Novel framework for determining TPS-calculated doses corresponding to detector locations using 3D camera in in vivo surface dosimetry

Purpose. To address the shortcomings of current procedures for evaluating the measured-to-planned dose agreement in in vivo dosimetry (IVD), this study aimed to develop an accurate and efficient novel framework to identify the detector location placed on a patient’s skin surface using a 3D camera and determine the planned dose at the same anatomical position corresponding to the detector location. Methods. Breast cancer treatment was simulated using an anthropomorphic adult female phantom (ATOM 702D; CIRS, Norfolk, VA, USA). An optically stimulated luminescent dosimeter was used for surface dose measurements (MyOSLchip, RadPro International GmbH, Germany) at six IVD points. Three-dimensional surface imaging (3DSI) of the phantom with the detector was performed in the treatment position using a 3D camera. The developed framework, iSMART, was designed to import 3DSI and treatment planning data for determining the position of the IVD detectors in the 3D treatment planning DICOM image. The clinical usefulness of iSMART was evaluated in terms of accuracy and efficiency, for comparison with the results obtained using cone-beam computed tomography (CBCT) image guidance. Results. The relative dose difference between the planned doses determined using iSMART and CBCT images displayed similar accuracies (within approximately ±2.0%) at all detector locations. The relative dose differences between the planned and measured doses at the six detector locations ranged from –4.8% to 3.1% for the CBCT images and –3.5% to 2.1% for iSMART. The total time required to read the planned doses at six detector locations averaged at 8.1 and 0.8 min for the CBCT images and iSMART, respectively. Conclusions. The proposed framework can improve the robustness of IVD analyses and aid in accurate and efficient evaluations of the measured-to-planned dose agreement.


Introduction
Radiation therapy (RT) dose plans calculated via a treatment planning system (TPS) should be safely and accurately delivered to patients. in vivo dosimetry (IVD) is performed by placing one or more detectors at representative locations or several relevant points on the skin surface to verify the consistency between the planned and delivered doses (Dipasquale et al 2014). For IVD measurements, various detectors such as optically stimulated luminescent dosimeters (OSLD), thermoluminescent dosimeters, and Gafchromic films are currently used in clinical practice (International Atomic Energy 2013, Esposito et al 2020, Verhaegen et al 2020).
To accurately evaluate the measured-planned dose agreement in IVD, the TPS-planned dose should be read and recorded at the same anatomical position as the detector location placed on the patient's surface. To this Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. end, the anatomical or spatial location of the detector should first be identified, and the detector position in the 3D treatment planning datasets should be determined. Significant differences in the measured-to-planned dose agreement may occur if the planned dose is read at regions different from the actual measurement location of the IVD detector, particularly in regions with high dose gradients and heterogeneities. In addition, detector position disagreements between the patient and the TPS can lead to inaccurate assessment of measured-to-planned dose agreement and erroneous clinical decisions.
In current clinical practice, the location coordinates of the detector placed on the patient's skin surface are generally identified using reference tattoos, light-field crosshairs, or the coordinates of the treatment couch (Townend et al 2020, Hanley et al 2021, Xu et al 2022. The TPS-planned dose for each detector is identified and read out from the treatment planning dataset by referencing the detector's manually defined location and relative coordinates. However, these procedures, which are mainly used for patients undergoing electron-beam RT or 2D/3D-RT, require additional time, resources, and well-trained staff. Such demanding and complicated processes can result in increased variability in the measured-to-planned dose agreement. Alternatively, the detector location observed in clinical cone-beam computed tomography (CBCT) images can be used to identify the corresponding detector location in the treatment planning images, for a comparison of the measured and planned doses (van Elmpt et al 2009). Although this method reduces the detector position disagreement, the patient's CBCT must be obtained for all detector positions, and it entails unnecessary radiation exposure to the patient. Furthermore, the detector position is difficult to identify if the detector is thin or artifacts occur in CBCT.
To overcome these limitations of the current procedures for evaluating the measured-to-planned dose agreement in IVD, new approaches aiming to reduce variability, offer facile usability, leverage the potential for automation, and increase clinical acceptance should be developed. Hence, this study aimed to develop a novel framework that can identify the IVD detector location placed on a patient's skin surface using a 3D camera and determine the planned dose at the same anatomical position as the detector location. Furthermore, we evaluated the clinical effectiveness of the developed framework in terms of its accuracy and efficiency and compared its results with those obtained from CBCT image guidance.

CT Simulation and treatment planning
In this study, breast cancer treatment was simulated using an anthropomorphic adult female phantom (ATOM 702-D; CIRS, Norfolk, VA, USA). Computed tomography (CT; Aquilion, Toshiba, Japan) images were acquired using a 3 mm-thick slice for RT treatment planning. Subsequently, the acquired sets of CT images were transferred to a RayStation (v10.0, RaySearch Laboratories, Stockholm, Sweden) treatment planning system.
The planning target volume (PTV) was defined as the contoured breast volume subtracted by 3 mm from the surface edge. The PTV of the hypothetically defined tumor bed was contoured separately, and the organs at risk, including the lungs and heart, were delineated. The external contour of the phantom surface was defined by the automatic segmentation method using a threshold of -350 Hounsfield units in CT. Volumetric modulated arc therapy (VMAT) planning using two coplanar partial arcs was performed using a linear accelerator (LINAC) with a 6 MV photon beam (Versa HD, ELEKTA, Crawley, UK). The first arc was initiated at 240°and ceased at 64°, and the second arc acted in reverse to the first. The collimator angles for the first and second arcs were 0°and 90°, respectively. Specifically, the prescription dose was 40.05 Gy in 15 fractions to the entire breast and 48 Gy to the tumor bed with a simultaneous integrated boost (2.67 and 3.2 Gy/fraction, respectively) (Chang et al 2020, Lee et al 2020). The objective of the plan was to cover at least 95% of the PTV with 95% of the prescribed dose. The dose calculation was performed using a collapsed cone algorithm with a dose grid size of 2 mm.

in vivo surface dose measurement
The dose prescribed according to the PTV was delivered to the anthropomorphic phantom using a 6 MV photon beam via the ELEKTA Versa HD LINAC. Prior to the treatment, optically stimulated luminescent dosimeters (OSLDs) were placed on the phantom surface for dose measurements. In this study, six IVD points were measured, and the IVD points were segmented into six regions of the upper, medial, lower, lateral, contralateral breast, and midcoronal line with respect to the center of the anthropomorphic phantom's breast, and their locations were defined as 1, 2, 3, 4, 5, and 6, respectively (figure 1). Four (IVD points 1 to 4) of the six OSLDs were placed inside the PTV and two (IVD points 5 and 6) placed in the low-dose area located outside the PTV. The gantry angles of the VMAT plan ranged from 240°to 64°, while the treatment beams were incident between 0°a nd 120°for individual OSLDs placed on the breast anthropomorphic phantom. All the OSLD doses measured in this study did not include angular dependence for an appropriate orientation. Based on a previous report (Kry et al 2020), a small angular dependence of the OSLD at megavoltage energies can be expected. The measuring locations of the detector for the IVD were marked on the anthropomorphic phantom's surface to ensure reproducible positioning throughout the treatment. All measurements were performed repeatedly for three consecutive days.
The surface dose measurements were performed using MyOSLchip (RadPro International GmbH, Germany) OSLDs, composed of a 4.7 mm × 4.7 mm × 0.5 mm Beryllium Oxide (BeO) square chip and enclosed in a 9.5 mm × 10 mm × 2 mm plastic housing fabricated using an acrylonitrile-butadiene-styrene (ABS) material (Sommer, Henniger 2006, Richter et al 2019, Gasparian et al 2022, RadPro 2022. The effective atomic number of BeO material is 7.2, near tissue equivalent, with a nominal density of 2.85 g cm −3 . The BeO OSL dosimeter has a dose range of 0.05 mGy-10 Gy, with dose linearity up to 10 Gy, and an energy response ranging from 16 keV to several MeV. At room temperature, the lifetime is 27 μs, with fading less than 1% in 6 months. The stabilization time required for detector readability is 30 min The signal depletion is decreased by 0.7% per reading (Madden et al 2021). Furthermore, the effective point of measurement for the OSLDs was situated at the center of the detector, at a depth of 1 mm from the detector surface, corresponding to a waterequivalent depth of 1.25 mm.
Device and system calibration was performed prior to the initial use of the OSLD according to the guideline proposed by the vendor. These initial calibrations included calibration of the OSLD based on LINAC, and individual calibration of the individual dosimeter's sensitivity at that energy. To minimize various uncertainties in the OSLD measurement values, the OSLD used in this study was additionally calibrated with a correction factor. Each OSLD was calibrated using a 6 MV beam and a dedicated solid water slab phantom (PTW, Germany, dimensions 30 cm × 30 cm × 20 cm), with a calibration dose chosen to be close to the expected skin dose used in this study. All OSLDs were irradiated to 250 cGy with a field size of 10 cm × 10 cm, and at depths of 10 cm and 100 cm source-to-surface distance. A correction factor was calculated as the ratio of the reference dose (250 cGy) delivered to the OSLD to the OSLD measured dose read by the myOSLchip reader. Using a commercial spreadsheet program, final OSLD measurements were calculated by applying corresponding correction factors to the associated OSLD readings.
Prior to each treatment, the CBCT images were obtained at the treatment position to confirm the variations in the phantom position. Subsequently, the anthropomorphic phantom was irradiated using a VMAT treatment plan. Raw measurements of OSLDs included the combined doses from imaging and treatment delivery, and the CBCT imaging dose is estimated to be approximately 1-2 cGy (Islam et al 2006). All OSLD measurements were read with the myOSLchip reader, with singular reading taken for each OSLD according to the vendor's procedure.
2.3. New IVD workflows using developed framework 2.3.1. 3D surface imaging acquisition A three-dimensional (3D) camera (Intel RealSense Depth Camera D435i, Intel, Santa Clara, CA) that could simultaneously scan color and depth images was employed to obtain the 3D spatial location information of the IVD detectors (OSLDs) located on the anthropomorphic phantom surface. Moreover, three-dimensional surface imaging (3DSI) of the phantom was performed at the treatment position. The depth information of an object was stored along with the RGB color images in the form of a point cloud (PC), which denotes a dataset of points detailing the height, width, and depth. In particular, the PC and RGB images were extracted from the scanned data files using Intel RealSense software development kit 2.0.

Design and development of proposed framework for IVD analyses
In this study, we developed a novel framework, named in vivo surface dosimetry measurement assessment in radiation therapy (iSMART), to identify and record the planned dose corresponding to the same anatomical position as the location of the IVD detectors placed on a patient's skin surface during in vivo skin dosimetry. This framework was implemented using MATLAB, version 2020b. The flowchart of iSMART is presented in figure 2.
iSMART was designed to import 3DSI and treatment planning data, including the RT structure, RT dose, and RT plan, in order to determine the position of the IVD detectors in the 3D treatment planning DICOM image. A 3D surface dose (3DSD) was extracted 2 mm below the external contour using RT-Dose and RT-Structure obtained from the TPS, which was then converted into a PC array containing the position and dose intensity. Additionally, registration was performed between 3DSI and 3DSD to project the surface dose distribution on 3DSI. The surface registration was optimized using iterative closest point algorithms with a 90% inlier rate and a maximum of 1000 iterations, along with a tolerance of 0.5 mm (Park et al 2022b). Moreover, a uniform sampling scheme was adopted to improve the efficiency of the PC registration. The 3DSD distribution, visualized on 3DSI, was reconstructed as two-dimensional projection images using a ray-tracing algorithm based on the camera intrinsic matrix (Park et al 2022a).
To record the planned dose corresponding to the detector location, the user reviews and selects the positions of the detectors visualized on the phantom surface image according to the rectangular location-bounding box created by iSMART, which is considered as the detector position. Thereafter, the user enters the detector depth to determine the subsurface depth and extract the surface dose. In this study, we adopted a rectangular locationbounding box with dimensions of 5 mm (W) × 5 mm (L) × 2 mm (H). The center position of each locationbounding box was reconstructed into a 3D coordinate system, and the planned dose of the TPS corresponding to the position of the location-bounding box was obtained using the 3DSI-to-3DSD registration information. Thus, the planned dose was evaluated as the average dose of the voxels included in the bounding box. Ultimately, iSMART outputs the planned dose values corresponding to all the visualized detector locations.

Performance evaluation of developed framework
The accuracy of the developed framework, iSMART, was evaluated by comparing the planned dose (PD iSMART ) analyzed by iSMART with the planned dose (PD CBCT ) evaluated from CBCT images (figure 3). The planned dose was determined at six OSLD locations, using three CBCTs and three 3DSIs acquired over three consecutive days, with the three readings at each location being averaged and compared. The method utilizing CBCT images entails the identification of detector locations using the obtained images to confirm the setup positioning prior to treatment. Here, CBCT images were imported into the TPS and fused with the planning CT images. Thereafter, two experienced medical physicists manually drew the regions of interest (ROIs) on the planning CT images in the anatomical positions corresponding to the detectors identified in the CBCT images (figures 3(a), (b)). In this study, the ROI was demarcated as a rectangular box with dimensions of 5 mm × 5 mm × 2 mm under an external contour considering the dose grid size used for dose calculation (Hafez et al 2021).
The planned dose corresponding to the detector location was calculated as the average dose in each ROI, and the relative dose difference between PD iSMART and PD CBCT was evaluated based on the following equation: The inter-operator variability of the planned dose readings between the two methods was investigated by comparing PD CBCT and PD iSMART which were determined independently by two medical physicists at the detector position. The planned dose determined using CBCT images and iSMART by one of the medical physicists was compared with those of other physicist, while the relative dose differences were also analyzed.
Additionally, the planned dose (PD iSMART and PD CBCT ) analyzed via both methods was compared with the measured dose (MD OSLD ) obtained from the OSLD. The relative dose difference between the PDs and MD OSLD was calculated as follows: The total time required for all the procedures to obtain the planned dose corresponding to the detector location was measured, to evaluate the time efficiency of iSMART. The results were compared with those obtained under CBCT guidance. The relative time difference between the total time analyzed using iSMART (T iSMART ) and CBCT images (T CBCT ) was calculated as follows:

Results
The TPS-planned doses at the six detector locations ranged from 37 to 252 cGy. At all the detector locations, the relative dose difference between the planned doses determined using CBCT images and iSMART exhibited similar accuracies within approximately ±2.0% (table 1). In comparison, the relative dose differences noted by the two medical physicists in the PD CBCT and PD iSMART results were within ±3.2% and ±1.1%, respectively. The largest difference between the two medical physicists was 3.2%, which was noted at point #1, as determined from the CBCT images ( figure 4).
Additionally, the planned doses determined using the CBCT images and iSMART were compared with the OSLD-measured doses. The relative difference between the planned and measured doses was approximately ±5.0%. At the six detector locations, the relative dose differences between the planned and measured doses ranged from -4.8% to 3.1% for the CBCT images and -3.5% to 2.1% for iSMART (table 2).
Furthermore, we assessed the total time required to measure the planned doses at the six detector locations. The average time was 8.1 ± 0.9 and 0.8 ± 0.1 min for the CBCT images and iSMART, respectively. Overall, using iSMART results in 7 min lesser (corresponding to 89% time reduction) IVD analysis time.

Discussion
In this study, we developed a novel framework that can identify the location of an IVD detector placed on a patient's skin surface using a 3D camera and read the planned dose at the same anatomical position as the location of the detector. To the best of our knowledge, this is the first study that determines the location information of a detector using a 3D camera to reduce the variability in the measured-to-planned dose agreement in IVD. The developed framework, iSMART, delivered comparable performance to CBCT image guidance in terms of accuracy and improved performance as compared with CBCT image guidance in terms of reproducibility and efficiency.   To reliably detect the error in the dose delivered to the patient, the agreement between the measured and planned doses should be compared. Therefore, RT staff should read and record the planned dose at the same anatomical point as the location of the detector on the patient's skin surface. Significant dose differences in the measured-to-planned agreement can occur if the planned dose is measured at a point different from the actual measurement location of the IVD detector. In particular, in regions with high dose gradients and dose heterogeneities, such as IMRT and VMAT, a disagreement in the detector position can create dramatic differences in the measured-to-planned dose agreement, potentially resulting in inaccuracies in clinical decisions and the evaluation of IVD results. Tariq et al investigated the magnitude of detector position disagreement and its impact on dosimetric errors (Tariq et al 2019). They observed an average detector position error of 9.7 ± 9.5 mm at 274 IVD points and reported a dosimetric disagreement of approximately 10% per cm of detector position disagreement. The placement error investigated in the previous study can be understood in the same context as the error in reading the planned dose at a position different from the actual detector's location, which is the background of our study. These findings emphasize the importance of accurately identifying the detector location and measuring the planned dose at a point corresponding to the detector location in the clinical workflow of IVD.
In current IVD measurements, well-trained RT staff manually determine the location of each detector attached to the patient's surface based on reference tattoos, light-field crosshairs, and the coordinates of the treatment couch. In addition, clinical CBCT scans have recently been used to identify the location of the detector visualized on the CBCT image to determine the planned dose according to the detector position. However, this current workflow of IVD is highly demanding in terms of complexity, resources, and the requirements of welltrained and skilled manpower. To resolve the challenges of these current IVD methods, a new approach that saves time and cost, while providing robust measured-to-planned dose agreement is required. Therefore, we developed a novel framework utilizing 3D cameras. Compared with current methods, we believe that the proposed framework, iSMART, offers several advantages in improving the reliability of IVD results.
First, the proposed approach ensures accurate and reproducible results, regardless of the worker's experience and skill level. In this study, the main objective of performance evaluation was to compare whether iSMART can read the planned dose from the detector location, similar to the accuracy of CBCT image guidance. CBCT is an excellent tool for accurately identifying the measured detector position, and it is difficult to find a better method in current clinical practice. The present findings demonstrated that the planned dose at the detector location determined by iSMART exhibited a highly consistent accuracy of approximately 2% when compared with the method using CBCT images. In particular, compared to PD CBCT , PD iSMART showed a relatively lower standard deviation of the three readings measured once a day for three consecutive days at points #1 and #5. Points #1 and #5 were located in the steep-slope area (50°) of the breast phantom and the penumbra region with a high dose gradient, respectively. This suggests that the planned dose determination using the manual method of CBCT image may result in reduced reproducibility in areas with geometric and dosimetric uncertainties. These results were also confirmed by the relative dose difference between the planned doses determined by two medical physicists using CBCT images and iSMART. The inter-operator variability of the two methods was verified by having independent planned dose readings taken by two medical physicists at the detector location. The relative dose difference between the readings taken by the two medical physicists was within ±3.2% and ±1.1% for PD CBCT and PD iSMART , respectively. Point #1, a steep-slope area, showed a greater difference in dose between the readings of the two physicists than at other points. In methods using CBCT images, inter-operator variability may arise as the medical physicist registers CBCT into planning CT images and manually identifies and delineates the detector in the CBCT image to determine the planned dose. On the other hand, iSMART accurately identifies the detector location with a 3D camera and uses a semiautomated system to determine the planned dose corresponding to the detector position, and hence being relatively less user-dependent. Consequently, semi-automatic iSMART using a 3D camera shows accurate and consistent performance regardless of the operator's experience and skill level.
In addition, planned doses determined using CBCT imaging and iSMART were consistent with the OSLDmeasured doses within ±4.8% and ±3.5%, respectively. However, the agreement between the measured and planned doses needs to be interpreted cautiously, considering the uncertainty and bias in this study. This study delivered the VMAT treatment plan with dual partial arcs (180-degree rotation) to the anthropomorphic breast phantom resulting in measured values of the detectors having uncertainty due to the angular sensitivity dependence of the OSLD. Furthermore, the effective measurement depth (water equivalent depth) of the OSLD used in this study was 1.25 mm, while the point where the planned dose was obtained at average depth of 2 mm below the skin surface due to the dose grid size. These uncertainties also exist in the method using CBCT currently applied in clinical practice. Nevertheless, iSMART has the potential to consider the angular dependence of each individual detector due to the use of a 3D coordinate system, and the planned dose can be calculated at the desired depth. The results of this study confirm that iSMART offers appropriate performance for clinical use.
Secondly, the simple and fast analysis process of iSMART provides an improved and efficient framework for detecting relative dose differences. Compared with the CBCT method, iSMART could reduce the time required to obtain the planned dose by approximately 89%. In total, iSMART could derive the planned dose at the six detector locations within 1 min Moreover, the anthropomorphic phantom surface on which the detector was attached to a 3D camera at the treatment position could be scanned for 3DSI acquisition within ∼30 s. Having being highlighted by several researchers, IVD methods are still not widely used owing to highly demanding workflow in terms of complexity, increased workload, time consumption, and cost-effectiveness (Essers, Mijnheer 1999, Mijnheer et al 2013, Falco et al 2018, Li et al 2018, Olaciregui-Ruiz et al 2020. In this respect, this study demonstrated that iSMART can improve the working efficiency in the IVD evaluation stage. Finally, the procedure of our novel framework enhances clinical workflow and increases clinical acceptance because it can be conveniently implemented in routine clinical practice and does not require additional radiation. The use of clinical CBCT scan of the patient can improve the measured-to-planned dose agreement, but the application of CBCT is limited to specific clinical situations. In practice, CBCT cannot be employed on patients owing to policy or cost concerns or in treatment techniques where CBCT is not routinely used, such as in 2D/3D-RT and electron-beam RT. For patients treated under treatment plans involving a long target or multiple isocenters, the short CBCT scan length may not be able to adequately cover the entire detector placed on the patient's surface. Further, the detector position cannot be easily identified if the detector is extremely small or thin, or if artifacts occur owing to the high-density material of the detector. In addition, CBCT imaging entails additional radiation doses for the patient. This study used an inexpensive and portable 3D camera that can easily acquire an accurate 3DSI. The relatively wide scan range of the 3D camera (Intel 2022) enables accurate identification of the detector's location, even in patients receiving treatment with long targets or treatment plans with multiple isocenters. Furthermore, as iSMART uses a 3D camera to detect the detector location, the location of any type of detector can be identified for visualization and measurement of the dose at that position. Therefore, the flexibility of iSMART enables users to create workflows optimized for clinical environments, IVD detectors, treatment techniques, and institutions.
Nonetheless, this study involved several limitations. First, iSMART is not a fully automated framework that can validate measured-to-planned-dose agreements. The current research is the first step toward developing a fully automated comprehensive framework, which will be realized through further research in the future. Second, the rigid image-registration algorithm used for the registration of 3DSD and 3DSI does not appropriately consider the variations in the dose distribution caused by anatomical variations. Thus, further research is required on deformed image registration to enable accurate correspondence as the patient's position varies. Finally, owing to the use of a 3D camera, iSMART can be applied only when a detector is located in a space where photography is possible. For patients where the detector is located under the bolus or a thermoplastic mask, the 3D camera can obtain the 3DSI once the bolus or thermoplastic mask is removed post-treatment.

Conclusions
This study developed a novel framework to identify and record a TPS-planned dose corresponding to the same anatomical position as the IVD detectors placed on the patient's surface. This method of utilizing a 3D camera provided highly accurate and reproducible results, in addition to minimizing cost and time consumption. The proposed framework can efficiently improve the robustness of IVD analysis and aid in the accurate evaluations of measured-to-planned dose agreements.