Stereo Cherenkov imaging as 3D absolute dose monitoring tool for total skin electron therapy (TSET)

Abstract

Cherenkov imaging provides a non-contact method for monitoring surface dose delivery during total skin electron therapy (TSET), where patients are treated in six postures using the Stanford standing technique. In this study, a multi-view Cherenkov imaging framework combined with patient-specific correction factors was investigated to improve quantitative three-dimensional (3D) surface dose estimation. Cherenkov emission was recorded during treatment using three synchronized time-gated cameras providing anterior and lateral coverage. Patient-specific anatomical models reconstructed from 3D surface scans were used to generate Monte Carlo (MC) reference dose distributions. Cherenkov images were corrected for background, camera response, and geometric distortions before conversion to pseudo-dose using in-vivo dosimetry calibration. Vertical Cherenkov intensity profiles were compared with MC dose profiles to derive patient-specific correction factors. Results showed that uncorrected Cherenkov-derived dose underestimated dose in distal and lateral body regions, while applying patient-specific corrections and multi-view imaging improved agreement with MC reference dose, supporting enhanced surface dose monitoring for TSET.

1. INTRODUCTION

Total skin electron therapy (TSET) is a specialized radiotherapy technique used to treat cutaneous lymphoma and leukemia [1–6] by delivering a uniform radiation dose across the entire skin surface while minimizing dose to underlying organs. The Stanford standing technique is widely implemented for clinical TSET delivery, where patients are treated in six sequential postures separated by approximately 60° rotations [7]. Dual-field electron beams are applied at each posture to improve vertical dose uniformity and overall surface coverage. Despite the clinical effectiveness of this technique, accurate verification of the delivered surface dose remains challenging because of complex body curvature, posture-dependent beam incidence, and daily positioning uncertainties.

Surface dose verification in TSET is traditionally performed using point-based dosimeters, including thermoluminescent dosimeters (TLDs), optically stimulated luminescence dosimeters (OSLDs), and diode detectors. While these methods provide reliable absolute dose measurements, they sample only limited anatomical locations and cannot fully characterize spatial dose distribution across the patient surface. Optical imaging based on Cherenkov emission has emerged as a promising non-contact approach for real-time visualization of superficial radiation dose delivery. [8–10].

Prior studies have demonstrated the feasibility of converting Cherenkov images into pseudo-dose distributions and projecting them onto patient-specific anatomical models to estimate cumulative surface dose during TSET. However, substantial discrepancies (up to 200–300%) between uncorrected Cherenkov-derived dose and actual surface dose have been observed in the lower extremities and remain unresolved.[11, 12]. In this study, we aim to reduce these discrepancies by introducing one-dimensional patient-specific correction factors derived from vertical Cherenkov intensity profiles and Monte Carlo dose simulations [13, 14]. Additionally, the multi-posture delivery of TSET introduces self-occlusion and viewing-angle dependence, which can lead to incomplete surface sampling and reduced quantitative accuracy when relying on a single imaging viewpoint. To address these limitations, we investigate a multi-view Cherenkov imaging framework combined with patient-specific vertical corrections to improve the accuracy and robustness of three-dimensional surface dose monitoring during TSET.

2. MATERIALS AND METHODS

2.1. Setup of the treatment, camera and measurements

Total skin electron therapy treatments were delivered using a 6 MeV electron beam from a clinical linear accelerator (TrueBeam, Varian Medical Systems). Patients were positioned in the Stanford standing postures inside a treatment box located 400 cm from the linac isocenter, corresponding to a source-to-surface distance (SSD) of approximately 500 cm. Each treatment cycle consisted of six postures (anterior–posterior and four oblique orientations separated by approximately 60° rotation). For each posture, dual electron fields with optimized gantry angles were delivered to improve dose uniformity along the vertical direction of the patient body.

A multi-camera Cherenkov imaging system was used to capture optical emission during irradiation. Three time-synchronized, time-gated intensified CMOS cameras (DoseOptics, NH, USA) were installed in the treatment room to provide full-body surface coverage. One camera was positioned behind the linear accelerator to image the patient front surface, while two additional cameras were mounted on adjacent walls to capture lateral views. The cameras were aligned toward the treatment box and operated in synchronization with radiation pulses. During each radiation field delivery, room lights were turned off to suppress ambient light contamination, and individual frames were accumulated to produce integrated Cherenkov images corresponding to each treatment posture. Prior to treatment initiation, three 3D scanners were employed to acquire patient 3D surface geometry for each posture, which was later used for patient-specific model reconstruction and subsequent analysis. Additional details of the imaging setup are described in our prior publication [12].

2.2. Monte Carlo Reference Dose

Patient-specific anatomical models were generated to serve as the geometric basis for 3D dose reconstruction and simulation. Surface geometry for each treatment posture was obtained using 3D scanning systems. The scanned surfaces were processed using computer animation and mesh registration techniques to ensure consistent vertex correspondence across all postures, enabling cumulative dose mapping throughout the entire treatment course[15].

Monte Carlo (MC) simulations were performed using patient DICOM datasets generated from the patient-specific 3D models for each posture to obtain reference surface dose distributions. The clinical TSET beam configuration, including beam energy, gantry angles, SSD, and treatment geometry, was reproduced in TOPAS. The commissioning of the simulation and detailed simulation setup have been described in previous publications [13, 14]. The simulated dose matrices were projected onto the corresponding patient surface models for each posture and summed to obtain cumulative surface dose distributions. The average MC dose profile along the patient height was calculated for eight patients and used as the reference for deriving one-dimensional correction factors for the Cherenkov intensity.

2.3. Patient-specific corrections

Raw Cherenkov images were preprocessed using background subtraction and median filtering. Flood-field and perspective corrections were applied to compensate for camera response variations and geometric differences between Cherenkov intensity and delivered dose [10]. Cherenkov intensity from the front camera was converted to pseudo-dose using patient- and day-specific conversion factors derived from same-day in-vivo dosimetry (IVD) measurements at the umbilicus, which serves as the clinical calibration reference point. Conversion factors relating the side cameras to the front camera were experimentally determined, allowing lateral Cherenkov views to be converted to pseudo-dose once the front-camera calibration factor was established.

Following dose conversion, Segment Anything Model v2 (SAMv2), a deep-learning-based image segmentation framework [16], was applied to extract the patient silhouette and compute the average Cherenkov-derived dose profile along the vertical direction. The resulting profile was compared with the averaged Monte Carlo (MC) dose profile to derive one-dimensional correction factors for each patient.

To evaluate the effectiveness of the patient-specific correction factors, the derived corrections were applied to each patient’s Cherenkov-derived dose maps and projected onto patient-specific 3D surface models to obtain cumulative dose distributions across all six treatment postures. The corrected and uncorrected Cherenkov-derived doses were compared with MC-simulated dose distributions to quantify improvements in reconstruction accuracy. In addition, dose estimates derived from the front camera alone were compared with multi-view reconstructions to assess the benefit of incorporating lateral imaging information. A total of eight patient datasets were analyzed in this study. Due to incomplete IVD measurements under certain treatment conditions, MC-simulated dose distributions were used as the primary reference for evaluation.

3. RESULTS

Figure 1 shows the Cherenkov-to-MC dose ratio for six postures for patient#95 and the derived 1D patient-specific correction factor. Figure 2 shows the Cherenkov-converted dose maps for patient#95 before and after applying patient-specific corrections. Figure 3 summarizes the patient-specific correction factors (CFs) for 20 patients. Table 1 shows the percentage difference between Cherenkov-dose (uncorrected, using the front camera only) and the reference Monte Carlo dose. Table 2 shows the percentage difference after patient-specific corrections using all three cameras

Figure 1.

Cherenkov-to-MC dose ratio profiles for six treatment postures (solid lines) and the derived 1D patient-specific correction factor (red dashed line) for patient #95. All curves are normalized to unity at pixel index 1000, corresponding to the projected radiation central axis. Pixel index 1800 approximately corresponds to the lower extremities, while index 800 approximately corresponds to the head region. The patient-specific correction factor was obtained by averaging the vertical Cherenkov-to-MC dose ratios across all postures and applying a cubic polynomial fit for smoothing. A symmetry constraint was imposed on the correction factor to reduce the impact of noise from facial features and hair, as well as absent head Cherenkov signal in patients treated with head blocking.

Figure 2.

Cherenkov intensity distributions for six TSET postures for patient #95 shown (a) before and (b) after applying patient-specific correction factors derived from Fig. 1. Each image represents the accumulated Cherenkov signal from the corresponding dual-field irradiation for each posture. After applying the correction factors, improved intensity uniformity is observed across the patient surface, particularly in the lower extremities and lateral body regions, where uncorrected images show systematic intensity underestimation.

Figure 3.

Summary of patient-specific correction factors for 20 patients as a function of vertical image index. Each curve represents a smoothed correction factor derived from Cherenkov-to-MC dose ratios. The torso region (approximately index 1200–800) was constrained to 1 based on prior studies indicating minimal correction requirement in this region. Larger correction factors are observed in distal anatomical regions, particularly in the shins. Lower shin correction factors were observed in patients with darker skin (1.81 ± 0.19 at index 1700) compared with lighter skin (2.37 ± 0.36).

Table 1.

Percentage difference between Cherenkov-dose (front view only, uncorrected) and Monte Carlo (MC) reference dose at five anatomical sites for eight patients. Mean and standard deviation (Std, %) are reported.

	Perineum	Shin	Chest	Lt. trunk	Rt. Trunk
Pat9(2)	−35%	−68%	3%	−24%	−17%
Pat37	−20%	−54%	4%	−18%	−16%
Pat39	−19%	−49%	6%	−20%	−27%
Pat40	−7%	−57%	0%	−22%	−25%
Pat41	−25%	−65%	−8%	−17%	−4%
Pat42	−13%	−73%	−3%	3%	−10%
Pat47	3%	−51%	−6%	−20%	−16%
Pat55	−23%	−70%	−8%	−10%	−13%
Mean	−17%	−61%	−2%	−16%	−16%
Std	11%	9%	6%	9%	7%

Table 2.

Percentage difference between Cherenkov-dose (three views combined, corrected) and Monte Carlo (MC) reference dose at five anatomical sites for eight patients. Mean and standard deviation (Std, %) are reported.

	Perineum	Shin	Chest	Lt. trunk	Rt. Trunk
Pat9(2)	−32%	−5%	3%	−4%	−7%
Pat37	−20%	−6%	3%	−8%	−8%
Pat39	−22%	3%	7%	−12%	−14%
Pat40	−4%	3%	0%	−12%	−12%
Pat41	−20%	3%	−10%	3%	2%
Pat42	−9%	8%	−3%	13%	−6%
Pat47	8%	−2%	−5%	−9%	−7%
Pat55	−15%	−8%	−8%	14%	−1%
Mean	−14%	0%	−2%	−2%	−7%
Std	12%	6%	6%	11%	5%

4. DISCUSSIONS

These findings demonstrate that combining patient-specific correction factors with multi-view Cherenkov imaging significantly improves quantitative surface dose estimation during TSET. Figure 1 illustrates the posture-dependent variation of the Cherenkov-to-Monte Carlo (MC) dose ratio for a representative patient. The ratio profiles show consistent trends across multiple postures but exhibit spatial deviations along the patient height. These deviations indicate systematic differences between Cherenkov-derived dose and MC reference dose, particularly in distal body regions. The derived one-dimensional correction factors effectively compensate for these discrepancies by normalizing Cherenkov intensity relative to the MC dose profile. The similarity of correction factor trends across postures suggests that vertical dose variation could be influenced by treatment geometry and optical sampling depth. The impact of applying patient-specific corrections is demonstrated in Figure 2, which compares Cherenkov-converted dose maps before and after correction. Prior to correction, Cherenkov intensity distributions show noticeable non-uniformities across several anatomical regions, particularly in the lower extremities and lateral body surfaces. After applying correction factors, improved spatial uniformity is observed across all postures. These results confirm that vertical intensity correction can effectively mitigate systematic dose underestimation while preserving posture-specific anatomical features.

Population-level variability of correction factors is summarized in Figure 3. The results indicate that correction magnitude varies across anatomical regions and between patients. The shins consistently require larger correction factors compared to other body regions, indicating reduced Cherenkov signal collection efficiency in distal extremities. This behavior is likely influenced by increased beam obliquity, variations in optical attenuation, and reduced visible surface area from the camera viewpoint. Additionally, differences in correction factors observed among patients suggest that tissue optical properties, including skin pigmentation, may contribute to variations in Cherenkov signal intensity.

Quantitative validation results presented in Tables 1 and 2 demonstrate the benefit of combining patient-specific correction factors with multi-view imaging. When Cherenkov dose is derived from the front camera alone, large discrepancies relative to MC dose are observed across several anatomical sites, particularly at lateral trunk regions. Incorporation of side-view Cherenkov imaging substantially improves dose agreement by recovering emission signals from surfaces not directly visible in the front view. After applying both multi-view integration and patient-specific correction factors, the overall difference between Cherenkov-derived dose and MC reference dose is significantly reduced across most anatomical locations.

Although these results demonstrate substantial improvement in dose reconstruction accuracy, residual discrepancies remain in anatomically complex regions such as the perineum and vertex. These regions are subject to strong geometric curvature, partial self-occlusion, and variable beam incidence angles, which may alter both Cherenkov generation and light transport. Further refinement of correction models incorporating patient-specific anatomical geometry and optical property variations may improve accuracy in these regions.

Overall, this study demonstrates that combining stereo Cherenkov imaging with patient-specific correction factors improves spatial dose estimation accuracy during TSET. Continued development of multi-view calibration strategies and expansion of patient cohorts will further improve the reliability of Cherenkov-based dose monitoring for clinical applications.

5. CONCLUSIONS

Applying a vertical correction substantially improves agreement between Cherenkov-derived dose and Monte Carlo dose in the shins. Adding side cameras further corrects the dose underestimation on the patient’s sides. Future work will refine the correction method, improve accuracy at other locations (e.g., perineum, vertex, and hands). We will also expand the patient cohort, and investigate the causes of the large correction factors in TSET Cherenkov imaging.