Offline Plan Adaptation in Robustly Optimized CBCT-Guided Pencil Beam Scanning Proton Beam Therapy: Lessons From a Real-World Cohort

Abstract

Purpose

Pencil beam scanning proton beam therapy (PBS-PBT) is highly sensitive to anatomic variations, often necessitating adaptive replanning (ARP) to ensure accurate dose delivery. While cone beam computed tomography (CBCT) is increasingly used for image guidance with PBT, many centers still rely on periodic quality assurance computed tomography scans (QACTs) to trigger ARP. This study reports a large real-world experience with a CBCT-guided offline adaptive PBS-PBT workflow.

Methods and Materials

We retrospectively analyzed 300 consecutive patients treated with robustly optimized PBS-PBT. All patients underwent CBCT guided treatments, with QACTs performed either periodically (P-QACT) or triggered by CBCT (T-QACT) changes. We assessed ARP frequency, dosimetric and anatomic triggers, temporal patterns, indications, and workflow impacts across treatment sites. Strategic modifications were introduced after the first 100 patients, and their effect on ARP frequency was evaluated.

Results

Of 761 QACTs performed, 541 were P-QACT and 220 were T-QACT. Overall, 94 ARPs were performed in 80 patients (27%), with the highest rates in head neck (62%) and thoracic (43%) cancers. The primary dosimetric trigger was organ-at-risk overdosage (52%), followed by target undercoverage (32%) and a combination of both (16%). T-QACT demonstrated 97% (91/94) sensitivity for triggering ARP with 41.4% (91/220) specificity. Most ARPs (64%) occurred in the first half of the treatment, with 32% in week 3. Beam path changes (52%) were the most frequent anatomic trigger, followed by target deformation (29%) and setup inconsistencies (13%). Strategic modifications reduced ARP frequency from 35% in the first 100 patients to 22.5% in the subsequent 200.

Conclusions

PBS-PBT frequently necessitates ARP. A CBCT guided offline adaptive workflow, combined with stringent image review protocols, can largely eliminate the need for routine QACTs and streamline replanning. This approach offers practical guidance for PBS-PBT programs.

Introduction

Proton beam therapy (PBT) using pencil beam scanning (PBS) offers superior dose conformity and normal tissue sparing compared with photon-based techniques for most clinical sites. However, the characteristics of PBS-PBT make it acutely sensitive to interfractional anatomic variations, including weight loss, tumor regression, organ motion, and setup uncertainties. These changes can compromise target coverage or lead to organs at risk (OARs) overdosage if not identified and corrected during the treatment course.^1,2

Adaptive replanning (ARP) has therefore become a critical component of modern proton therapy. While many adaptive concepts have originated in photon therapy, their translation to PBT requires separate evaluation due to fundamental differences in beam properties, range uncertainties, and sensitivity to tissue heterogeneities. Offline adaptive PBT, relying on additional imaging to evaluate the need for plan modification, is currently the predominant approach worldwide. A recent survey reported that over 80% of proton centers perform some form of offline adaptation during treatment.³ In contrast, online ARP is only available at a few highly specialized institutions. Recent publications have demonstrated the feasibility of daily online adaptive PBT, and ongoing trials such as NCT06310655 are evaluating clinical workflows. However, large-scale adoption remains limited due to the lack of commercially available solutions, as well as the complexity of imaging, contouring, planning, and quality assurance within the time constraints of daily treatment.4, 5, 6, 7, 8, 9, 10

Imaging plays a central role in triggering adaptation. Historically, many proton centers without in-room volumetric imaging relied on periodic quality assurance (QA) computed tomography (P-QACT) to evaluate cumulative dose perturbations and determine the need for ARP. With the integration of cone beam computed tomography (CBCT) into newer PBT systems, daily volumetric image guidance has become feasible, enabling identification of anatomic variations at the time of treatment. However, QACTs are still required to quantify dose perturbations and to decide on ARP implementation, which is either triggered by CBCT findings (T-QACT) or performed periodically (P-QACT) according to institutional policy for each site. Importantly, despite the availability of CBCT on most modern PBT systems, P-QACT remains a common practice at many centers. Frequent QACT as part of the offline adaptive strategy is cumbersome as it not only affects the institutional workflow, requiring extra workforce and work-hours, but can also potentially impact the overall treatment time.^11,12 In addition, frequent QA imaging can add to the financial burden and expose patients to additional doses of radiation.¹³

In this study, we report our institutional experience of implementing a CBCT guided, offline adaptive PBS-PBT workflow in a real-world cohort of 300 consecutive patients. We describe the frequency, dosimetric and anatomic triggers, temporal patterns, and site-specific trends of ARP across diverse clinical indications. Furthermore, we evaluate the impact of strategic workflow modifications introduced after an internal audit, designed to improve efficiency and reduce the frequency of replanning. Our findings provide practical lessons for centers establishing or refining adaptive proton therapy protocols, highlighting both clinical implications and resource considerations.

Methods and Materials

We conducted a retrospective analysis of the first 300 consecutive patients treated with robustly optimized, image guided PBS-PBT at our center. Patients were selected for PBT through shared decision-making in multidisciplinary tumor board meetings, following a systematic institutional protocol with peer review before treatment (Fig. E1). The study was approved by the institutional review board and ethics committee.

Treatment planning

Standard immobilization and simulation techniques were applied as per institutional practice. Targets and OARs were delineated, and doses prescribed according to international guidelines using the RayStation treatment planning system (v.12A, RaySearch). Treatment plans aimed to deliver a Gross Tumour Volume (GTV) / Clinial Target Volume (CTV) D98 (dose to 98% of the volume) exceeding 98% of the prescribed dose on nominal plans, with robustness evaluated for CTV D95 >95% under 3 to 5 mm setup uncertainty and 3.5% to 4% range uncertainty. Pediatric patients requiring sedation were managed under general anesthesia according to institutional pediatric protocols.

Image guidance and adaptive workflow

Patients were treated with daily image guidance using kV x-rays and daily CBCTs, which were evaluated online by radiation therapists. Radiation oncologists performed additional offline reviews at least twice weekly using Adapt Insight/Mosaiq OIS. For tumors with motion, surface guidance with AlignRT (VisionRT) complemented standard motion mitigation strategies.

When anatomic variations were noted or anticipated, a QACT was performed to evaluate their impact on the integrity of the treatment plan. Two types of QACTs were undertaken: triggered QACTs (T-QACT), which were initiated in response to CBCT verification findings, and periodic QACTs (P-QACT), which were performed at the treating physician’s discretion. The frequency of P-QACTs varied, ranging from weekly in cases where changes in plan quality were anticipated at treatment initiation to additional scans during treatment when prompted by clinical concerns.

All treatments were delivered on an IBA ProteusPLUS system equipped with in-room CBCT, with planning on RayStation v12A. This hardware–software combination enabled integration of daily CBCT with QACT-based plan evaluation as summarized in Fig. 1, with additional operational details provided in Appendix E1 (Supplemental Material).

Figure 1.

A detailed workflow of our institutional proton beam therapy (PBT)-IGRT and adaptive replanning protocol. ¹Intracranial tumors—cone beam computed tomography (CBCT) twice weekly; other sites—daily CBCT. ²Patient setup accuracy, beam path changes, and target deformation reviewed. ³Offline review of CBCT by treating oncologist twice weekly and triggered quality assurance computed tomography scan (T-QACT) is performed, if needed. ⁴At physician’s discretion, these processes are elaborately described in Appendix E1.

Abbreviations: IGRT = Image-guided radiation therapy; PSQA = Patient-specific quality assurance.

QACT trigger criteria

QACT acquisition and escalation were initiated by radiation therapists based on daily CBCT findings, with final decisions on ARP made by the treating physician. Triggers were subjective and typically related to 1 or more of the following: gross CTV deformation (generally >3-5 mm change in any dimension), changes in beam path due to anatomic or density variations (such as weight loss >3%-5% of body weight, tumor shrinkage, sinus aeration affecting >30% of cavity volume, bowel or gastric filling, or pleural effusion), setup reproducibility issues (such as persistent shifts >3 mm across multiple fractions), or a combination of these factors. These parameters served as clinical prompts for additional imaging rather than rigid predefined thresholds.

QACT evaluation and ARP

All QACTs (both T-QACT and P-QACT) were imported into the treatment planning system and rigidly registered to the original planning CT. The target and organ-at-risk contours were copied from the original planning CT to QACT and were rationalized to incorporate anatomic variations. The nominal treatment plan was recalculated on the QACT data set using identical plan parameters. Dose distributions and dose–volume histograms were compared between the nominal plan and the recalculated plan to assess dosimetric adequacy.

The decision to generate an ARP was ultimately physician-dependent and based on multiple dosimetric considerations. Thresholds for initiating ARP were not uniform across physicians, reflecting real-world variations in clinical practice. These included inadequate CTV coverage, the emergence of new hotspots, or OAR overdosing compared with the nominal plan. Criteria for OAR overdosage were subjective and varied according to the organ type: more stringent for serial structures such as the spinal cord, brainstem, and optic apparatus; more flexible for parallel organs such as the lungs or parotids; and individualized for skin depending on beam arrangement and anticipated tolerance. All ARPs were generated using the same robust optimization parameters as initial plans and underwent repeat physician approval and patient-specific QA before clinical implementation. In cases where dosimetric deviations were identified that warranted ARP but remained only mildly suboptimal, treatment was continued until the ARP could be implemented. Conversely, if the evaluated plan was deemed unacceptable, treatment was interrupted until the ARP could be generated.

Data collection and analysis

Data on patient characteristics, treatment sites, immobilization devices, QACT timing and triggers (T-QACT or P-QACT), QACT outcomes, ARP frequency, reasons, and temporal patterns were extracted from the oncology information system (Mosaiq OIS) and analyzed using SPSS version 26. All available data were extracted, and no cases were excluded for incomplete records. QACT outcomes were documented as either acceptable (no ARP required) or requiring ARP. The anatomic triggers for ARP were classified as setup reproducibility-related, target/CTV deformation, beam path changes, or miscellaneous. Temporal patterns were analyzed by treatment week. Descriptive statistics were used to summarize ARP rates overall, by site, and for preaudit versus postaudit cohorts.

Workflow interventions

After treating the first 100 patients, an internal quality improvement audit was performed as part of our institutional proton therapy program's planned evaluation strategy. This audit, conducted independently of the current retrospective analysis, was triggered as a predetermined checkpoint to identify opportunities for optimization. Based on clinical patterns observed during this audit, several strategic modifications were implemented. These modifications included mandatory preplanning audits to streamline immobilization and minimize setup reproducibility-related causes for ARP; use of a higher number of beams (at least 3 beams) and more frequent use of noncoplanar beams; anatomic robust optimization; use of dose-assigned multifield optimization to reduce plan heterogeneities, and 4D robust optimization for moving targets. Trends in ARP frequency in the first 100 and the next 200 patients (before and after these changes were implemented) were compared.

Statistical analysis

Descriptive statistics were used to summarize patient demographics, treatment sites, and ARP events. Categorical variables, including tumor site, QACT type, and ARP triggers, were reported as frequencies and percentages, while continuous variables, such as the number of QACTs per patient and timing of ARP, were summarized using mean and range or median and interquartile range, as appropriate. The weekly distribution of ARP frequency was assessed using the Chi-square goodness-of-fit test. Comparisons of ARP rates between P-QACT and T-QACT and ARP frequency between the first 100-patient and the next 200-patient cohorts were performed using Fisher’s exact test. All statistical analyses were conducted using SPSS version 25.0 (IBM Corp.), and a 2-sided P value of less than .05 was considered statistically significant.

Results

Patient characteristics, treatment sites, immobilization details, and dose/fractionation schedules are summarized in Table 1. The median patient age was 41 years (range, 0.8-84), with 70% male. At presentation, most patients were adults; 67 (22%) were under 16 years, 56 (19%) were under 10 years, and 28 (9%) required sedation during simulation and treatment.

Table 1.

Patient characteristics, site of treatment, immobilization equipment description, and treatment dose/fractionation schedule

Characteristics	All patients (N = 300)	N (%)
Patient characteristics
Age	Years (median, range)	41 (0.8-84)
Gender	Male	210 (70%)
	Female	90 (30%)
Treatment characteristics
Site of treatment	Immobilization description
Brain	Mold care, thermoplastic mask (3-clamp)	105 (35%)
Head and neck	Mold care, thermoplastic mask (5-clamp) ± mouth bite	60 (20%)
Pelvis	Knee/ankle support, arm rest ± rectal balloon (prostate) ± vaginal obturator (vaginal vault)	48 (16%)
Cranio-spinal axis	Mold care, thermoplastic mask (3-clamp), vacloc	30 (10%)
Thorax	Vaclok + arm rest ± compression belt	30 (10%)
Abdomen	Vaclok + arm rest + knee rest ± compression belt	15 (5%)
Miscellaneous	-	12 (4%)
Dose and fractionation	Total dose (median, range) in CGE	58 (21-70.4)
	No of fractions (median, range)	29 (3-35)
	Dose per fraction (median, range) in CGE	2 (1.67-8)

Abbreviation: CGE = cobalt gray equivalent.

A total of 761 QACTs were performed (541 P-QACT and 220 T-QACT), averaging 2.54 per patient (range, 0-7). By subgroup, averages were 1.25 (range, 0-4) per sedated child, 1.87 (range, 1-6) per nonsedated child, and 2.8 (range, 1-7) per adult. Site-specific averages were 3.62 (range, 2-7) for head and neck, 3.2 (2-6) for abdomen, 3.08 (1-6) for pelvis, 2.77 (1-7) for thorax, 2.4 (1-6) for craniospinal irradiation (CSI), and 1.64 (0-4) for brain.

Of the 300 patients, 80 (27%) required ARP at least once. ARP frequency was highest in the head and neck (62%, P < .005) and thoracic (43%, P = .032) cohorts, with site-specific distributions shown in Fig. 2. Temporal patterns (Fig. 3) demonstrated that 51 of 80 ARPs (64%) occurred within the first 3 weeks of treatment, with a significant peak in week 3 (24 ARPs, 32%; P = .003). Of the 11 ARPs within the first week, 4 each involved CSI and head and neck cases. By week 3, ARPs were distributed across all sites. In the latter half of treatment, 29 ARPs (36%) were performed, predominantly in head and neck and pelvic patients, usually to correct cumulative OAR overdoses. Only 2 of the 80 patients requiring ARP (2.5%) experienced a 1-day treatment interruption; all others (97.5%) had no unscheduled delays.

Figure 2.

The distribution of patients undergoing adaptive replanning (ARP) with site-wise ARP frequency.

Abbreviation: CSI = craniospinal irradiation.

Figure 3.

Heatmap showing temporal patterns of site-wise adaptive replanning.

Abbreviations: ARP = adaptive replanning; CSI = craniospinal irradiation.

In total, 94 ARPs were generated for 80 patients: 70 patients (86%) required 1 ARP, while 10 (14%) underwent multiple ARPs, 7 (9%) with 2, 2 (3%) with 3, and 1 (1%) with 4. Multiple ARPs were significantly more frequent in the head and neck cohort (5 with 2 ARPs and 2 with 3; P = .0009), with single instances in thoracic (4 ARPs) and pelvic (2 ARPs) groups.

Of the 94 ARPs, 91 (97%) were triggered by T-QACT, while only 3 (3%) followed P-QACT without a CBCT trigger. This translated to 41.4% of T-QACTs resulting in ARP compared with just 0.6% of P-QACTs (P < .0001), a finding consistent across disease sites (Table 2).

Table 2.

Site-wise distribution of P-QACT/T-QACT and association with adaptive replanning

Site	QACT description	Number of QACT	Number of adaptive replanning	Number of ARP/QACT (%)	P value
All	P-QACT	541	3 (3%)	0.6	<.0001
	T-QACT	220	91 (97%)	41.4
Brain	P-QACT	152	0	0	<.0001
	T-QACT	20	13	65
Head neck	P-QACT	122	0	0	<.0001
	T-QACT	95	46	48.4
Pelvis	P-QACT	110	0	0	<.0001
	T-QACT	38	11	28.9
Craniospinal irradiation	P-QACT	60	1	1.7	.002
	T-QACT	12	4	33.3
Thorax	P-QACT	53	2	3.8	<.0001
	T-QACT	30	15	50
Abdomen	P-QACT	28	0	0	.487
	T-QACT	20	2	10
Miscellaneous	P-QACT	15	0	0	Cannot be evaluated
	T-QACT	6	0	0

Abbreviations: ARP = adaptive replanning; CT = computed tomography; P-QACT = periodic QACT; QACT = quality assurance CT scan; T-QACT = triggered QACT. Bold values indicate the higher yield of T-QACT for adaptive replanning compared to P-QACT.

Dosimetric triggers for ARP included OAR overdose in 49 cases (52%), target undercoverage in 30 (32%), and both in 15 (16%). By site, target dose compromise predominated in brain and CSI cohorts, while OAR overdose was the leading cause in head and neck and thoracic patients (Fig. E2). The median interval from QACT trigger to clinical implementation of ARP was 1 working day (range, 1-4).

Anatomic triggers leading to ARP included beam path changes in 52%, CTV deformation in 29%, and setup inconsistencies in 13%. Beam path changes were the predominant driver among head and neck, brain, and pelvic patients, whereas CTV deformation was more frequent in thoracic patients. Temporally, CTV deformation clustered in weeks 2 to 3, while beam path changes were more evenly distributed across the treatment course. Setup inconsistencies were most often identified and corrected in week 1. Figure 4a and b illustrates the distribution of ARP triggers by site and treatment week.

Figure 4.

(a) and (b) Anatomic triggers for site-wise adaptive replanning and the heatmap showing the temporal patterns.

Abbreviation: CSI = craniospinal irradiation; CTV = Clinical Target Volume.

After the first 100 patients, an internal audit prompted workflow modifications. ARP frequency subsequently decreased from 35% in the first 100 patients to 22.5% in the next 200 (P = .027). This reduction was most pronounced for brain, CSI, and pelvic cohorts (P < .05; Fig. E3).

Discussion

This retrospective analysis of 300 consecutive patients represents, to our knowledge, the largest reported real-world data set evaluating CBCT guided offline adaptive proton therapy with PBS-PBT. By systematically characterizing the triggers, timing, and outcomes of ARP, and by quantifying the impact of workflow modifications, this study provides practical insights for proton centers seeking to establish or refine offline adaptive protocols.

Nearly one-fourth of patients required ARP, which is consistent with prior reports documenting ARP frequencies between 10% and 30%.14, 15, 16 Within the head and neck cohort, 62% of patients required ARP, slightly higher than the 25% to 65% range reported in earlier studies.^2,16, 17, 18, 19, 20 This variation likely reflects differences in subsite distribution, dosimetric evaluation methods, and institutional thresholds for replanning. Deiter et al¹⁸ reported a 63% ARP frequency in head and neck cases when QACT-based dosimetric evaluation was used, whereas Stanforth et al¹⁶ observed only 25% when relying on CBCT-deformed planning CTs. In thoracic cancers, our 43% ARP frequency aligned with published reports of 24% to 61%.^21,22 Together, these findings highlight that although ARP frequency varies, systematic CBCT guided workflows reliably detect clinically meaningful dosimetric deviations.

Of 761 QACTs performed in our study, 12.4% prompted ARP, consistent with prior single-site reports. Biswal et al²³ documented 437 QACTs among 97 patients with head and neck cancer, while Hamza et al²² reported 360 QACTs in 80 patients with thoracic cancer. Importantly, T-QACT demonstrated substantially higher sensitivity for detecting the need for ARP compared to P-QACT: 97% of ARP triggers arose from T-QACT, while P-QACT contributed minimally in the presence of daily CBCT. Although the specificity of T-QACT was modest (41.4%), advances in CBCT quality and the development of synthetic CT dose reconstruction algorithms are expected to enhance predictive accuracy.²⁴ Notably, 3 ARPs in our cohort were prompted only by P-QACT—1 in CSI and 2 in thoracic tumors, reflecting limitations of routine CBCT in visualizing inferior spinal fields or subtle lung changes. These exceptions suggest that selective P-QACT remains justified in certain clinical scenarios.

The anatomic triggers necessitating ARP varied depending on the treatment site. They were primarily attributed to beam path changes (such as weight loss, sinus or bowel filling, pneumonitis, or meningocele), CTV deformation (including tumor response, edema, or cavity alterations), setup inconsistencies, or other miscellaneous factors (such as unexplained acute radiation toxicity or technical reasons preventing plan delivery). For instance, in head and neck cancer cases, common triggers included sinus filling, weight loss, CTV deformation, and setup inconsistencies. In contrast, for thoracic tumors, CTV deformation and adjacent lung changes were the predominant triggers.

We also analyzed the temporal distribution of ARPs based on the treatment site and the underlying anatomic triggers (Figure 3, Figure 4). Notably, 50% of ARPs occurred in the first half of the treatment schedule. In the first week, setup inconsistencies were the primary anatomic trigger, while during weeks 2 and 3, ARPs were frequently triggered due to CTV deformation (often an increase in target lesion volume) or changes in the beam path caused by bowel or sinus filling. In weeks 4 and 5, ARPs were mainly related to tumor response-related CTV deformation or weight loss. While the anatomic triggers evolved throughout the treatment course, patients with head and neck cancer required ARPs consistently across all treatment weeks, without a distinct temporal pattern.

Postaudit modifications after the first 100 patients significantly reduced ARP frequency from 35% to 22.5%. These strategic interventions were multifaceted, targeting both simulation and planning phases. First, mandatory preplanning meetings were instituted to optimize immobilization and simulation strategies, addressing setup inconsistencies—a frequent early trigger linked to our initial learning curve with PBT-specific devices (eg, custom neck rests, baseplates). These meetings facilitated tailored approaches, such as prone positioning, mouth bites, or avoiding vacuum bags, minimizing beam path variability.²⁵ Second, anatomic robust optimization was introduced for sites prone to density changes (eg, sinuses, bowel), incorporating synthetic CTs to simulate variations and enhance plan resilience.^26,27 Third, noncoplanar beams were more frequently employed to bypass unstable regions (eg meningocele, postoperative cavities), reducing beam path-related ARPs. Fourth, dose-assigned multifield optimization was implemented, assigning specific dose constraints to individual beams during optimization (a minimum dose, maximum dose, or occasionally both) to ensure that the individual beam dose contribution is not skewed, and is more similar to a single-field optimized plan with respect to plan robustness, which particularly benefited pelvic and head/neck cases. Finally, for moving tumors, 4D robust optimization integrated multiple breath-hold or 4DCT phases, improving 4D plan robustness.^28,29 Although this was effective, the optimization was time intensive, especially for large targets, and in selected patients, 4D robust evaluation was performed where a plan was first generated on an average scan and then evaluated across all 4D phases. All these strategic modifications required initial investment in training and planning time but yielded a sustained reduction in ARP across sites (Fig. E3), demonstrating their practical efficacy in streamlining workflow and enhancing treatment robustness.

The offline ARP workflow described here (Fig. 1) is resource demanding, involving repeat CTs, recontouring, dose recalculation, and patient-specific QA. This increased workload for physicians, physicists, and therapists, and required more treatment planning licenses, CT simulation slots, and treatment room availability. Such implications must be considered when new proton centers plan infrastructure and human resources. Nevertheless, treatment continuity was largely preserved, with only 2.5% of patients experiencing unscheduled delays, showing that offline ARP is feasible even in a high-volume environment.

This study has limitations. The retrospective design carries the risk of selection and reporting biases. Although this represents the largest data set of its kind, subgroup sizes for certain disease sites remain limited. Dosimetric thresholds for ARP were institution-specific and physician-dependent, potentially reducing generalizability. Moreover, while our data demonstrate improved efficiency after workflow modifications, we did not conduct a formal cost-effectiveness analysis. Such evaluation is essential to balance the trade-offs of improved robustness against added imaging dose, resource consumption, and financial burden. Despite these limitations, this study provides the largest real-world data set of CBCT guided offline ARP in proton therapy, outlining its frequency, triggers, and outcomes. Our findings show that structured workflows and strategic modifications can reduce ARP burden while maintaining treatment quality, offering practical guidance for both established and emerging proton centers.

Conclusion

PBS-PBT frequently requires ARP. In the context of rapidly evolving PBT, our study offers valuable workflow insights for implementing a practical and efficient offline adaptive strategy. The high sensitivity of CBCT-based triggers observed underscores that routine periodic QACTs may be largely redundant if meticulous daily CBCT imaging and stringent offline review protocols are adhered to. While online adaptation is likely to gain traction as technology and infrastructure mature, our findings present a scalable and resource-conscious approach for centers optimizing PBS-PBT workflows.

Disclosures

None.

Appendix. Supplementary materials

Adaptive_Supplemental File

Supplementary material

Figure 1: Institutional protocol for patient selection, peer-review and quality assurance

Figure 2: Dosimetric triggers for site-wise adaptive replanning

Figure 3: Overall frequency in first 100 and next 200 patients and site-wise distribution

Appendix-1: Institutional PBT-IGRT protocol and QACT evaluation protocol