space, differences in their distributional characteristics were still evident.
Conclusions
Within the same CBCT acquisition, the selection of anatomical registration regions may influence both the distribution and structural characteristics of 6DoF setup errors in head and neck radiotherapy. These findings suggest that the potential impact of registration region selection should be considered during clinical setup evaluation.
Keywords: Head and neck radiotherapy, Cone-beam CT, Six-degree-of-freedom registration error, Principal component analysis, Registration region
Introduction
Cone-beam computed tomography (CBCT) has been widely incorporated into head and neck radiotherapy for patient setup evaluation and error correction [1]. By registering pre-treatment or intra-treatment CBCT images with planning CT images [2], patient positioning deviations can be quantitatively assessed, thereby providing essential support for the delivery of high-precision radiotherapy. Given the anatomical complexity of the head and neck region and the close spatial relationship between target volumes and adjacent organs at risk [3], accurate CBCT-based registration plays a critical role in routine clinical radiotherapy workflows.
In clinical practice, CBCT registration is typically performed within predefined anatomical registration regions [4]. These regions differ in spatial location, anatomical composition, and rigidity characteristics, which may influence the range and distribution of image information involved in the registration process. As a result, setup errors derived from different registration regions may exhibit distinct numerical characteristics across translational and rotational directions [5, 6]. Therefore, the selection of registration regions represents not only a technical step within the clinical workflow but also a potential factor affecting the quantitative evaluation of patient positioning errors.
Previous studies on CBCT-guided radiotherapy have mainly focused on setup accuracy, error distributions, or margin determination [7, 8]. Some investigations have analyzed setup errors under different treatment sites, immobilization strategies, or imaging conditions, providing valuable references for clinical setup management. However, in many study designs [9, 10], comparisons among different registration regions were conducted based on CBCT images acquired at different treatment fractions or under varying imaging conditions. Differences in the imaging conditions themselves may therefore confound the observed setup error characteristics. To reduce this potential confounding, the present study uses a within-scan comparison framework to quantify how the choice of anatomical registration region influences 6DoF setup corrections when evaluated on the same CBCT acquisition.
From the perspective of error composition, setup errors in head and neck radiotherapy generally consist of multiple translational and rotational components, which may exhibit certain degrees of correlation. Analyzing individual directional components in isolation may not fully reflect the overall characteristics of setup errors [11, 12]. Multivariate approaches that consider the joint behavior of multiple error components allow for a more comprehensive description of setup error characteristics. In this context, principal component analysis (PCA) has been applied to characterize the structural patterns of multi-dimensional errors in head and neck radiotherapy [13–16], providing additional insight beyond univariate analyses.
Based on the considerations above, this study retrospectively analyzed CBCT data acquired during routine head and neck radiotherapy. Under identical CBCT imaging conditions, 6DoF setup errors associated with different anatomical registration regions were systematically compared, and their distributional and structural characteristics were evaluated. By assessing the influence of registration region selection under unified imaging conditions, this study aims to provide more objective reference information for the selection and interpretation of registration regions in clinical setup evaluation.
Methods
Study design and data source
This study was designed as a single-center retrospective analysis based on cone-beam computed tomography (CBCT) guidance data routinely acquired during the radiotherapy course of patients with head and neck cancer. All CBCT datasets were obtained from actual clinical treatment workflows and were originally used for patient setup evaluation and registration-based correction.
During the radiotherapy course, individual patients could undergo multiple CBCT scans. For each CBCT scan, in order to meet clinical setup evaluation requirements, different anatomical registration regions were independently applied based on the same CBCT image dataset, and the corresponding six-degree-of-freedom (6DoF) setup errors were recorded. Accordingly, the data had a nested structure across patient–CBCT scan–registration region levels. Four commonly used anatomical registration regions were included: the head–neck combined frame ROI, the Nasopharyngeal frame ROI, the lower neck frame ROI, and the lower cervical vertebral body frame ROI. Each registration record consisted of three translational components—left–right (LR), superior–inferior (SI), and anterior–posterior (AP)—and three rotational components, including rotation (RTN), pitch, and roll.
A total of 82 patients with head and neck cancer were included in the analysis. All patients received radiotherapy for the first time and had pathologically confirmed poorly differentiated squamous cell carcinoma. Staging was performed according to the AJCC 8th edition criteria (stage I, n = 4; stage II, n = 14; stage III, n = 26; and stage IV, n = 38). The age ranged from 30 to 68 years (median, 43 years). All patients were treated with standard IMRT, and the prescription dose was 70 Gy. Across the entire radiotherapy course, these patients underwent 486 CBCT scans. For each scan, four region-specific registrations were available, yielding 1,944 region-specific 6DoF records (486 × 4).All four regions were implemented using the same registration system and the same CBCT dataset, with standardized rectangular ROIs defined by fixed anatomical landmarks, to ensure comparability across regions.In routine clinical practice, one clinically selected region was used for couch correction; the remaining regions were used for research-only comparisons.
CBCT acquisition was performed according to routine clinical practice rather than a fixed study protocol. The frequency of CBCT imaging varied among patients depending on clinical needs, typically including initial setup verification and additional scans when positional uncertainty was suspected. All scans were acquired using the institutional head CBCT protocol (100 kV, 150 mAs; full-fan with half-trajectory; 500 projections; scan diameter 26.2 cm and longitudinal range 18.5 cm; reconstruction matrix 512 with a slice thickness of 2.0 mm; CTDIw 0.32 cGy).All available CBCT scans during the treatment course that met image quality and registration requirements were included in the analysis, and no additional case selection was performed for research purposes. Quality control: All ROI-specific registration results were reviewed prior to analysis to confirm completeness of 6DoF outputs, successful image registration, and adequate CBCT image quality within the selected region.After quality control, all 1,944 ROI-specific registration records were retained for analysis.
This study was approved by the Ethics Committee of Sun Yat-sen University Cancer Center (Approval No. B2025-874-01).The study was conducted in accordance with the Declaration of Helsinki and relevant national regulations.Given the retrospective nature of the study and the use of anonymized clinical data, the requirement for informed consent was waived by the ethics committee.
CBCT registration workflow and error definition
All CBCT images were acquired according to routine clinical radiotherapy procedures. Registration between CBCT and planning CT was performed using the image registration module in Elekta MOSAIQ (software version 2.80 SP3) with an automatic grayscale-based rigid registration approach. For each predefined registration region (ROI), the system reported six-degree-of-freedom (6DoF) setup errors, including three translational components (LR, SI, and AP) and three rotational components (RTN, pitch, and roll), which were exported for subsequent analyses. All registrations were reviewed by experienced radiation therapists; manual adjustments were applied when necessary to ensure that the final registration met clinical setup evaluation standards.
In routine clinical practice, both translational and rotational corrections were applied using a six-degree-of-freedom (6DoF) treatment couch based on the clinically selected registration region. During treatment, the clinically selected registration region was the head–neck combined frame ROI, whereas the other three ROIs were generated retrospectively by re-drawing the registration window on the same CBCT dataset. Patients were immobilized with individualized devices composed of a foam cradle for posterior support and a head–neck–shoulder thermoplastic mask; one set was manufactured per patient and discarded after completion of treatment.The retrospective registrations followed the same automatic registration and therapist review workflow used in routine practice to ensure clinically acceptable matching.
For each individual CBCT scan, registration was independently performed within different anatomical registration regions, and the corresponding six-degree-of-freedom (6DoF) setup errors reported by the system were recorded. Translational setup errors were defined along the left–right (LR), superior–inferior (SI), and anterior–posterior (AP) directions. Rotational setup errors included rotation (RTN), pitch, and roll.
All setup error data analyzed in this study were directly obtained from the registration system outputs. No additional image-derived features, geometric descriptors, or dosimetric parameters were introduced during data processing or analysis.
Supplementary description of registration regions
The four registration regions were selected because they represent commonly used matching strategies in routine head-and-neck CBCT-guided setup evaluation at our institution. To ensure consistency and reproducibility, all regions were defined using fixed anatomical landmarks on CBCT images and were implemented as standardized rectangular ROIs during clinical image matching. Representative examples of the four ROIs on axial, sagittal, and coronal CBCT views are shown in Fig. 1.
Fig. 1.
Representative CBCT views showing the four predefined rectangular registration regions (ROIs) for image matching
Head–neck combined frame ROI: The left–right extent included the outer margins of both auricles. The superior–inferior extent ranged from the inferior orbital rim to the C7 vertebra. The anterior–posterior extent covered from the nasal tip to the posterior margin of the occipital bone.
Nasopharyngeal frame ROI: The left–right and anterior–posterior extents were identical to those of the head–neck combined frame ROI. The superior–inferior extent ranged from the inferior orbital rim to the superior border of C1.
Lower neck frame ROI: The left–right extent was identical to that of the head–neck combined frame ROI. The anterior–posterior extent covered from the anterior margin of the hyoid bone to the posterior margin of the occipital bone. The superior–inferior extent ranged from the inferior border of C3 to the inferior border of C7.
Lower cervical vertebral body frame ROI: The left–right extent was defined by the lateral borders of the cervical vertebral bodies, and the anterior–posterior extent by the anterior and posterior borders of the vertebral bodies. The superior–inferior extent was identical to that of the lower neck frame ROI (from the inferior border of C3 to the inferior border of C7).
All registration regions were independently applied to the same CBCT dataset, and six-degree-of-freedom (6DoF) registration results were recorded separately for each region.
Statistical analysis
The primary focus of the statistical analysis was to compare the distributional characteristics of six-degree-of-freedom (6DoF) setup errors associated with different anatomical registration regions within the same CBCT acquisition. Given that multiple registration results corresponding to different anatomical regions were derived from the same CBCT scan, all comparisons among registration regions were conducted using paired statistical approaches, with the CBCT scan as the unit of pairing.
For each translational and rotational error component, the Friedman test was first applied to assess whether overall differences existed among the registration regions. When the global test indicated statistical significance, post hoc pairwise comparisons were subsequently performed using the Wilcoxon signed-rank test. To account for multiple comparisons, Holm’s method was applied to adjust the resulting p-values. Setup error measurements were summarized using medians and interquartile ranges, and statistical significance was defined as a two-sided p-value of less than 0.05.
To facilitate visualization of setup error behavior over the radiotherapy course, the medians and interquartile ranges of 6DoF setup errors for each registration region were calculated at different CBCT scan indices and presented graphically. The scan index was defined as the chronological order of CBCT acquisition for each patient; in routine practice, the prescribed number of CBCT scans was approximately evenly distributed across treatment fractions (i.e., after the first CBCT, subsequent scans were scheduled at similar fraction intervals). This analysis was intended to illustrate the overall distributional patterns and variability of setup errors across CBCT scans, rather than to support inferential or trend-based statistical conclusions.
In addition, principal component analysis (PCA) was performed on standardized 6DoF setup error components to describe major patterns of multivariate variation. PCA results were summarized in terms of explained variance ratios, component loading matrices, and sample distributions in the principal component space. This analysis was used to provide a structural description of the relationships among translational and rotational error components, without introducing classification, discrimination, or predictive modeling procedures.
Results
Distributional characteristics of 6DoF setup errors across different registration regions
Under the same CBCT acquisition, six-degree-of-freedom (6DoF) setup errors associated with different anatomical registration regions exhibited differences in their numerical distributions. The median values and dispersion of both translational components (LR, SI, and AP) and rotational components (RTN, pitch, and roll) were not fully consistent across registration regions. The overall distributional characteristics of 6DoF setup errors for different registration regions are illustrated in Fig. 2.
Fig. 2.
Distribution of 6DoF setup errors across different anatomical registration regions. Note: Box plots represent the median and interquartile range (IQR), with whiskers indicating the data range. Translational components include left–right (LR), superior–inferior (SI), and anterior–posterior (AP) directions, while rotational components include rotation (RTN), pitch, and roll
Paired statistical analysis demonstrated that, for certain error components, overall differences existed among registration regions. Results of the Friedman test indicated statistically significant differences among registration regions for several components, particularly within the rotational directions, whereas differences among translational components were less consistently observed. Detailed results of the Friedman global tests are summarized in Table 1.
Table 1.
Friedman test results for 6DoF setup errors across different anatomical registration regions
| Component | Friedman_stat | Friedman_p |
|---|---|---|
| SI | 37.53 | < 0.001 |
| LR | 15.60 | < 0.001 |
| AP | 42.78 | < 0.001 |
| RTN | 44.14 | < 0.001 |
| pitch | 46.74 | < 0.001 |
| roll | 44.35 | < 0.001 |
Friedman statistics and corresponding p values are reported for each translational and rotational component
Following the global tests, pairwise comparisons between registration regions were conducted using the Wilcoxon signed-rank test. The magnitude and statistical significance of differences varied across error components and registration region pairs, indicating that the extent of distributional differences was direction-dependent. Results of the pairwise comparisons, together with Holm-adjusted p-values, are presented in Table 2.
Table 2.
Pairwise comparisons of 6DoF setup errors between registration regions (Wilcoxon signed-rank test, Holm-adjusted p values)
| Component | Region_A | Region_B | Wilcoxon_stat | p_raw | p_holm |
|---|---|---|---|---|---|
| SI | HR | NR | 1350.5 | < 0.001 | < 0.001 |
| HR | LNR | 18187.5 | 0.54 | 0.72 | |
| HR | LCVR | 21,103 | < 0.001 | < 0.001 | |
| NR | LNR | 18,011 | < 0.001 | < 0.001 | |
| NR | LCVR | 23,735 | 0.36 | 0.72 | |
| LNR | LCVR | 17,885 | < 0.001 | < 0.001 | |
| LR | HR | NR | 4308 | < 0.001 | < 0.001 |
| HR | LNR | 21156.5 | 0.24 | 0.61 | |
| HR | LCVR | 27233.5 | 0.06 | 0.27 | |
| NR | LNR | 22,755 | 0.20 | 0.61 | |
| NR | LCVR | 29,479 | 0.77 | 0.77 | |
| LNR | LCVR | 25223.5 | 0.05 | 0.27 | |
| AP | HR | NR | 15,083 | 0.04 | 0.07 |
| HR | LNR | 28070.5 | 0.04 | 0.07 | |
| HR | LCVR | 27,041 | < 0.001 | < 0.001 | |
| NR | LNR | 32,850 | 0.02 | 0.05 | |
| NR | LCVR | 28,621 | < 0.001 | < 0.001 | |
| LNR | LCVR | 14183.5 | < 0.001 | < 0.001 | |
| RTN | HR | NR | 25979.5 | < 0.001 | < 0.001 |
| HR | LNR | 44754.5 | 0.04 | 0.06 | |
| HR | LCVR | 45094.5 | < 0.001 | 0.01 | |
| NR | LNR | 39,964 | < 0.001 | < 0.001 | |
| NR | LCVR | 39,923 | < 0.001 | < 0.001 | |
| LNR | LCVR | 45,870 | 0.03 | 0.06 | |
| pitch | HR | NR | 42,048 | < 0.001 | 0.01 |
| HR | LNR | 35,483 | < 0.001 | < 0.001 | |
| HR | LCVR | 53409.5 | 0.51 | 0.51 | |
| NR | LNR | 35,256 | < 0.001 | < 0.001 | |
| NR | LCVR | 50709.5 | 0.20 | 0.41 | |
| LNR | LCVR | 39,410 | < 0.001 | < 0.001 | |
| roll | HR | NR | 28781.5 | < 0.001 | < 0.001 |
| HR | LNR | 43,658 | < 0.001 | < 0.001 | |
| HR | LCVR | 42963.5 | < 0.001 | < 0.001 | |
| NR | LNR | 42392.5 | < 0.001 | < 0.001 | |
| NR | LCVR | 41,073 | < 0.001 | < 0.001 | |
| LNR | LCVR | 45,394 | < 0.001 | < 0.001 |
P values were adjusted using Holm’s method to account for multiple comparisons. p_raw indicates unadjusted p values, and p_holm indicates Holm-adjusted p values
Abbreviations: HR Head–neck combined frame ROI, NR Nasopharyngeal frame ROI, LNR Lower neck frame ROI, LCVR Lower cervical vertebral body frame ROI
From a descriptive perspective, different registration regions showed distinct distributional patterns across translational and rotational components. For some registration regions, rotational components exhibited larger median values or greater dispersion, whereas translational components tended to be more concentrated. In contrast, other registration regions demonstrated varying degrees of dispersion across different directional components. To further quantify these characteristics, the median values and interquartile ranges of each error component were calculated for each registration region, with the results summarized in Table 3.
Table 3.
Medians and interquartile ranges of 6DoF setup errors for different anatomical registration regions
| Registration Region | Component | Median | Q1 | Q3 | IQR |
|---|---|---|---|---|---|
| HR | SI | 0.1 | 0 | 0.1 | 0.1 |
| HR | LR | 0 | 0 | 0.1 | 0.1 |
| HR | AP | 0.1 | 0 | 0.1 | 0.1 |
| HR | RTN | -0.1 | -0.5 | 0.275 | 0.775 |
| HR | pitch | 0.2 | -0.2 | 0.6 | 0.8 |
| HR | roll | -0.4 | -1 | 0.1 | 1.1 |
| NR | SI | 0.1 | 0 | 0.2 | 0.2 |
| NR | LR | 0.1 | 0 | 0.1 | 0.1 |
| NR | AP | 0 | -0.1 | 0.1 | 0.2 |
| NR | RTN | 0 | -0.4 | 0.5 | 0.9 |
| NR | pitch | 0.2 | -0.4 | 0.7 | 1.1 |
| NR | roll | -0.5 | -1.2 | 0.1 | 1.3 |
| LNR | SI | 0.1 | 0 | 0.2 | 0.2 |
| LNR | LR | 0 | -0.1 | 0.15 | 0.25 |
| LNR | AP | 0.1 | -0.1 | 0.2 | 0.3 |
| LNR | RTN | -0.2 | -0.6 | 0.2 | 0.8 |
| LNR | pitch | 0.6 | -0.1 | 1.4 | 1.5 |
| LNR | roll | -0.2 | -1 | 0.6 | 1.6 |
| LCVR | SI | 0.1 | 0 | 0.2 | 0.2 |
| LCVR | LR | 0.1 | -0.1 | 0.2 | 0.3 |
| LCVR | AP | 0.1 | -0.1 | 0.3 | 0.4 |
| LCVR | RTN | -0.2 | -1.1 | 0.5 | 1.6 |
| LCVR | pitch | 0.3 | -0.9 | 1.6 | 2.5 |
| LCVR | roll | 0 | -1.1 | 1 | 2.1 |
Values are presented as median (Q1–Q3).Translational errors are reported in mm, and rotational errors are reported in degrees (°)
Abbreviations: HR Head–neck combined frame ROI, NR Nasopharyngeal frame ROI, LNR Lower neck frame ROI, LCVR Lower cervical vertebral body frame ROI
Structural characteristics of setup errors: principal component analysis
To characterize the overall structural features of six-degree-of-freedom (6DoF) setup errors from a multivariate perspective, principal component analysis (PCA) was performed on the standardized translational and rotational error components. The PCA results showed that variance was distributed across multiple principal components, with PC1 and PC2 accounting for a substantial share of the total variance, while PC3 and PC4 also contributed non-negligibly The distribution of explained variance across principal components is illustrated in Fig. 3.
Fig. 3.
Scree plot showing the variance explained by each principal component. Note: Principal components are ordered by decreasing explained variance
Analysis of the principal component loadings indicated that translational and rotational error components contributed differently to the principal components. Rotational-related components (RTN, pitch, and roll) and translational-related components (LR, SI, and AP) exhibited distinct loading patterns across the principal components. The detailed loading matrix for each component is summarized in Table 4.
Table 4.
Principal component loading matrix for the 6DoF setup error components
| Component | PC1 | PC2 | PC3 | PC4 | PC5 | PC6 |
|---|---|---|---|---|---|---|
| SI | -0.04 | -0.48 | -0.67 | -0.19 | -0.14 | -0.51 |
| LR | 0.26 | -0.20 | 0.40 | -0.86 | -0.01 | -0.03 |
| AP | 0.34 | 0.37 | -0.61 | -0.29 | 0.00 | 0.54 |
| RTN | -0.62 | 0.21 | 0.00 | -0.24 | -0.70 | 0.12 |
| pitch | -0.24 | -0.71 | -0.01 | 0.06 | 0.09 | 0.65 |
| roll | 0.60 | -0.22 | 0.13 | 0.30 | -0.69 | 0.06 |
Loadings reflect the contribution of each error component to the corresponding principal component. Variables were standardized prior to analysis
In the principal component space, samples corresponding to different anatomical registration regions demonstrated partial overlap while also exhibiting differences in their distributional patterns. The distribution of setup error samples projected onto the first two principal components (PC1–PC2) is shown in Fig. 4.
Fig. 4.
PCA score plots in the PC1–PC2 space for the four registration regions. Note: Each point represents one registration instance. Panel A shows the overall distribution with all regions overlaid, and Panels B–E show the ROI-specific distributions using the same axis limits
Variation of setup errors over the radiotherapy course
To describe the variation of six-degree-of-freedom (6DoF) setup errors over the radiotherapy course, setup error measurements associated with different anatomical registration regions were summarized according to CBCT scan indices. The distributional characteristics of setup errors across successive CBCT scans are illustrated in Fig. 5.
Fig. 5.
Distribution of 6DoF setup errors across CBCT scan indices for different anatomical registration regions. Note: Data are summarized using medians and interquartile ranges (IQRs). CBCT scan indices represent the chronological order of image acquisition during treatment. Translational directions follow the patient-anatomical convention: +LR indicates left, +SI indicates superior, and + AP indicates anterior (negative values indicate the opposite directions)
Across different directional components, setup errors associated with all registration regions exhibited varying degrees of fluctuation over the radiotherapy course. For certain components, changes in median values were observed across CBCT scan indices, accompanied by substantial inter-individual variability, as reflected by the interquartile ranges. The extent of variability differed among directional components and registration regions, indicating that setup error behavior over time was not uniform across all dimensions.
These results provide an overview of the overall distribution and variability of setup errors across the treatment course and are intended to illustrate the heterogeneity of setup error behavior under routine clinical conditions, rather than to establish temporal trends or infer systematic changes.
Discussion
Based on CBCT data acquired during routine head-and-neck radiotherapy, we compared six-degree-of-freedom (6DoF) setup errors derived from four anatomical registration regions within the same CBCT acquisition. The error distributions were not identical across regions. Differences were more noticeable in several rotational components, while translational components varied less across regions.
These findings suggest that even when the same CBCT acquisition is used, the numerical values of reported setup errors can change with the chosen registration region. In practice, this means that setup error magnitudes should be interpreted together with the registration region used, rather than as region-independent measurements.
Across regions, rotational components appeared more sensitive to ROI selection than translational components. This pattern is consistent with the idea that the anatomical content within the ROI can influence how rotations are resolved during matching. Similar behavior of rotational setup components has been described previously, including those by Kaiser et al. [17], and Hirotaki et al. [18]. In line with these reports, our PCA suggests that the dominant variability patterns can be captured by coupled rotational and translational modes rather than isolated directional changes.
The head-and-neck region is anatomically complex, and ROIs differ in their spatial location and the mix of rigid and less rigid structures. These factors may partly explain why rotational error distributions differed between ROIs [19].That said, we did not examine the internal weighting of the registration algorithm or perform a structure-specific sensitivity analysis. Therefore, the mechanistic interpretation should be regarded as a plausible explanation rather than a demonstrated cause.
PCA was used here as a descriptive tool to summarize how the six error components co-vary. Based on the loading patterns (Table 4), PC1 was primarily driven by rotational components (RTN and roll), whereas PC2 was mainly driven by pitch together with SI, indicating that the dominant variability can be captured by clinically recognizable coupled modes rather than isolated directional changes. From a practical standpoint, this supports interpreting setup errors as coupled, multicomponent behavior rather than six isolated numbers.
In the PCA space, ROIs showed partial overlap but also displayed shifts in their distributions.The overlap indicates that a substantial portion of setup behavior is shared across ROIs under the same CBCT acquisition, whereas the shifts suggest ROI-dependent weighting of error patterns. Importantly, PCA is not used here to rank ROIs or to claim superior stability of any region; it only provides a compact description of multivariate structure [20].
In the descriptive analysis of setup error variation over the radiotherapy course [21], setup errors associated with different registration regions and directional components exhibited noticeable fluctuations across CBCT scan indices, accompanied by substantial inter-individual variability. Given that individual patients may undergo multiple CBCT scans during treatment and that setup errors can be influenced by a range of factors—including patient positioning, immobilization effectiveness, and patient condition—no inferential statistical analysis was performed for this component.
Accordingly, these plots are intended to illustrate variability under routine conditions rather than to demonstrate time trends or learning effects.From a clinical standpoint, ROI-dependent differences—especially in rotational components—may translate into meaningful geometric discrepancies for targets and organs at risk in the head-and-neck region, where steep dose gradients are common. Although the present study did not perform dose recalculation, the findings support incorporating ROI selection into routine interpretation of setup corrections and into department-level matching protocols. Workflow considerations are also relevant: a larger ROI may improve global robustness but can increase the chance of blending rigid and less-rigid anatomy, whereas a more focused ROI may better represent the intended anatomy but may be more sensitive to local deformation. Finally, CBCT imaging frequency reflects a balance between positional assurance and additional imaging dose and time; using a standardized ROI strategy may help optimize this balance in daily practice.
Several limitations of this study should be acknowledged [22]. First, this was a single-center retrospective analysis, and the results may be influenced by center-specific clinical workflows and patient characteristics. Second, although paired statistical methods were applied to compare setup errors across registration regions within the same CBCT acquisition, mixed-effects modeling approaches [23] were not employed to further account for the hierarchical structure of repeated measurements. Third, this study focused exclusively on setup error outputs and did not incorporate image features, dosimetric parameters, or internal registration algorithm characteristics, which limits mechanistic interpretation.In addition, LR was interpreted in the patient left–right sense, and the predominance of positive LR values indicates that lateral shifts tended to occur more often toward the + LR direction across the cohort. This may reflect a consistent setup tendency (e.g., patient head–neck positioning preference or immobilization-related asymmetry) rather than a ROI-specific effect, and was not specifically evaluated in the current dataset.
These limitations should be considered when interpreting the findings and assessing their generalizability to other clinical settings.
Despite these limitations, this study provides a systematic comparison of six-degree-of-freedom setup errors across different anatomical registration regions using a within-scan comparison framework. By focusing on both distributional and structural characteristics, the results highlight that registration region selection can influence the numerical presentation of setup errors in head and neck radiotherapy. These findings may serve as a reference for clinical setup evaluation and registration strategy selection in routine practice.
Conclusion
This retrospective study compared 6DoF setup errors derived from four ROI-based CBCT-to-CT registration strategies in head-and-neck radiotherapy. Setup error distributions differed across ROIs, with larger discrepancies observed in rotational components than in translational components. PCA identified coupled multicomponent patterns of variation, supporting interpretation of setup errors as structured behavior rather than six independent values. Clinically, registration results should be interpreted with explicit consideration of the selected ROI, as ROI choice can change the reported error characteristics. Further multi-center studies are warranted to confirm the generalizability of these findings.
Authors’ contributions
Weixiang Lin and Liangjie Xiao jointly conceived the study and drafted the main manuscript text. Junwei Chen was responsible for data curation and analysis, and prepared the figures for the paper. Xiaosheng Lin participated in experimental implementation and contributed to data collection. Zhanwei Li assisted with literature searching, organization of results, and manuscript revision and polishing. Jianlan Fang (Second Corresponding Author) provided guidance on study design and execution, supervised parts of the research, and critically reviewed the manuscript. Yongwen Fang (First Corresponding Author) led the project, oversaw the overall study design and direction, finalized the manuscript, and handled submission and correspondence. All authors reviewed and approved the final version of the manuscript.
Funding
This research received no external funding.
Data availability
The datasets generated and/or analyzed during the current study are not publicly available due to institutional and ethical restrictions, as the data contain detailed CBCT-based registration information derived from routine clinical workflows, which may pose a potential risk of patient re-identification even after anonymization. However, the data are available from the corresponding author upon reasonable request, subject to approval by the institutional ethics committee and in compliance with relevant data protection regulations. All statistical analyses were conducted using standard, well-established methods, and the analytical procedures are described in sufficient detail in the Methods section to support reproducibility of the study findings.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the Declaration of Helsinki. The study protocol was reviewed and approved by the Institutional Ethics Committee (Approval No. B2025-874-01). Given the retrospective nature of the study and the use of anonymized clinical data, the requirement for informed consent was waived by the Ethics Committee.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Weixiang Lin and Liangjie Xiao contributed equally to this work and should be considered co-first authors.
Contributor Information
Jianlan Fang, Email: fangjl@sysucc.org.cn.
Yongwen Fang, Email: fangyw@sysucc.org.cn.
References
- 1.Wen C, Jiqing S, Rui W, et al. Different setup errors assessed by weekly cone-beam computed tomography on different registration in nasopharyngeal carcinoma treated with intensity-modulated radiation therapy. Oncotargets Therapy. 2015;8:2545–53. 10.2147/OTT.S87159. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Cheo T, Yvonne C, et al. Measuring radiotherapy setup errors at multiple neck levels in nasopharyngeal cancer: A case for differential PTV expansion. Radiother Oncol. 2015;117(3):419–24. 10.1016/j.radonc.2015.09.032. [DOI] [PubMed] [Google Scholar]
- 3.Masi L, Casamassima F, Polli C, et al. Cone beam CT image guidance for intracranial stereotactic treatments: Comparison with a frame-guided setup. Int J Radiat Oncol Biol Phys. 2008;71(3):926–33. 10.1016/j.ijrobp.2008.03.006. [DOI] [PubMed] [Google Scholar]
- 4.Morley L, Court LE, Kiteley R et al. The Effect of Registration Volume Extent on Residual Errors Assessed Using Cone-Beam Computed Tomography in Radiation Treatment of Head and Neck Cancer,Journal of Medical Imaging and Radiation Sciences. 43, Issue 2,2012,Pages 95–102. ISSN 1939–8654. 10.1016/j.jmir.2012.01.002. [DOI] [PubMed]
- 5.Ding Y, Ma P, Li W, et al. Effect of surgical mask on setup error in head and neck radiotherapy. Technol Cancer Res Treat. 2020;19:1533033820974021. 10.1177/1533033820974021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Strbac B, Jokic VS. Evaluation of setup errors in head and neck radiotherapy using electronic portal imaging. Physica Med. 2013;29(5):531–6. 10.1016/j.ejmp.2012.12.001. [DOI] [PubMed] [Google Scholar]
- 7.Cubillos Mesías M, Boda-Heggemann J, Thoelking J, et al. Quantification and assessment of interfraction setup errors based on cone beam CT and determination of safety margins for radiotherapy. PLoS ONE. 2016;11(3):e0150326. 10.1371/journal.pone.0150326. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Mohandass P, Khanna D, Nishaanth B, et al. Impact of three different matching methods on patient setup error in X-ray volumetric imaging for head and neck cancer. Rep Practical Oncol Radiotherapy. 2020;25(6):906–12. 10.1016/j.rpor.2020.09.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Sikdar D, Krishnan AS, Namitha RS, et al. MV CBCT-based assessment of setup uncertainties and planning target volume margins in head and neck cancer. Rep Practical Oncol Radiotherapy. 2024;29(2):141–7. 10.5603/rpor.99361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.van Herk M. Errors and margins in radiotherapy. Semin Radiat Oncol. 2004;14(1):52–64. 10.1053/j.semradonc.2003.10.003. [DOI] [PubMed] [Google Scholar]
- 11.Tsiamas P, Bagher-Ebadian H, Siddiqui F, et al. Principal component analysis modeling of head-and-neck anatomy using daily cone beam CT images. Med Phys. 2018;45(12):5366–75. 10.1002/mp.13233. [DOI] [PubMed] [Google Scholar]
- 12.Kapanen M, Laaksomaa M, Tulijoki T, et al. Estimation of adequate setup margins and thresholds for position errors requiring immediate attention in head and neck cancer radiotherapy based on 2D image guidance. Radiat Oncol. 2013;8:212. 10.1186/1748-717X-8-212. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Chetvertkov MA, Siddiqui F, Kim J, et al. Use of regularized principal component analysis to model anatomical changes during head and neck radiation therapy for treatment adaptation and response assessment. Med Phys. 2016;43(10):5307. 10.1118/1.4961746. [DOI] [PubMed] [Google Scholar]
- 14.Liu J, Lyman KM, Ding Z, Zhou L. Assessment of the therapeutic accuracy of cone beam computed tomography–guided nasopharyngeal carcinoma radiotherapy. Oncol Lett. 2019;18:1071–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Robbins J, van Herk M, Eiben B, et al. Probabilistic evaluation of plan quality for time-dependent anatomical deformations in head and neck cancer patients. Physica Med. 2023;109:102579. 10.1016/j.ejmp.2023.102579. [DOI] [PubMed] [Google Scholar]
- 16.Jain V, Soni TP, Singh DK, et al. A prospective study to assess and quantify setup errors using cone-beam computed tomography in head-and-neck cancer image-guided radiotherapy. J Cancer Res Ther. 2023;19(3):783–7. 10.4103/jcrt.jcrt_2006_21. [DOI] [PubMed] [Google Scholar]
- 17.Kaiser A, Schultheiss TE, Wong JY, et al. pitch, roll, and yaw variations in patient positioning. Int J Radiat Oncol Biol Phys. 2006;66(3):949–55. 10.1016/j.ijrobp.2006.05.055. [DOI] [PubMed] [Google Scholar]
- 18.Hirotaki K, Tomizawa K, Moriya S et al. Impact of Anatomical Position Errors on Dose Distribution in Head and Neck Radiotherapy and Robust Image Registration Against Anatomical Changes. Anticancer Res. 2023;43(4):1827–1834. 10.21873/anticanres.16336. PMID: 36974799. [DOI] [PubMed]
- 19.Delishaj D, Ursino S, Pasqualetti F, et al. Setup errors in head and neck cancer treated with IMRT assessed by cone-beam computed tomography: A feasible protocol. Radiation Oncol J. 2018;36(1):54–62. 10.3857/roj.2017.00493. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Contesini M, Guberti M, Saccani R, et al. Setup errors in patients with head and neck cancer treated with IMRT: Impact on customized immobilization systems, patient pain, and anxiety. Radiat Oncol. 2017;12:72. 10.1186/s13014-017-0807-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Graf R, Boehmer D, Budach V, et al. Interfraction rotation of the prostate evaluated by kilovoltage X-ray fiducial marker imaging in intensity-modulated radiotherapy. Med Dosim. 2012;37(4):396–400. 10.1016/j.meddos.2012.02.006. [DOI] [PubMed] [Google Scholar]
- 22.De Kerf G, Claessens M, Mollaert I, et al. Machine learning–based treatment couch parameter prediction in support of surface-guided radiation therapy. Tech Innovations Patient Support Radiation Oncol. 2022;23:15–20. 10.1016/j.tipsro.2022.08.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Verbeke G, Molenberghs G. Linear Mixed Models for Longitudinal Data. Springer Series in Statistics. Springer; 2000. 10.1007/978-1-4419-0300-6.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The datasets generated and/or analyzed during the current study are not publicly available due to institutional and ethical restrictions, as the data contain detailed CBCT-based registration information derived from routine clinical workflows, which may pose a potential risk of patient re-identification even after anonymization. However, the data are available from the corresponding author upon reasonable request, subject to approval by the institutional ethics committee and in compliance with relevant data protection regulations. All statistical analyses were conducted using standard, well-established methods, and the analytical procedures are described in sufficient detail in the Methods section to support reproducibility of the study findings.