On-board cone-beam computed tomography (CBCT) plays a considerable role in modern radiotherapy, as it is able to provide three-dimensional anatomy images with high-resolution. The use of CBCT facilitates patient positioning verification and target localization by anatomy alignment with reference to planning CT images, therefore substantially improving the accuracy, safety and effectiveness of image-guided radiotherapy (IGRT) (1–3). Despite the numerous advantages, CBCT comes at increased x-ray exposure compared with 2D radiographic imaging (4, 5). Since patients mostly receive recurrent CBCT scans over the entire treatment course, the health risk of imaging radiation induced by CBCT has raised serious concerns (6–11). For instance, it was estimated by Zhou et al. (8) that the average lifetime risk of cancer attributed to cumulative imaging doses per 100,000 individuals was 78 for brain cancer, 271 for lung cancer, and 510 for leukemia. A recent international survey on imaging practices in radiotherapy (12) shows that most centers used vendor-provided CBCT protocols in routine practice and less than 50% optimize the protocols to some extent. Growing efforts have been made to reduce imaging dose, and several strategies are recommended (4, 13, 14). A common and clinically available strategy of dose reduction is to reduce x-ray exposure via three techniques. The first technique is to use a short-arc scan over about 200 degrees instead of a 360-degree full-arc scan. The data acquired over short arcs are sufficient to reconstruct images of small objects, and therefore short-arc configuration has become the default for head-and-neck and pediatric protocols. The second is to decrease tube output (mAs) so as to reduce incident x-ray photons per projection, and the third is to increase angular separation to reduce exposure frequency. Several studies demonstrate that reduced exposure leads to CBCT image quality degradation. Wood et al. (15) evaluated mAs reduction on a pelvis phantom and made size-based recommendation of mAs settings. Yan et al. (16) investigated the relationship between the image quality and imaging dose and established a mAs-dependent empirical function. Lu et al. (17) evaluated the effect of reducing projection number on image quality and registration accuracy, concluding the projection-number reduction technique could achieve uncompromised registration accuracy under conscientious care. Similarly, Men et al. (18) evaluated the image quality, dose and registration errors acquired with different CBCT angular separations, indicating that CBCT images with fewer projections impair image quality and registration accuracy. Olch et al. (19) compared the image quality and registration accuracy between kV-pair imaging and low-dose CBCT acquired with reduced mAs and projections. The study demonstrates that CBCT dose can be reduced to a level comparable to kV-pair imaging with similar or enhanced patient positioning accuracy.

It is important to note that all the aforementioned studies are all phantom-based without any patient-involved validation. To our knowledge, there were only two relevant clinical study on low-dose CBCT for IGRT. The first study was by Alcorn et al. (20), which was an international study to evaluate the feasibility of a pre-defined low-dose CBCT protocol in pediatrics that received central neural system radiotherapy. The other study was by Bryce-Atkinson et al. (21) using CBCT scans from 7 pediatric patients to simulate low-dose scans by either adding noise or halving the number of projections. Due to the limited patient category, imaging site and cohort size, further validation is required. To this end, this study aims to comprehensively evaluate the image quality, image dose, and registration accuracy of CBCT images reconstructed using projections acquired with different combinations of angular range. The key innovation and highlight are: 1) we used a virtual clinical trial-based approach retrospectively enrolling a total of 71 real patients, and 2) we evaluated the clinical viability of low-dose CBCT for head, thorax and pelvis IGRT.

Figure 1 shows representative reconstructed images of real patients that received head (a-row), thorax (b-row) and pelvis (c-row) scans respectively with the default (S₁ ) and projection-reduced protocols (S_1/2, S_1/4, S_1/8 ). As we reduced the projection number, image quality deteriorated as the dose was lowered. Compared with the full-dose reference images (left-column in Figure 1 ), image noise and strike artifacts in dose-reduced protocols became obvious, which were induced by increased angular separations. While image details were blurred, the loss of soft-tissue contrast was relatively greater than that of bony structures.

The 6DoF registration errors of projection-reduced head, thorax, and pelvis scans relative to the full-dose scans are summarized in Figure 2 , showing the median, mean ± standard deviation (SD) and 25–75% interpercentile range (Q1-Q3). We can clearly see that, regardless of imaging site or scan protocol, the mean and median values of relative errors across six directions were quite close to zero, ranging within ±0.4 mm in translation and within ±0.2 degrees in rotation. We use Q1-Q3 and mean ± SD to indicate uncertainties in relative errors. For translational directions, the Q1-Q3 values across all protocols ranged within ±0.5 mm in translation except that in LNG for pelvis scan protocols (± 0.7 mm), and a similar trend was identified for mean ± SD. For rotational directions, the uncertainties in head protocols were relatively larger than in those in thorax and pelvis protocols. This may be attributed to the uncertainties induced by cervical/neck flexion. In addition, the couch shifts of S_1/2, S_1/4 and S_1/8 were paired with those of S₁ in each direction to perform Wilcoxon paired signed-rank test. The p-values across all the tests were > 0.05, among which the smallest value was about 0.17, indicating the couch shift differences between projection-reduced protocols and the full-dose protocol were not statistically significant.

The CNR, Tx and Rx values over the testing patients were summarized in Table 2 and Figure 3 . As the projection numbers were reduced, the imaging dose decreased, leading to significant descending in CNR performance and ascending in Tx and Rx values. For CNR, while the S_1/2 protocol exhibited inferior performance to the full-dose protocol, the mean values across the three sites were >3.0, indicating the soft-tissue contrast performance was generally acceptable even when the projection number was halved from the default setting. For Tx, the mean values were mostly less than 1 mm except for S_1/8 protocol in pelvis scan. This is similar to the trend in Figure 2E , where S_1/8 exhibited the largest uncertainties. For Rx, while the mean values for head scans were between 0.4 and 0.5 degree, those for thorax and pelvis scans were all less than 0.4 degree.

Reducing imaging dose of CBCT has been a prominent concern due to the potential health risks associated with recurrent CBCT scans over the entire treatment course. A direct strategy to reducing dose is to obtain projections as few as feasible through expanding the angular gap in sampling, but with the challenge of ensuring that image quality and registration accuracy are not compromised. In the meantime, how to effectively evaluate the impact of projection reduction on image quality and registration accuracy in the clinical context without impairing treatment accuracy is a challenging issue as well.

Our study proposed a virtual clinical trial-based strategy to perform comprehensive clinical evaluation of projection-reduced low-dose CBCT for IGRT application. In this work, we enrolled a total of 71 real-world patients to quantitatively evaluate image quality and registration accuracy for head, thorax and pelvis CBCT scans. Hence, this virtual clinical trial-based study is more accurate in reflecting the clinical context than previous studies that are either highly controlled (20, 21) or phantom-based (15–19). In the meantime, we used the validated open-source research toolkit for image reconstruction (TIGRE-VarianCBCT) and registration (3D Slicer), justifying the reproducibility of all the results. In this context, our analysis yielded important insights.

Firstly, while the reduction of projection numbers led to increased image noise and strike artifacts, acceptable clinical image quality was maintained up to a decrease by half from the default projection number. This was justified by the CNR results in Table 2 , demonstrating the potential to reduce the imaging dose without significant loss in image quality. However, as the projection number further decreased to 1/4 and 1/8 of the full dose, the CNR values fell below the acceptable threshold, implying substantial deterioration in soft-tissue contrast. This also indicates a limitation in the extent of projection-reduced low-dose CBCT while preserving adequate imaging quality.

Secondly, our study showed that the relative registration errors of projection-reduced CBCT scans compared to full-dose scans were negligible across six directions with mean error< 0.5 mm in translation and<0.4 degree in rotation. Moreover, the error magnitude was< 1.0 mm in translation (Tx) and< 0.4 degree in rotation (Rx). This further demonstrates that a reduction in projection numbers does not necessarily impair the registration accuracy and suggest the maintenance of registration accuracy even at reduced doses, the key concern in IGRT. This finding is in line with those in phantom-based studies by Men et al. (18) and Olch et al. (19). Also, as the projection number was further reduced, slight increase in uncertainties were observed, especially in the LNG direction in both thorax pelvis scans. This phenomenon we assume can be attributed to subtle pelvic anatomy changes along the LNG direction.

Moreover, it is worth noting that no statistically significant shift differences were identified between the dose-reduced protocols and the full-dose reference. The consistency of our findings with the literature (19, 21) reinforces the reliability and potential clinical viability of the projection-reduction strategy.

Our study has several limitations. First, we used auto-matching in CBCT-to-CT registration to minimize any personal bias. However, in clinical practice, therapists or physicists are probably to perform couch shifts with manual corrections. There may be some discrepancy here between the research world and how the shifts would look in a true clinical setting. Second, while the strategy of virtual clinical trial facilitated patient enrollment [for instance, only 6 patients were enrolled in the study by Bryce-Atkinson et al (21)], due to the limitation of the IRB approval, we only enrolled adult patients. It is well known that age is a key factor attributable to estimating patients’ risks (13), as it is pediatric patients that will benefit the most from imaging dose reduction. While we believe some findings in this study are also applicable to pediatric patients, further efforts including more extensive clinical trials are required for solid validation. Thirdly, while we used a virtual clinical trial design to ensure clinical applicability, further prospective clinical trials are needed for a more robust and comprehensive validation, especially in combination with other low-dose techniques. Besides, the retrospective study design, scan sites and the dose-reduction method are also limitations.

Our study indicates that while the impact of dose-reduction on CBCT couch shifts is not significant, the impact on CNR values is significant. Further validation on optimizing CBCT imaging dose is required.

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding authors.

The studies involving humans were approved by Peking University Cancer Hospital. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements.

MW: Writing – original draft. KY: Writing – original draft. YZ: Writing – review & editing. JG: Writing – review & editing. XZ: Writing – review & editing. ZL: Writing – review & editing. YL: Writing – review & editing. HW: Writing – review & editing. YD: Writing – review & editing.