CBCT-based oART using the Varian Ethos system is conducted with a pre-defined workflow. The process begins with the planning CT (pCT), where the treatment intent—including dose prescription, planning objectives, and delineation of OARs—is established.

A reference treatment plan is generated on the planning CT using one of several predefined beam configurations: intensity-modulated radiotherapy (IMRT) with 7, 9, or 12 equidistant fields, or volumetric modulated arc therapy (VMAT) with two or three arcs. Once this reference plan is approved, it becomes available for daily treatment. Our early clinical experience indicated that VMAT plan calculation required considerably more time while offering only marginal dosimetric benefit compared with IMRT. For this reason, VMAT plans (Ethos 1.0) were not used in routine clinical practice at our institution.

At each treatment session, the patient is positioned and a CBCT scan is acquired. Following a quality check of the image, the system automatically propagates the planning contours to the CBCT of the day. These propagated contours must then be reviewed and, if necessary, edited by the user. Using deformable image registration, the CBCT anatomy is mapped back to the planning CT to preserve Hounsfield unit accuracy (synthetic CT). On this basis, two dose distributions are calculated: (1) the dose from the scheduled (non-adapted) plan applied to the current anatomy, and (2) a newly re-optimized adapted plan, generated using the original treatment intent and constraints, tailored to the anatomy of the day ( Figure 1 ).

The clinician then compares both plans and selects the one to be delivered. In practice, the adapted plan typically offers superior dosimetric quality, and at our institution it is selected in > 99% of sessions for treatment.

Between July 2023 and October 2024, a total of 72 patients were treated with the Ethos system at our institution. The majority of patients underwent pelvic radiotherapy, primarily for prostate cancer. Patients with primary prostate cancer radiotherapy are treated at our institution with the in-house protocol based on the CHHiP trial (14).

For this post-hoc analysis, we selected all patients with a confirmed diagnosis of prostate cancer, who were treated with the 2 SIBs concept at our institution and whose data could be fully exported from the Ethos system. These 49 patients had been treated prior to the commencement of this study, making this an exploratory analysis.

Planning target volume (PTV), SIB1, and SIB2 are structures derived from prostate and seminal vesicles contours, which is necessary for the adaptive treatment workflow since they will be automatically generated by the system based on the adapted prostate and seminal vesicle contours. SIB2 is defined as the prostate with 3 mm margins (posteriorly 0 mm). SIB1 includes the prostate and the proximal 1 cm of the seminal vesicles with 6 mm margins (posteriorly 3 mm). PTV is defined the same as SIB1 but includes the proximal 2 cm of the seminal vesicles with 6 mm margins in all directions including posterior. The cumulative prescribed doses for PTV, SIB1, and SIB2 are respectively 48 Gy, 57.6 Gy, and 60 Gy.

Dose objectives for OARs in this study were aligned with our institution’s in-house protocol (15), which is based on the guidelines from the CHHiP (14), PROFIT (16), PACE-B (17), and PACE-C (18) trials. In our institution, a posterior rectum wall (PRW) is used as an additional OAR (reasoning and PRW contouring have been described previously (19)).

To ensure better bladder sparing, the patients are instructed to follow our in-house “Bladder and bowel preparation instructions” (15), which aim at a reproducibly empty rectum and a comfortably full bladder.

As previously discussed (19), a systematic bias exists in which prostate contours on the pCT tend to be smaller than those on CBCT. This discrepancy does not indicate an error but arises from the ESTRO ACROP contouring guidelines (20), which recommend assuming equal levator ani muscle thickness adjacent to the prostate and rectum on CT, while on MRI (magnetic resonance imaging) these structures can be clearly distinguished. Consequently, MRI-based contouring yields smaller target volumes by avoiding unnecessary inclusion of the levator ani muscles. In CT-only workflows, the Santorini plexus is also frequently included due to limited soft-tissue contrast, further enlarging prostate, CTV, and PTV volumes.

At our institution, MRI is used to support pCT contouring but is not always referenced during adaptive workflows, occasionally leading to larger prostate contours in adapted datasets. Large discrepancies between the prostate contour volume on the pCT (used for the scheduled plan) and on the CBCT (used for the adapted plan) can introduce artifacts in dosimetric comparison (19).

To minimize variability and enhance data homogeneity, we applied a threshold of 15% to the pro-state volume for each session: Δ V = | V p r o s t a t e , p C T − V p r o s t a t e , C B C T | ≤ 15 % . This threshold allows to homogenize the data while accounting for physiological prostate swelling often observed during the radiotherapy (21). Fractions exceeding this threshold were excluded, resulting in the removal of 555 out of 980 fractions due to pronounced contour discrepancies (see Figure 2 for prostate volume distributions). The excluded fractions were analyzed separately. Consequently, six patients were entirely excluded from the study.

Additionally, three interrupted sessions were excluded. The final dataset comprised 422 fractions (ranging from 1 to 20 fractions per patient) from a total of 43 patients, providing a consistent basis for analysis. Among these 422 fractions, the scheduled plan was selected for treatment in only three sessions.

Since this is a retrospective study, all data was available prior to the beginning of the study. Dose and structure DICOM files were exported from the Ethos system. Dose-volume histograms were computed using a custom-developed Python script based on the dicompyler-core package (version 0.5.6) (22).

For each dose-volume metric, we calculated the difference between the metric value obtained with the scheduled plan and that obtained with the adapted plan for each fraction. To assess the significance of these differences, we applied the Wilcoxon signed-rank test for paired data with a two-sided alternative hypothesis.

Additionally, we evaluated the homogeneity index H I = ( D 1 % − D 99 % ) / D p r e s c r i b e d  (23) for SIB2 as well as the conformation number C N = T V 95 % / T V × T V 95 % / V 95 % for PTV, where T V 95 % and V 95 % are respectively the volume of PTV and the volume of tissue covered by 95% of the PTV prescribed dose, TV is the total volume of PTV (24). The CN quantifies both the target coverage (the first term of the formula) and the healthy tissue sparing (the second term).

To estimate both the central tendency and dispersion of non-normally distributed data, we calculated the Hodges-Lehmann median along with the corresponding quartiles (Q1 and Q3).

A custom Python script was developed for this analysis, utilizing core libraries such as NumPy, SciPy, and statistics. Given the exploratory nature of this study, p-values are considered descriptive, with p<0.05 interpreted as indicative of statistical significance. No Bonferroni correction was applied; instead, we always present an absolute p-value if p ≥ 0.001.

A total of 43 patients ( Table 1 ) with a confirmed diagnosis of prostate cancer were included in this study. Clinical staging revealed that 24 patients (55.8%) had T1 tumors, 18 (41.9%) had T2 tumors, and 1 patient (2.3%) was classified as T3. Androgen deprivation therapy was administered to 17 patients, depending on clinical indications and risk stratification. Adaptive radiotherapy was delivered in most cases using IMRT techniques. Most patients received either 9-beam or 12-beam IMRT; four patients were treated with different IMRT beam arrangements in different sessions, and one patient received either VMAT or IMRT, although all VMAT-treated fractions were excluded by the prostate-volume criterion.

Adaptation significantly enhanced target coverage as measured by D95% (p<0.001) for all targets, with the improvement ranging from 1.5 to 6.0%. (Hereafter, we estimate metric changes in terms of Hodges-Lehmann median of the difference distributions, all values refer to absolute dose changes, e.g. percentage points.) Additionally, V95% increased on average by 2.3 to 5.8% ( Figure 3 ; Supplementary Table S1 in the Supplementary Material ). Furthermore, the interquartile range (IQR = Q3 - Q1) decreased with adaptation for all target metrics.

The adapted plan demonstrated markedly improved homogeneity and conformality ( Figure 4 ). Specifically, the homogeneity for SIB2 (ideal value of 0) improved from 0.092[0.080, 0.115] (Hodges-Lehmann median [Q1, Q3]) to 0.056[0.054, 0.059] (p<0.001), with its maximum value decreasing from 0.75 to 0.11. The conformation number for PTV (ideal value of 1) increased from 0.66[0.64, 0.68] to 0.685[0.671, 0.699] (p<0.001), with its minimum value increasing from 0.53 to 0.62.

For the adapted plan, all alternative target constraints (except SIB2 Dmean) were satisfied in 418 out of 422 fractions: D95%≥ 95% and V95%≥ 95%. Only in four fractions did both SIB1 constraints fail, while those for SIB2 and PTV were consistently met ( Table 2 ).

Bladder V60Gy remained unchanged with adaptation, whereas V48Gy and V40Gy exhibited modest but statistically significant (p< 0.001) improvements ( Figure 5 ; Supplementary Table S1 in the Supplementary Material ). The percentage of fractions meeting the optimal constraints for the bladder metrics was higher with the adapted plan than with the scheduled one, and was ranging between 97.6% and 99.5% ( Table 2 ).

Among the evaluated rectum metrics (V56Gy, V52Gy, V48Gy, V40Gy, V32Gy, V24Gy, D2ccm), five showed statistically significant changes: the first four metrics experienced a slight deterioration (less than 0.8%) with adaptation, while V24Gy showed a minor improvement. Nevertheless, the adapted plan met all optimal rectum constraints in all fractions ( Table 2 ; Supplementary Table S1 in the Supplementary Material ).

Furthermore, the adapted plan outperformed the scheduled plan in terms of the PRW metrics: the dose to 2 ccm decreased by 0.08 Gy, the maximum dose was reduced by 0.17 Gy, and V37Gy improved by 0.65% (in all three cases p<0.001), while the percentage of fractions meeting the optimal constraint increased for V37Gy from 86% (with the scheduled plan) to 99% (with the adapted one), and for Dmax from 69% to 92% ( Table 2 ; Supplementary Table S1 in the Supplementary Material ).

The IQR decreased with adaptation for all bladder, rectum, and PRW metrics.

Bowel metrics did not exhibit any significant differences with adaptation.

When the prostate contour on CBCT exceeded the 15% threshold (e.g. a bigger prostate on CBCT, 485 sessions), median reductions in the target metrics D95% and V95% ranged from 4.5% to 14.9% ( Supplementary Figure S1 in the Supplementary Material ). The scheduled plan could not account for such a big prostate on the daily CBCT satisfying the goals for these target metrics in much fewer sessions in comparison with the adapted plan (see Table 2 ). For the sessions with a smaller prostate on CBCT (70 sessions), the adapted plan still conferred statistically significant dosimetric improvements over the scheduled plan, although the magnitude of benefit was reduced (between 0.8% and 2.3%) relative to the cases with a prostate volume close to V _pCT ( Supplementary Figure S2 in the Supplementary Material ). The scheduled plan could also fulfill the goals in the most sessions.

Moreover, an enlarged prostate contour on CBCT (and hence larger targets) artificially favored the scheduled plan for OAR metrics—they appeared slightly lower than with adaptation ( Supplementary Figure S1 in the Supplementary Material ), whereas the converse held true for a smaller prostate contour ( Supplementary Figure S2 in the Supplementary Material ).