Twenty postoperative patients receiving adjuvant radiotherapy at Subei People’s Hospital of Jiangsu Province (Yangzhou, China) from January 2022 to December 2024 were retrospectively enrolled (mean age: 43 ± 4.3 years). All patients had undergone breast-conserving surgery or modified radical mastectomy, with histopathological confirmation of stage I–III disease. Inclusion criteria: (i) postoperative left-sided breast cancer receiving conventional fractionated radiotherapy; (ii) availability of complete pCT, pre-treatment CBCT, and planning data; (iii) completion of full treatment course. Exclusion criteria: (i) poor image quality or severe artifacts; (ii) prior thoracic irradiation; (iii) severe cardiopulmonary comorbidities affecting dosimetric assessment. Ethical approval was obtained from the institutional review board.

All patients were immobilized in the supine position with both arms raised using vacuum cushions. Simulation CT was performed at 120 kV with 3 mm slice thickness. The prescription dose was 50 Gy in 25 fractions over 5 weeks. For selected patients, regional nodal irradiation (axillary/supraclavicular) was delivered with the same prescription. Radiotherapy techniques included intensity-modulated radiotherapy (IMRT) and conformal radiotherapy (CRT), ensuring that at least 95% of the planning target volume (PTV) received the prescribed dose.

Varian TrueBeam linear accelerator (Varian Medical Systems, USA); GE large-bore CT scanner (GE Healthcare, USA); Varian simulation localizer (Varian Medical Systems, USA); TPS version 15.6 treatment planning system; PTW Mephysto Navigator (PTW Dosimetry, Freiburg, Germany).

CBCT scans were obtained prior to the first treatment session using on-board imaging. For pCT,images were acquired with a large-bore CT scanner (120 kV, 80 mA, 25 ms, slice thickness 5 mm). CBCT acquisition parameters included: 100 kV, 15 mA, 150 mAs, 2 mm slice thickness, 360° rotation, 650 projections, and a 40×40 cm² field of view. A patient-specific HU-RED calibration curve was constructed for CBCT dose calculation, while pCT plans were calculated with the standard HU-RED curve. All CBCT and CT images were rigidly registered based on bony landmarks, followed by manual verification and minor adjustments as necessary. For supplementary analysis, regions of interest (ROIs) were selected in lung, soft tissue, and bone on both pCT and CBCT images after rigid registration. The ROIs were placed in relatively homogeneous areas and matched in anatomical location between the paired image sets. Mean HU values were extracted for each tissue ROI. For RED analysis, the corresponding mean HU values from CBCT were converted to RED using the patient-specific HU–RED calibration curve and compared with those derived from pCT. The ROI-based HU and RED analyses are presented in Supplementary Figures 1 and S2.

Treatment plans (IMRT and conformal radiotherapy [CRT]) were initially designed on pCT images and subsequently transferred to CBCT for direct dose recalculation, with all optimization parameters unchanged.

Dosimetric parameters for PTV included D2, D50, D98, Dmean, conformity index (CI), and homogeneity index (HI). OAR evaluation covered cardiac Dmean, V20, V30, V50 and bilateral lung Dmean, V20, V30, V50. Dose verification was performed with gamma analysis under 3%/3 mm and 2%/2 mm criteria, using a 10% dose threshold.

All contours were independently delineated by two senior radiation oncologists and reviewed for consistency. Statistical analyses were performed using SPSS 26.0. Paired t-tests or Wilcoxon signed-rank tests were applied as appropriate.

To determine equivalence between pCT- and CBCT-based dose calculations, the two one-sided tests (TOST) procedure was conducted. Equivalence margins (Δ) were predefined according to clinically acceptable dosimetric tolerances (PTV: ± 0.5 Gy; OARs: ± 2%). Statistical equivalence was established when the 90% confidence interval of the mean difference fell entirely within Δ and the TOST P-value was <0.05. Lin’s concordance correlation coefficient (CCC) and intraclass correlation coefficient ICC(A,1) were calculated to assess agreement. Bland–Altman plots were generated to evaluate systematic bias and limits of agreement. Statistical significance was set at P < 0.05.

CBCT- and pCT-derived HU-RED curves demonstrated similar trends, suggesting general consistency in tissue density quantification. However, greater HU discrepancies were observed in high-density tissues (e.g., bone) compared with low-density tissues (e.g., soft tissue and air) (Figure 1). In addition, ROI-based comparisons showed that the mean HU differences between pCT andCBCT for representative tissues (lung, soft tissue, and bone) were small, and the corresponding ROI-based RED differences derived from the HU–RED calibration were also limited (Supplementary Figures 1-S2).

In IMRT, CBCT plans yielded slightly higher D2, D50, and Dmean, with marginally lower D98 compared with pCT; none of the differences were statistically significant (P > 0.05). In CRT, CBCT-based D2, D50, D98, and Dmean were slightly lower than those from pCT, also without statistical significance. CI and HI values showed minimal variation between the two modalities under both techniques (Tables 1, 2).

In both IMRT and CRT, CBCT and pCT plans exhibited non-significant differences in cardiac Dmean, V20, V30, V50 and bilateral lung dosimetric parameters. Differences were small in magnitude, with CBCT values occasionally slightly higher or lower than pCT values (Tables 3, 4).

In this study, the absolute differences in dosimetric parameters between CBCT-based and CT-based plans were generally minimal. For the planning target volume, the mean deviations in PTV D2, D50, D98, and Dmean remained within 1.5% across all patients, with the largest discrepancy observed in a single case for D2 (1.53%). The overall dose–volume distributions of the PTV on both image sets were highly comparable, and no clinically meaningful hotspots or cold spots were identified.

For organs at risk (OARs), the differences between the two planning methods were similarly small. Mean deviations were consistently below 1.75%, and the greatest variation occurred in the cardiac V20 parameter, which reached a maximum of 2.11% in one patient. The DVH curves of the heart and lungs demonstrated almost overlapping patterns between CBCT- and CT-based plans, further supporting the high level of dosimetric agreement.

These findings collectively indicate that CBCT-based dose calculations exhibit a high degree of consistency with CT-based planning across both target volumes and critical organs (Figure 2). Representative dose distributions and DVH comparisons are presented in Figure 3. In addition, representative isodose overlays for two individual patients (Patient 1 and Patient 2) are provided in Supplementary Figure 3 to further visually support the consistency between CT- and CBCT-based dose calculations.

Dose verification software was employed to perform two-dimensional gamma analysis for both pCT- and CBCT-based plans, with the detailed results presented in Table 5. When the dose threshold was set at 10%, both treatment techniques demonstrated high gamma passing rates. For IMRT, the passing rates under the 3%/3 mm criterion were 97.8% ± 1.6% for the pCT plan and 97.5% ± 1.7% for the CBCT plan; under the more stringent 2%/2 mm criterion, the corresponding values were 93.6% ± 2.5% and 93.1% ± 2.8%. None of the differences were statistically significant (P > 0.05). Similarly, for CRT plans, the gamma passing rates under the 3%/3 mm criterion were 98.6% ± 1.2% (pCT) and 98.3% ± 1.3% (CBCT), whereas under the 2%/2 mm criterion, they were 94.5% ± 2.2% and 94.2% ± 2.3%, respectively, with no statistically significant differences observed (P > 0.05).

Equivalence between pCT- and CBCT-based dose calculations was assessed using the two one-sided tests (TOST). For all PTV-related dosimetric parameters—including Dmean, D98, D50, D2, HI, and CI—the mean differences and their 90% confidence intervals fell entirely within the predefined equivalence margins (Δ). All TOST P-values were below 0.05, demonstrating statistical equivalence between the two calculation methods. The mean differences were small (typically ≤ 0.1 Gy), and none of the 90% CIs approached the equivalence limits. Lin’s CCC and ICC(A,1) exceeded 0.98 across all PTV parameters, indicating excellent agreement. Bland–Altman analyses showed minimal bias with narrow limits of agreement, further supporting the equivalence findings.

Similarly, strong consistency was observed for the OAR parameters. Heart Dmean, V30, V50, as well as Lung Dmean, V20, and V30, all achieved statistical equivalence (TOST P < 0.05). The mean differences (0.10–0.30% or 0.10–0.12 Gy) were well below clinically meaningful thresholds. Both CCC and ICC ranged from 0.96 to 0.98, indicating excellent reliability. Bland–Altman plots revealed negligible systematic bias with clinically acceptable limits of agreement (Table 6).