Clinical implementation and impact of real-time Cherenkov imaging for treatment verification and patient safety in radiation therapy

Graphical abstract

Highlights

• •Cherenkov imaging was integrated into routine radiation therapy at two centers.
• •Fifteen percent of 1,196 treatment plans showed actionable deviations.
• •Most deviations were detected within the first three treatment fractions.
• •Real-time imaging prompted replanning in 1.8% and workflow changes in 2.5%
• •Early-fraction review is a practical model for clinical implementation.

Abstract

Background and purpose

Real-time Cherenkov imaging (CI) visualizes surface dose distribution during external beam radiotherapy (EBRT). Although feasibility has been demonstrated, sustained multi-site implementation with predefined clinical impact metrics remains less well characterized. We evaluated clinic-wide CI integration across two centers over two years and quantified clinical impact, timing of detection, and corrective actions.

Materials and methods

CI (DoseRT, Vision RT) was incorporated into routine treatment delivery on Varian TrueBeam linear accelerators. All techniques (3D, IMRT, SRS/SBRT) were monitored prospectively. Findings were assigned the highest plan-level clinical impact category: 0–1 (informational/non-actionable), 2 (setup refinement), 3 (workflow/planning modification), and 4 (immediate intervention). During implementation, mandatory physics review of CI images from the first five fractions of new treatment starts was adopted. Outcomes included actionable rate (Categories 2–4), time to first detection, and corrective endpoints (replanning, workflow changes).

Results

Among 1,196 plans, 15% were actionable (Category 2: 8%; Category 3: 4%; Category 4: 3%). Twenty-one plans (1.8%) required partial or full replanning and 30 (2.5%) prompted workflow or policy modifications. Approximately half of Category 4 deviations were identified during beam-on and required immediate interruption and correction. Deviations were concentrated in early fractions: 76% were detected by fraction 3 and 89% by fraction 5 (median 2.6).

Conclusions

Clinic-wide CI integration is feasible and identifies actionable deviations in 15% of treatment plans, predominantly within the first five fractions. Real-time visualization supported immediate intervention and system-level workflow refinement without measurable disruption to clinical throughput, supporting CI as a complementary verification layer to IGRT/SGRT for patient safety.

1. Introduction

Modern external beam radiation therapy (EBRT) is delivered within a multilayered quality assurance (QA) framework that includes treatment planning review, image-guided radiotherapy (IGRT), surface-guided radiotherapy (SGRT), and routine physics oversight. These safeguards provide geometric validation and pre-treatment dosimetric assurance [1], [2], [3], [4] but do not directly visualize the interaction of delivered radiation with patient surface anatomy during beam-on. As treatment techniques become increasingly conformal and margins decrease, confirmation that delivered beam placement aligns with planning intent in real time remains clinically relevant.

Cherenkov imaging (CI) enables visualization of optical emission generated when high-energy charged particles traverse tissue during radiation delivery [5], [6], [7]. Time-gated camera systems synchronized to linear accelerator pulses isolate Cherenkov emission under ambient lighting conditions, permitting acquisition during routine treatment delivery [8]. The resulting images provide a qualitative representation of the surface dose distribution and field geometry [6], [9], [10], providing a verification perspective distinct from geometric alignment tools.

Initial clinical reports demonstrated feasibility of CI during EBRT and suggested applications in surface dose monitoring [8], [11]. More recently, clinic-wide deployment studies have described quality improvement opportunities identified through routine CI use [6]. However, much of the literature has emphasized feasibility, spatial conformity analysis, or retrospective case discovery. Less emphasis has been placed on structured, sustained multi-site implementation using predefined clinical impact categories, temporal detection analysis, and documented corrective endpoints embedded within routine clinical workflow.

Our group previously reviewed the evolution of CI in radiation therapy [12] and reported a case demonstrating detection of insufficient skin flash during breast VMAT [10]. Here, we report a structured two-year implementation across two independent radiation oncology centers. The objectives were to quantify plan-level clinical impact using predefined categories, characterize timing of deviation detection, evaluate escalation patterns, and assess how CI influenced corrective action and workflow refinement under sustained clinical use.

2. Materials and methods

2.1. Clinical sites and equipment

Cherenkov imaging (CI; DoseRT, Vision RT, London, UK) was implemented at two radiation oncology centers: AdventHealth Celebration (Florida) and AdventHealth Parker (Colorado). Both sites operate Varian TrueBeam linear accelerators equipped with time-gated CI capability integrated alongside existing SGRT infrastructure.

Implementation occurred within routine clinical operations without dedicated research personnel. This retrospective implementation review was conducted under an institutional quality improvement framework. Data collection and analysis were performed in accordance with applicable institutional policies. DoseRT did not experience unplanned downtime during the study period; all plans meeting institutional monitoring criteria were monitored as intended. In the event of an equipment or software issue, the predefined contingency workflow was to proceed with standard-of-care treatment delivery and resume CI monitoring on subsequent fractions once functionality was restored. Occasional non-capture events (e.g., room lighting outside recommended limits, camera not armed, or planned non-use) were infrequent and not systematically recorded in this retrospective implementation review.

2.2. Acquisition parameters and image review

CI images were acquired using a pulse-synchronized, time-gated camera system integrated with the linear accelerator. All acquisition parameters, including synchronization timing, exposure settings, gain, and geometric calibration, were established during system installation and commissioning and were not user-adjustable during routine clinical operation. The only operational variable under clinical control was ambient room lighting, which was maintained within manufacturer-recommended limits to preserve image contrast while allowing safe patient setup and monitoring.

Real-time CI was displayed at the treatment console during beam delivery and observed by radiation therapists for all monitored fractions. Therapists evaluated CI signal patterns relative to expected field geometry, accessory placement, and anatomic landmarks. Deviations from anticipated emission distribution, including unexpected surface illumination, absence of expected signal, or asymmetric patterns, were documented and escalated according to the predefined clinical impact framework.

Midway through the implementation period, a structured multidisciplinary review process was formalized. For all new treatment starts, physics review of CI images from the first five fractions was required. Physicians were consulted when deviations suggested potential target coverage compromise or normal tissue exposure. This early-fraction review process facilitated consistent interpretation, standardized category adjudication, and documentation of corrective actions.

The DoseRT system provides two-dimensional surface Cherenkov emission images rather than three-dimensional dose reconstruction. System output includes a time-synchronized video replay showing real-time Cherenkov signal overlaid on the patient surface during beam-on and a cumulative signal intensity map for the entire delivered fraction. These outputs provide a qualitative, surface-based representation of delivered field geometry and relative superficial dose patterns. Clinical interpretation was based on pattern comparison relative to expected field geometry and anatomic landmarks. Deviations such as unexpected surface illumination outside the intended treatment region, absence of expected signal, or inconsistent patterns across fractions served as action triggers prompting review of setup, accessories, imaging alignment, and planned beam geometry. When observed patterns suggested possible impact on target coverage or unintended normal tissue exposure, cases were escalated to physics and, when indicated, to the treating physician, with target and organ-at-risk implications evaluated through the standard treatment planning process. CI thus served as the detection and verification modality rather than the dosimetric calculation tool.

CI findings were documented in a tiered manner. Radiation therapists flagged incidents within the DoseRT system interface to facilitate follow-up review. Physics was notified and events were investigated and assigned a clinical impact category with the multidisciplinary team as needed. Events were documented in an internal quality and implementation log; items resulting in changes to clinical care (e.g., re-simulation, partial or full replanning, or physician-directed treatment modification) were additionally documented in the patient chart consistent with standard clinical practice.

2.3. Patient and plan cohort

Over a two-year period, 1,196 treatment plans were monitored across both centers: 550 three-dimensional conformal radiotherapy (3DCRT) plans, 534 intensity-modulated radiation therapy (IMRT) plans, and 112 stereotactic (SRS/SBRT) plans (Table 1).

Table 1.

Distribution of monitored treatment plans across two radiation oncology centers during the two-year study period. CI was applied uniformly across 3D, IMRT, and SRS/SBRT techniques.

Technique	Number of Plans	Percentage (%)
3D Conformal (3D)	550	46.0
IMRT	534	44.7
SRS/SBRT	112	9.3
Total	1196	100

Because most patients were treated with a single plan, plan-level analysis was considered representative of patient-level impact. CI monitoring was applied uniformly across techniques, disease sites, and fractionation regimens. For SRS/SBRT patients, the CI review workflow was effectively more intensive than the standard first-five-fraction physics review protocol because these courses contain five or fewer fractions. For SBRT courses (3–5 fractions), CI monitoring was performed by the radiation therapists during delivery for each fraction, with physics perform in room, real-time oversight for the entire first fraction and post fraction CI review for the remaining fractions, effectively encompassing the full treatment course. For single-fraction SRS, physics was present for the entire delivery with real-time CI review, enabling immediate detection and correction if an unexpected signal pattern was observed. Patient-level demographic variables were not analyzed in this implementation-focused review; future prospective studies may incorporate demographics to assess generalizability across patient populations.

2.4. Workflow integration and escalation framework

CI was prospectively intgrated into routine treatment delivery. During beam-on, therapists observed real-time Cherenkov images at the treatment console. Deviations from expected emission patterns were documented and escalated according to the predefined clinical impact framework.

Findings were evaluated both during treatment delivery and during structured multidisciplinary review. Early implementation experience demonstrated clustering of actionable deviations within initial treatment fractions, prompting formal adoption of mandatory physics review of Cherenkov images from the first five fractions for all new treatment starts (Fig. 1). This policy institutionalized consistent early-course evaluation and standardized escalation pathways across providers.

Fig. 1.

Schematic representation of CI integration into routine treatment delivery. Real-time therapist monitoring during beam delivery is supplemented by structured multidisciplinary review during the first five fractions. Findings are categorized according to predefined clinical impact levels, with escalation pathways guiding corrective action.

2.5. Clinical impact classification

Each plan was categorized according to the highest impact level reached: (Category 0–1) informational or non-actionable; (Category 2) minor deviation requiring setup refinement; (Category 3) moderate deviation requiring workflow or planning modification; (Category 4) major deviation requiring immediate intervention or treatment interruption. Categories 2–4 were considered clinically actionable.

If multiple findings occurred, the highest category was recorded. Timing of first detection was defined as the fraction during which the highest-impact deviation was first identified. Descriptive statistics were used for categorical variables. No inferential comparisons were performed given the implementation-focused study design.

Category assignment was based on the following criteria. Category 2 (minor deviation, setup refinement) was assigned when a CI deviation was attributable to patient positioning or accessory placement and was correctable through therapist-directed repositioning (e.g., arm, shoulder, or head adjustment, immobilization re-indexing, or minor bolus or accessory alignment correction) without changing the treatment plan or documented workflow. Category 3 (moderate deviation, workflow or planning modification) was assigned when the deviation could not be reliably prevented through simple repositioning and instead required a system-level corrective action, such as changes to setup instructions, immobilization strategy, simulation process, accessory placement policy, or partial or full replanning. Category 4 (major deviation, immediate intervention) was assigned when the deviation required same-day treatment interruption with potential implications for target coverage or organ-at-risk exposure.

Medical physics reviewed CI images for all new treatment starts (required for the first five fractions after workflow formalization) and led the preliminary categorization for actionable findings. Radiation oncologist input was obtained when deviations suggested potential compromise of target coverage or unexpected normal tissue exposure. Final adjudication followed consensus discussion between the attending physicist and treating physician. Standardization across the two clinical sites was ensured through a shared clinical impact framework (Categories 0–4) applied at both centers, fixed acquisition parameters established at commissioning and not user-adjustable during clinical use, a shared early-fraction review policy with consistent escalation triggers, and periodic joint case review between sites to harmonize interpretation. Clinical significance was defined operationally as any CI finding that triggered a corrective clinical action beyond documentation (Categories 2–4).

2.6. Ethics approval

This retrospective quality improvement review was conducted under institutional review board oversight at our institution. Informed consent was waived in accordance with applicable institutional policies for retrospective quality improvement studies. All data were de-identified prior to analysis.

3. Results

3.1. Plan-level clinical impact

Of 1,196 monitored plans, 85% were classified as Category 0–1 and required no intervention. Actionable deviations (Categories 2–4) occurred in 15% of plans, including 8% Category 2, 4% Category 3, and 3% Category 4. Major deviations were therefore uncommon but clinically significant. Actionable findings were observed across all treatment techniques without clustering within a single modality (Fig. 2).

Fig. 2.

Distribution of highest-impact CI categories per treatment plan (n = 1196). Categories 2–4 were considered clinically actionable. The majority of plans (85%) demonstrated no actionable deviation (category 0–1).

3.2. Corrective interventions

Twenty-one plans (1.8%) required partial or full replanning based on CI findings. Thirty plans (2.5%) prompted workflow modifications, including adjustments to immobilization, accessory placement, setup instructions, or simulation protocols (Table 2). Replanning and workflow adjustments represent subsets of Category 3–4 deviations. Category 3–4 deviations were observed across disease sites, with breast plans accounting for the largest proportion (55%), followed by pelvis (15%), lung and abdomen (10%), brain (5%), and other sites (15%) (Table 3).

Table 2.

Corrective actions initiated following CI-detected deviations. Replanning included full or partial plan revisions. Workflow modifications include changes to setup protocols, immobilization strategy, or accessory placement. Percentages are calculated relative to total monitored plans (n = 1196).

Corrective Action Type	Number of Plans	Percent of Total Plans (%)
Setup refinement only (Category 2)	96	8.0
Workflow/planning modification (Category 3)	48	4.0
Immediate intervention (Category 4)	36	3.0
Replan (partial or full)	21	1.8
Workflow adjustment (policy/process)	30	2.5

Table 3.

Distribution of Category 3–4 plans by disease site (n = 84). Percentages are calculated relative to the total number of Category 3–4 plans. Breast plans accounted for the majority of major deviations, consistent with the high representation of breast treatments in the monitored cohort and the inherent sensitivity of CI to surface-level deviations in breast radiotherapy.

Disease Site	Category 3 (n)	Category 4 (n)	Total (%)
Breast	27	19	46 (55%)
Pelvis	8	5	13 (15%)
Lung/Abdomen	4	4	8 (15%)
Brain	2	2	4 (10%)
Other	7	6	13(5%)
Total	48	36	84 (100%)

Approximately half of Category 4 deviations were identified during beam delivery and required immediate interruption and repositioning. The remainder were identified during structured multidisciplinary review and corrected prior to subsequent fractions. No recurrent deviations of the same type were observed within individual treatment courses following intervention, suggesting durable resolution after detection.

3.3. Temporal detection pattern

Deviations were concentrated in early treatment fractions (Fig. 3). Seventy-six percent were identified within the first three fractions and 89% within the first five fractions (median 2.6). Late-course first detections were uncommon and were generally associated with accessory positioning changes or patient comfort-related adjustments during longer treatment courses.

Fig. 3.

Cumulative fraction of plans with first detected CI deviation as a function of treatment fraction number. 76% of deviations were identified within the first three fractions and 89% within the first five fractions. The median fraction at first detection was 2.6.

3.4. Escalation of findings

Approximately 60–70% of Category 3–4 plans initially presented as Category 1 findings during real-time delivery, with multidisciplinary review prompting reclassification. Subtle surface-emission irregularities frequently triggered further evaluation, occasionally revealing clinically significant deviations requiring intervention.

This escalation pattern highlights the value of structured multidisciplinary review in distinguishing informational findings from clinically actionable deviations within a real-time monitoring framework.

3.5. Representative clinical cases

Category 1 – Verification of Intended Sparing: During deep inspiration breath-hold breast treatment, CI consistently demonstrated absence of signal outside the expected treatment area, confirming appropriate beam geometry and normal tissue sparing.

Category 2 – Minor Setup Refinement: During right breast irradiation, CI revealed signal on the lower right arm above the axilla following minor patient repositioning. Treatment was paused and the arm repositioned; no further deviations occurred.

Category 3 – Workflow Modification: Unintended contralateral breast and chin CI signal prompted review of simulation alignment, treatment planning, and setup reproducibility. Workflow adjustments improved reproducibility, and subsequent fractions demonstrated expected CI signal patterns.

Category 4 – Major Deviation: In a previously reported case [10], absence of expected CI signal revealed inadequate flash margin during breast VMAT. Treatment was interrupted and replanning was performed to restore appropriate coverage. Representative examples are shown in Fig. 4.

Fig. 4.

Representative Cherenkov imaging findings across clinical impact categories. (A) Category 1: Verification of expected Cherenkov signal distribution during DIBH breast irradiation without contralateral exposure. (B) Category 2: Minor unexpected Cherenkov signal during right breast treatment associated with patient repositioning, resolved with setup refinement. (C) Category 3: Visualization of unintended contralateral breast and chin irradiation prompting workflow modification. (D) Category 4: Absence of expected surface emission due to insufficient flash margin, requiring treatment interruption and replanning [10].

For breast cases, including the Category 2 and 3 events described above, radiation therapists performed a light field check as part of routine setup verification. However, the light field is most informative for select anterior and tangential geometric checks and does not reflect all delivered apertures or segments (e.g., field-in-field segments, MLC-defined subfields, or modulated delivery). CI provided a cumulative beam-on visualization of surface emission reflecting the entire delivered plan, thereby revealing issues such as intrafraction patient motion, arm or accessory drift after initial setup, and subfield-specific geometric deviations that may not be apparent during a static light field check. CI was therefore considered complementary to, rather than a replacement for, standard setup verification.

The representative cases presented above primarily involve breast treatments, reflecting both the predominance of breast plans among Category 3–4 events (55%, Table 3) and the broad range of CI findings observed in this patient population. Surface proximity, large field sizes, deep inspiration breath-hold setup complexity, and sensitivity to intrafraction motion make breast radiotherapy particularly well suited to demonstrating the range of CI-detected deviations. Category 3–4 events did occur across other disease sites as shown in Table 3, including pelvis, lung and abdomen, and brain.

4. Discussion

This multi-site implementation demonstrates that Cherenkov imaging can be sustainably embedded within routine clinical practice and demostrate measurable clinical utility. Fifteen percent of plans demonstrated actionable deviations, with 3% classified as major. Approximately half of major deviations were identified during beam delivery and corrected immediately, underscoring the patient safety implications of real-time surface visualization.

Detection was strongly concentrated within early treatment fractions (median detection fraction 2.6), supporting structured early-course review as a practical implementation model. These findings extend prior reports. Jarvis et al. [11] described six incidents in an initial 64-patient experience, and Alexander et al. [13] reported quality improvement opportunities during one year of clinic-wide deployment at a single institution. The present study adds a predefined clinical impact classification system, formal temporal analysis, and a multi-site cohort of 1,196 plans, providing a more structured characterization of actionable CI yield under sustained routine use.

Compared with IGRT and SGRT, which verify patient geometry and positioning [14], [15], CI provides complementary beam-on confirmation of delivered superficial dose distribution patterns. Rather than replacing volumetric imaging, CI functions as an additional safety layer that visualizes radiation–tissue interaction during treatment delivery. From a systems perspective, it bridges the gap between planning intent and observable surface dose manifestation in real time.

Cherenkov imaging provides a verification perspective distinct from other beam-on and post-fraction methods. In-vivo point dose measurements provide quantitative confirmation of delivered dose but sample only limited locations and may miss deviations elsewhere. Portal dosimetry and EPID-based methods offer high-resolution assessment of delivered fluence or exit dose but do not directly visualize beam interaction with patient surface anatomy. CI complements these approaches by providing real-time, full-field visualization of surface emission patterns, making it particularly useful for detecting positioning, immobilization, and intrafraction changes, although it remains qualitative and is most informative when clinically relevant effects are expressed at the patient surface.

Limitations include reliance on qualitative assessment rather than absolute dose quantification, potential observer variability, and limited applicability for deep-seated targets without surface manifestation. The current implementation was performed exclusively on Varian TrueBeam linear accelerators using DoseRT (Vision RT), and generalizability to other platforms has not been assessed. Category adjudication was performed by the clinical care team without formal inter-observer reliability testing; future work should quantify inter-observer agreement and explore automated detection approaches using image similarity or pattern-recognition metrics.

Implementation-level findings included institutional workflow refinements (2.5% of plans) and replanning (1.8%). No recurrent deviations of the same type were observed within individual treatment courses following corrective intervention, suggesting that CI-detected findings prompted effective and durable corrections. No measurable disruption to clinical throughput was reported; formal time-motion analysis is planned for future study. Notably, approximately half of Category 4 deviations were identified not during beam delivery but through the structured multidisciplinary physics review process, underscoring that real-time therapist observation alone is insufficient for comprehensive detection and that a systematic review process adds substantial clinical value. Furthermore, 60–70% of plans ultimately classified as Category 3–4 were initially categorized as Category 1 (informational) by radiation therapists during real-time delivery, with subsequent physicist-led multidisciplinary review prompting reclassification to a higher impact level. This finding reinforces the importance of the early-fraction review model and suggests that the clinical yield of CI is substantially enhanced when real-time monitoring is paired with structured retrospective review by the physics team.

Real-time CI was successfully integrated across two radiation oncology clinics and sustained over a two-year period. Using predefined clinical impact categories, 15% of monitored treatment plans demonstrated actionable deviations, including 3% major deviations. Most deviations were identified within the first three fractions.

The temporal concentration of findings during early treatment directly motivated adoption of mandatory multidisciplinary review of the first five fractions for all new treatment starts. This early-course review model may serve as a practical policy framework for institutions implementing CI.

CI supported replanning in 1.8% of plans and workflow modification in 2.5%, reinforcing its role as a complementary real-time verification modality that enhances early-course assessment and patient safety without measurable disruption to clinical operations.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Adi Robinson: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Michael Tallhamer: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: The authors have a relationship with VisionRT that includes consulting or advisory and travel reimbursement. Our institution is a Center of Excellence for the vendor, and it receives equipment at reduced cost for evaluation purposes as part of a Professional Services Agreement. This relationship did not influence the work reported in this manuscript.