Automated robotic-assisted patient positioning method and dosimetric impact analysis for boron neutron capture therapy (2024)

Boron Neutron Capture Therapy (BNCT) represents a novel paradigm in radiotherapy characterized by the use of the natural isotope boron-10 with high cross-sectional affinity for thermal neutrons¹. Boron compounds selectively accumulate in tumor tissues; when subjected to a precisely modulated thermal neutron beam, these compounds facilitate the release of high-energy alpha particles and lithium-7 isotopes. This targeted nuclear reaction can selectively eradicate cancer cells and deliver biologically targeted radiotherapy at the cellular level². In contrast to conventional radiotherapies, BNCT is particularly advantageous for treating head and neck tumors as it minimizes collateral damage to surrounding healthy tissues. The development of accelerator-based BNCT (AB-BNCT) equipment has catalyzed international research and clinical studies, with Japan taking the lead by integrating BNCT into its health insurance system as of June 2020^3,4,5,6.

Traditional radiotherapy typically employs rotatable gantries to guide the radiation beam. In BNCT, however, the trajectory of beam is largely determined by the Beam Shaping Assembly (BSA). Given the stationary nature of the neutron beam and the unique physical properties of neutrons, precise patient positioning near the beam port is a crucial component for effective BNCT treatment. Generally, positioning is driven by a Treatment Planning System (TPS), which calculates theoretical positions within a virtual environment. Nevertheless, discrepancies between these theoretical calculations and practical patient positioning can result in suboptimal dose delivery, underscoring the need for precise and accurate positioning.

Given that BNCT is inherently a biotargeted treatment modality, patient positioning tolerances are not as strict as with other forms of radiotherapy. The study of Lee⁷ showed the mean dose of deep-seated tumor changes by 1% and 5% for lateral and outward shifts from the source plane, respectively, while the mean dose of superficial tumor without significant variation by 1cm shift or 15˚ tilt. The research of Kakino⁸ illustrated a significant reduction of tumor dose due to increased distance from the collimator, and the smaller the collimator, the more severe the tumor dose reduction. However, they considered a cylindrical phantom to investigate the effect of placement deviation on dose distribution. The cylindrical phantom differs somewhat from contour of patient’s head.

In BNCT positioning application, Palmer et al. utilized thermoplastic masks and physical restraints like seatbelts and headrests⁹, but this approach falls short in terms of both accuracy and repeatability. Wielopolski et al. proposed a two-step positioning method using surface markers¹⁰, which lacked automation. Kumada et al. introduced a laser coordinate system to improve accuracy^11,12,13, but this method still necessitates manual intervention. Kiger et al. focused on theoretical calculations for positioning¹⁴, likewise, automated patient positioning remained unattained. Image-Guided Radiation Therapy (IGRT), although effective in conventional radiation therapy, is not applicable in the BNCT context due to neutron activation issues with MRI or cone-beam CT devices^15,16. As a result of relying on a fixed radiation port, current BNCT practice broadly employs a movable treatment table for positioning patient, but automated patient positioning has not yet been achieved.

To address these challenges, this article introduces an automated patient positioning system that incorporates robotic technology commonly used in industrial applications, such as mechanical cutting and welding. The control parameters for the robotic system are algorithmically calculated based on beam data provided by NeuMANTA (Multifunctional Arithmetic for Neutron Transportation Analysis)¹⁷, a TPS developed by Neuboron Medical Group. Validation of robotic-assisted patient positioning and discussion of dosimetric impact using anthropomorphic phantom culminated in a fully automated patient positioning protocol, thereby enhancing the efficacy and accuracy of BNCT treatment plans.

Integration

The robotic patient positioning system is designed to leverage the precision and reliability commonly associated with industrial robots. The system comprises three primary components: a robotic arm, a laser alignment system, and a collimator, as shown in Fig.1. The robotic arm is a KUKA R2700 model with six degrees of freedom and a maximum load capacity of 2 tons. A fitting support structure, complete with beams and trusses, is securely attached to the ceiling using anchoring points to support the weight and movement of the robot. This fixture then serves as the principal platform for mounting the robot’s brackets, as depicted in Fig.2. To prevent neutron activation, the robot’s surface is clad with boron-reinforced fiberglass, a substance that shields against neutrons. A carbon fiber treatment couch is affixed to the robot’s flange using a connector, as illustrated in Fig.3. These connectors are custom-designed to facilitate rapid and secure attachment, offering a dependable and efficient method for patient positioning. The laser alignment system features three strategically placed laser lights mounted on the wall to the left, right, and above the collimator. The intersection of their cross-lines serves as the origin of the laser coordinate system. The collimator remains in a fixed position and is an integral part of beam shaping.

During the treatment planning phase, NeuMANTA reads patient images with fiducial markers and determines the position of positioning reference points and the beam parameters of the treatment plan based on its coordinate system. During the execution phase, the system leverages reference points to align the robot’s coordinate system with the imaging coordinate system to confirm initial position. Subsequently, the robotic arm system calculates control parameters through an algorithm based on the source position, and automatically moves the robotic arm to align the patient with the collimator beam port to achieve the final position.

To rigorously validate the accuracy of the automated patient positioning algorithm, an anthropomorphic phantom model was employed. Multiple fiducial markers were strategically affixed to the phantom surface. Of these, three served as seed points for initial positioning, while the remainder functioned as reference points. The configuration is illustrated in Fig.4a, with the corresponding CT model displayed in Fig.4b. The validation followed a protocol that mirrored the actual treatment procedure. Initially, the phantom underwent CT scanning, as depicted in Fig.5a. Prior to the scan, seed points were laser-aligned within the CT room to ensure their capture in the same CT slice. These seed points were subsequently identifiable in NeuMANTA, as shown in Fig.5b. Once the CT images were imported into NeuMANTA, the treatment plan was developed and the position of the phantom was confirmed. The plan is then transferred to the patient positioning system.

NeuMANTA works in synergy with robotic patient positioning system to ensure accuracy during irradiation. The software-controlled robotic arm positions the patient according to the parameters set in NeuMANTA. To improve user-friendliness and operational efficiency, NeuMANTA eliminates the need to manually adjust the patient’s image model during positioning. Instead, the system employs a virtual collimator to fine-tune the patient’s position and orientation relative to the source plane. This achieves precise patient positioning in accordance with the treatment plan.

The representation of source position parameters in NeuMANTA is analogous to that in traditional radiotherapy, with the physical collimator is affixed to the gantry, as illustrated in Fig.6. Within the image coordinate system of the treatment planning platform, the spatial coordinate center of the collimator is denoted as (SrcX, SrcY, SrcZ). It is worth noting that the effects of collimator rotation about the beam centerline can be neglected during treatment. This simplifies the positioning problem by reducing the degrees of freedom involved in collimator adjustments. Consequently, only two angular parameters — θ (theta) and φ (phi) — require to be considerated. Here, θ represents the angle between the beam centerline and the Y-axis, while φ is the angle between the beam centerline and the Z-axis. These parameters uniquely define the relative spatial relationship between the collimator center and the patient, thus facilitating precise treatment execution.

The positioning parameters for the collimator and patient are represented in coordinate systems C₁ and C₂, respectively. The origin of the C₁ has is located at (SrcX, SrcY, SrcZ), with the Z₁-axis aligned along the beam centerline. The relationship between Z₁ and the angular parameters defined in NeuMANTA, can be elucidated by Eq.1.

\(X={Z_1} \cdot \sin (\varphi ) \cdot \sin \left( {~\theta } \right)\)

\(Y=~{Z_1} \cdot \sin (\varphi ) \cdot \cos \left( {~\theta } \right)\)

Considering the circular geometry of the collimator outer surface, the X₁-axis is defined along the horizontal plane of the surface. Once the origin and the two principal axes are established, C₁ becomes a well-defined coordinate system. Similarly, by identifying three non-collinear reference points on the patient’s surface image, a unique coordinate system C₂ can be established to represent the patient’s position.

Robotic systems are known for their high positioning accuracy, operational flexibility, and repeatability. When augmented with a treatment couch at its flange, a six-axis robot becomes a formidable asset in patient positioning for BNCT. The flange’s spatial coordinates are uniquely determined by six parameters: T_x, T_y, T_z, R_a, R_b, and R_c. Here, T_x, T_y, and T_z signify translational displacements along the X, Y, and Z axes, respectively. R_a, R_b, and R_c correspond to rotational parameters around these axes. These position and rotation parameters are encapsulated in a 4 × 4 transformation matrix, as described in Eq.2.

In this matrix, R₁, R₂, and R₃ are derived from Eq.3 to 5, respectively.

In the robotic coordinate system, any vector X in homogeneous coordinates transitions from an initial position X₀ to a final position X₁ through the transformation matrix, as expressed in Eq.6.

The transformation matrix M serves as a mathematical tool to elucidate the relationship between multiple coordinate systems. For instance, in the robotic coordinate system, the relationship between the collimator coordinate system C₁’ and the patient coordinate system C₂’ can be expressed as a transformation matrix, illustrated in Eq.7.

Similarly, within NeuMANTA’s imaging coordinate system, the collimator and patient surface coordinate systems — C₁ and C₂ — share an analogous relationship, as shown in Eq.8.

Utilizing the parameters of transformation matrix M₁’, a corresponding set of robotic arm control parameters can be algorithmically determined, thereby achieving the objective of automated patient positioning.

To examine the impact of positioning deviation on dosimetry, anthropomorphic phantom images were employed to simulate superficial and deep-seated tumors, both modeled as spheroids with a diameter of 3cm, and organs at risk (OAR) including skin, soft tissue, cranium, and brain with homogeneous material retrieved from ICRU report 46¹⁸. The measured depths from the skin to the superficial and deep-seated tumor centers were 35mm and 75mm, respectively, with a baseline SSD of 5mm, as shown in Fig.7. Note that SSD is the distance from the center of the collimator outlet surface to the skin. These simulations were run through the NeuMANTA and the dose distributions were calculated using the Monte Carlo dose calculation engine, COMPASS (COMpact PArticle Simulation System)^19,20,21.

The NeuPex AB-BNCT source²², 10cm collimator aperture, and assuming boron concentrations of 25 ppm in OARs and 75 ppm in tumor were used to evaluate the dose distribution for different positioning deviations. As calculating BNCT bioequivalent dose, the relative biological effectiveness (RBE) was 3.2 for neutron, and compound biological effectiveness (CBE) were 2.5, 1.3, 1.3, 1.3, and 3.8 for skin, soft tissue, cranium, brain, and tumor, respectively²³.

Validation

The seed points used to identify phantom CT images in NeuMANTA are shown in Fig.5b, and their coordinates are listed in Table1. This step is critical to achieve accurate and reliable fiducial mark image registration. After clinical acceptance of the treatment plan, the position of phantom was confirmed as shown in Fig.8. The source center position of the collimator port in the image coordinate system is SrcX = -80.3mm, SrcY = 48.5mm, SrcZ = 76.2mm, and the source angles is θ = 90 °, φ = 0.0 °.

The phantom was securely affixed to the robot-mounted treatment couch and aligned at the starting position. Laser alignment ensured that the seed points corresponded precisely with their intended positions, as illustrated in Table2; Fig.9a. Based on this initial alignment, the final robot position was calculated and the robot was subsequently repositioned to match the designated coordinates in NeuMANTA. This process is documented in Fig.9b, c.

Accuracy

To ascertain the stability and precision of the positioning method, four representative beam position parameters were selected for validation. These parameters are enumerated in Table3, while the corresponding robot control parameters derived from them are listed in Table4. Post-positioning, discrepancies were measured in two key dimensions: the SSD and the surface deviation (i.e., the offset between the beam centerline on the phantom surface and the designated reference point). As summarized in Table5, the maximum deviation observed in the SSD dimension was under 3mm, and the surface deviation was less than 2mm.

The validation of innovative robotic-assisted patient positioning technique was performed using an anthropomorphic phantom, revealing a maximum deviation of approximately 3mm in the SSD direction and 2mm along the surface. Intriguingly, the analysis revealed that the minimum positional deviation was realized when there was no rotation of the robotic arm, whereas the maximum deviation occurred when the rotation angle approached 90°. As we know, greater rotation angles are more likely to result in errors. For a robot, when the rotational angle error remains constant, increasing the radius leads to larger potential deviations. In the robot’s coordinate system, the origin is established at the support structure fixed in the ceiling, so the rotation radius is approximately 2m, underscoring the significant influence of the robot’s rotation angle on positioning accuracy.

The dose outcomes for the nominal treatment plan served as the reference are shown as the solid lines in Fig.10. The dashed line and dash-dotted line were the dose results when the SSD is shortened by 2mm and increased by 3mm respectively, and the movement represented by the dotted line is the dose result of the shift perpendicular to the beam direction (y, z). The impact of positioning on robust BNCT dose is most significant for superficial tumor, followed by deep-seated tumor and has a lower impact on OARs. Tables6 and 7, and 8 present the dosimetric effects of positional deviations on normal tissues, superficial tumors, and deep-seated tumors, respectively. It has been observed that under the nominal SSD, deviations in the y- and z-direction of positioning result in maximum dose rate (D_max)change of less than 1.2% for the OARs and mean dose rate (D_mean) change of less than 1.8% for the brain. However, for tumor, whether superficial or deep-seated, the variation in D_mean and D₈₀ is within 1.4%, and the minimum dose rate (D_min) is within 2.5%. When the source incident is correct but the SSD was shortened by 2mm from the nominal, the dose rate increase within 2.7% for OARs and within 2.2% for tumor. On the contrary, when the SSD is increased by 3mm, the dose rates of OAR and tumor are reduced within 4.1%. Nevertheless, as the positioning of the SSD and the incident point are both different from the nominal treatment plan, the dose rate changes in OAR and tumor increased slightly, but the dose rate error not exceed 4.7%.

It is imperative to consider that the extent of positioning deviation correlates positively with the increase in SSD. Although this may impose limitations on the method’s efficacy for higher SSD values, it should be noted that in real-world treatment scenarios, the patient is generally positioned as close to the collimator as feasible. This helps to control SSD deviation effectively, mitigating its impact on treatment accuracy.

This study introduces a novel robotic-assisted patient positioning system specifically tailored for BNCT. Leveraging industrial-grade robotic technology, the system bridges the gap between theoretical treatment planning and its actual execution, offering a marked advancement over current patient positioning methods in BNCT. Our system, fully integrated with the NeuMANTA treatment planning system, automates the traditionally manual and often error-prone process of patient positioning. Validation using an anthropomorphic phantom model confirms the system’s high level of accuracy, with a maximum observed deviation of 3mm in the SSD and 2mm along the surface. The system shows particular promise for enhancing the precision and effectiveness of BNCT treatments, specifically in settings where precise targeting is required to mitigate collateral damage to surrounding tissues.

The dosimetric impact of positional deviations was also thoroughly examined, revealing that even if the surface orientation deviation exceeds the maximum deviation of the positioning system, the resulting dose change is less than 1.8% for OARs and 2.5% for tumor within a 4mm range. When the SSD deviation is within 3mm, the change in the D_mean and D₈₀ is approximately 3%. In conclusion, the robotic patient positioning system proposed in this study demonstrates substantial promise for enhancing the precision and effectiveness of BNCT treatments. However, the limitations identified, such as the influence of robot rotation and SSD on positioning accuracy, warrant further investigation and potential methodological refinements to optimize treatment outcomes.

Future work

While the current study lays a solid foundation, several avenues for future research and development emerge. Rotation Sensitivity: The study points to the sensitivity of positioning accuracy to the angle of robot rotation. Future work could focus on improving the robotic arm’s rotational algorithms to reduce this sensitivity. To guarantee accurate and precise positioning in a real clinical treatment practice, it may be beneficial to employ secondary verification methods such as optical tracking. This verification method would serve to confirm patient positioning. Dosimetric Impact: The present work outlines the dosimetric consequences of positional inaccuracies. A logical next step would be the development of real-time dosimetry to provide instant feedback during treatment.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Authors and Affiliations

1. Neuboron Therapy System Ltd, Xiamen, Fujian Province, P.R. China

   Jiang Chen,Yi-Chiao Teng,Diyun Shu,Qiuping Gong,Qiaolin Xie&Yuan-Hao Liu

2. Nanjing Vocational University of Industry Technology, Nanjing, Jiangsu Province, P.R. China

   Jiang Chen

3. National Tsing Hua University, Hsinchu, 30013, Taiwan, Republic of China

   Yi-Chiao Teng

4. Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu Province, P.R. China

   Yuan-Hao Liu

5. Neuboron Medtech Ltd, Nanjing, Jiangsu Province, P.R. China

   Yuan-Hao Liu

6. Xiamen Humanity Hospital, Xiamen, Fujian Province, P.R. China

   Yuan-Hao Liu

Contributions

Jiang Chen contributed to the concept of this study and writing; Diyun Shu, Qiuping Gong, Qiaolin Xie performed the experiment; Yi-Chiao Teng investigated the dosimetric influence of position deviation; Yuan-Hao Liu contributed to reviewing and editing.

Corresponding author

Correspondence to Yuan-Hao Liu.

Competing interests

The authors declare no competing interests.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions