Automatic quality assurance of radiotherapy treatment plans using Bayesian networks: A multi-institutional study.

Purpose: Artificial intelligence applications in radiation oncology have been the focus of study in the last decade. The introduction of automated and intelligent solutions for routine clinical tasks, such as treatment planning and quality assurance, has the potential to increase safety and efficiency of radiotherapy. In this work, we present a multi-institutional study across three different institutions internationally on a Bayesian network (BN)-based initial plan review assistive tool that alerts radiotherapy professionals for potential erroneous or suboptimal treatment plans. Methods: Clinical data were collected from the oncology information systems in three institutes in Europe (Maastro clinic - 8753 patients treated between 2012 and 2020) and the United States of America (University of Vermont Medical Center [UVMMC] - 2733 patients, University of Washington [UW] - 6180 patients, treated between 2018 and 2021). We trained the BN model to detect potential errors in radiotherapy treatment plans using different combinations of institutional data and performed single-site and cross-site validation with simulated plans with embedded errors. The simulated errors consisted of three different categories: i) patient setup, ii) treatment planning and iii) prescription. We also compared the strategy of using only diagnostic parameters or all variables as evidence for the BN. We evaluated the model performance utilizing the area under the receiver-operating characteristic curve (AUC). Results: The best network performance was observed when the BN model is trained and validated using the dataset in the same center. In particular, the testing and validation using UVMMC data has achieved an AUC of 0.92 with all parameters used as evidence. In cross-validation studies, we observed that the BN model performed better when it was trained and validated in institutes with similar technology and treatment protocols (for instance, when testing on UVMMC data, the model trained on UW data achieved an AUC of 0.84, compared with an AUC of 0.64 for the model trained on Maastro data). Also, combining training data from larger clinics (UW and Maastro clinic) and using it on smaller clinics (UVMMC) leads to satisfactory performance with an AUC of 0.85. Lastly, we found that in general the BN model performed better when all variables are considered as evidence. Conclusion: We have developed and validated a Bayesian network model to assist initial treatment plan review using multi-institutional data with different technology and clinical practices. The model has shown good performance even when trained on data from clinics with divergent profiles, suggesting that the model is able to adapt to different data distributions.

Automated planning stage tracking and analysis through an integrated whiteboard system.

PURPOSE: One of the common challenges in delivering complex healthcare procedures such as radiation oncology is the organization and sharing of information in ways that facilitate workflow and prevent treatment delays. Within the major vendors of Oncology Information Systems (OIS) is a lack of tools and displays to assist in task timing and workflow processes. To address this issue, we developed an electronic whiteboard integrated with a local OIS to track, record, and evaluate time frames associated with clinical radiation oncology treatment planning processes. METHODS: We developed software using an R environment hosted on a local web-server at Seattle Cancer Care Alliance (SCCA) in 2017. The planning process was divided into stages, and time-stamped moves between planning stages were recorded automatically via Mosaiq (Elekta, Sweden) Quality Check Lists (QCLs). Whiteboard logs were merged with Mosaiq-extracted diagnostic factors and evaluated for significance. Interventional changes to task time expectations were evaluated for 6 months in 2021 and compared with 6 month periods in 2018 and 2019. RESULTS: Whiteboard/Mosaiq data from the SCCA show that treatment intent, number of prescriptions, and nodal involvement were main factors influencing overall time to plan completion. Contouring and Planning times were improved by 2.6 days (p<10-14) and 2.5 days (p<10-11), respectively. Overall time to plan completion was reduced by 33% (5.1 days; p<10-11). CONCLUSIONS: This report establishes the utility of real-time task tracking tools in a radiotherapy planning process. The whiteboard results provide data-driven evidence to add justification for practice change implementations.

The dosimetric benefit of in-advance respiratory training for deep inspiration breath holding is realized during daily treatment in left breast radiotherapy: A comparative retrospective study of serial surface motion tracking.

INTRODUCTION: A novel approach of in-advance preparatory respiratory training and practice for deep inspiration breath holding (DIBH) has been shown to further reduce cardiac dose in breast cancer radiotherapy patients, enabled by deeper (extended) DIBH. Here we investigated the consistency and stability of such training-induced extended DIBH after training completion and throughout the daily radiotherapy course. METHODS: Daily chestwall motion from real-time surface tracking transponder data was analysed in 67 left breast radiotherapy patients treated in DIBH. Twenty-seven received preparatory DIBH training/practice (prep Trn) 1-2 weeks prior to CT simulation, resulting in an extended DIBH (ext DIBH) and reduced cardiac dose at simulation. Forty had only conventional immediate pre-procedure DIBH instruction without prep Trn and without extended DIBH (non-Trn group). Day-to-day variability in chestwall excursion pattern during radiotherapy was compared among the groups. RESULTS: The average of daily maximum chestwall excursions was overall similar, 2.5 ± 0.6 mm for prep Trn/ext DIBH vs. 2.9 ± 0.8 mm for non-Trn patients (P = 0.24). Chestwall excursions beyond the 3-mm tolerance threshold were less common in the prep Trn/ext DIBH group (18.8% vs. 37.5% of all fractions within the respective groups, P = 0.038). Among patients with cardiopulmonary disease those with prep Trn/ext DIBH had fewer chestwall excursions beyond 3 mm (9.4% vs. 46.7%, P = 0.023) and smaller average maximum excursions than non-Trn patients (2.4 ± 0.3 vs. 3.0 ± 0.6 mm, P = 0.047, respectively). CONCLUSION: Similar stability of daily DIBH among patients with and without preparatory training/practice suggests that the training-induced extended DIBH and cardiac dose reductions were effectively sustained throughout the radiotherapy course. Training further reduced beyond-tolerance chestwall excursions, particularly in patients with cardiopulmonary disease.

Reducing Cardiac Radiation Dose From Breast Cancer Radiation Therapy With Breath Hold Training and Cognitive Behavioral Therapy.

The delivery of radiation therapy shares many of the challenges encountered in imaging procedures. As in imaging, such as MRI, organ motion must be reduced to a minimum, often for lengthy time periods, to effectively target the tumor during imaging-guided therapy while reducing radiation dose to nearby normal tissues. For patients, radiation therapy is frequently a stress- and anxiety-provoking medical procedure, evoking fear from negative perceptions about irradiation, confinement from immobilization devices, claustrophobia, unease with equipment, physical discomfort, and overall cancer fear. Such stress can be a profound challenge for cancer patients' emotional coping and tolerance to treatment, and particularly interferes with advanced radiation therapy procedures where active, complex and repetitive high-level cooperation is often required from the patient.In breast cancer, the most common cancer in women worldwide, radiation therapy is an indispensable component of treatment to improve tumor control and outcome in both breast-conserving therapy for early-stage disease and in advanced-stage patients. High technological complexity and high patient cooperation is required to mitigate the known cardiac toxicity and mortality from breast cancer radiation by reducing the unintended radiation dose to the heart from left breast or left chest wall irradiation. To address this, radiation treatment in daily deep inspiration breath hold (DIBH), to create greater distance between the treatment target and the heart, is increasingly practiced. While holding the promise to decrease cardiac toxicity, DIBH procedures often augment patients' baseline stress and anxiety reaction toward radiation treatment. Patients are often overwhelmed by the physical and mental demands of daily DIBH, including the nonintuitive timed and sustained coordination of abdominal thoracic muscles for prolonged breath holding.While technologies, such as DIBH, have advanced to millimeter-precision in treatment delivery and motion tracking, the "human factor" of patients' ability to cooperate and perform has been addressed much less. Both are needed to optimally deliver advanced radiation therapy with minimized normal tissue effects, while alleviating physical and cognitive distress during this challenging phase of breast cancer therapy.This article discusses physical training and psychotherapeutic integrative health approaches, applied to radiation oncology, to leverage and augment the gains enabled by advanced technology-based high-precision radiation treatment in breast cancer. Such combinations of advanced technologies with training and cognitive integrative health interventions hold the promise to provide simple feasible and low-cost means to improve patient experience, emotional outcomes and quality of life, while optimizing patient performance for advanced imaging-guided treatment procedures - paving the way to improve cardiac outcomes in breast cancer survivors.

Radiation Therapy Quality Assurance Tasks and Tools: The Many Roles of Machine Learning.

The recent explosion in machine learning efforts in the quality assurance (QA) space has produced a variety of proofs-of-concept many with promising results. Expected outcomes of model implementation include improvements in planning time, plan quality, advanced dosimetric QA, predictive machine maintenance, increased safety checks, and developments key for new QA paradigms driven by adaptive planning. In this article, we outline several areas of research and discuss some of the unique challenges each area presents.

Characterizing a deformable registration algorithm for surface-guided breast radiotherapy.

PURPOSE: Surface-guided radiation therapy (SGRT) is a nonionizing imaging approach for patient setup guidance, intra-fraction monitoring, and automated breath-hold gating of radiation treatments. SGRT employs the premise that the external patient surface correlates to the internal anatomy, to infer the treatment isocenter position at time of treatment delivery. Deformations and posture variations are known to impact the correlation between external and internal anatomy. However, the degree, magnitude, and algorithm dependence of this impact are not intuitive and currently no methods exist to assess this relationship. The primary aim of this work was to develop a framework to investigate and understand how a commercial optical surface imaging system (C-RAD, Uppsala, Sweden), which uses a nonrigid registration algorithm, handles rotations and surface deformations. METHODS: A workflow consisting of a female torso phantom and software-introduced transformations to the corresponding digital reference surface was developed. To benchmark and validate the approach, known rigid translations and rotations were first applied. Relevant breast radiotherapy deformations related to breast size, hunching/arching back, distended/deflated abdomen, and an irregular surface to mimic a cover sheet over the lower part of the torso were investigated. The difference between rigid and deformed surfaces was evaluated as a function of isocenter location. RESULTS: For all introduced rigid body transformations, C-RAD computed isocenter shifts were determined within 1 mm and 1˚. Additional translational shifts to correct for rotations as a function of isocenter location were determined with the same accuracy. For yaw setup errors, the difference in shift corrections between a plan with an isocenter placed in the center of the breast (BrstIso) and one located 12 cm superiorly (SCFIso) was 2.3 mm/1˚ in lateral direction. Pitch setup errors resulted in a difference of 2.1 mm/1˚ in vertical direction. For some of the deformation scenarios, much larger differences up to 16 mm and 7˚ in the calculated shifts between BrstIso and SCFIso were observed that could lead to large unintended gaps or overlap between adjacent matched fields if uncorrected. CONCLUSIONS: The methodology developed lends itself well for quality assurance (QA) of SGRT systems. The deformable C-RAD algorithm determined accurate shifts for rigid transformations, and this was independent of isocenter location. For surface deformations, the position of the isocenter had considerable impact on the registration result. It is recommended to avoid off-axis isocenters during treatment planning to optimally utilize the capabilities of the deformable image registration algorithm, especially when multiple isocenters are used with fields that share a field edge.

Predictors of cardiac and lung dose sparing in DIBH for left breast treatment.

This retrospective study of left breast radiation therapy (RT) investigates the correlation between anatomical parameters and dose to heart or/and left lung in deep inspiration breath-hold (DIBH) compared to free-breathing (FB) technique. Anatomical parameters of sixty-seven patients, treated with a step-and-shoot technique to 50 Gy or 50.4 Gy were included. They consisted of the cardiac contact distances in axial (CCDax) and parasagittal (CCDps) planes, and the lateral heart-to-chest distance (HCD). Correlation analysis was performed to identify predictors for heart and lung dose sparing. Paired t-test and linear regression were used for data analysis with significance level of p = 0.05. All dose metrics for heart and lung were significantly reduced with DIBH, however 21% of patients analyzed had less than 1.0 Gy mean heart dose reduction. Both FB-CCDpsdistance and FB-HCD correlated with FB mean heart dose and mean DIBH heart dose reduction. The strongest correlation was observed for the ratio of FB-CCDpsand FB-HCD with heart dose sparing. A FB-CCDps and FB-HCD model was developed to predict DIBH induced mean heart dose reduction, with 1.04 Gy per unit of FB-CCDps/FB-HCD. Variation between predicted and actual mean heart dose reduction ranged from -0.6 Gy to 0.6 Gy. In this study, FB-CCDps and FB-HCD distance served as predictors for heart dose reduction with DIBH equally, with FB-CCDps/FB-HCD as a stronger predictor. These parameters and the prediction model could be further investigated for use as a tool to better select patients who will benefit from DIBH.

Development and Validation of a Bayesian Network Method to Detect External Beam Radiation Therapy Physician Order Errors.

PURPOSE: To investigate a Bayesian network (BN)-based method to detect errors in external beam radiation therapy physician orders. METHODS AND MATERIALS: A total of 4431 external beam radiation therapy orders from 2008 to 2017 at the authors' institution were obtained from clinical treatment management systems and divided into 3 groups: single prescription, concurrent boost, and sequential boost. Multiple BNs were developed for each group to detect errors in new orders using joint posterior probabilities of the order parameters, given disease information. Each BN was trained with a group of orders using a Bayesian learning algorithm. A procedure was developed to select the optimal BN for each treatment site in each group and to determine site-specific parameters and error detection thresholds. Potential clinical errors, created both manually and automatically, were applied to test error detection performance. RESULTS: The average true-positive rate (TPR) and false-positive rate (FPR) of error detection were 95.72% and 1.99%, respectively, for the single-prescription cohort with 9 treatment sites. For the concurrent-boost cohort, the TPR and FPR were 92.94% and 14.53%, respectively. For the sequential-boost cohort, the TPR and FPR were 100% and 9.48%, respectively, for the prescribed dose values and 100% and 4.34%, respectively, for the remaining order parameters. For the patient simulation and imaging parameters for 9 treatment sites, the TPR and FPR were 100% and 4.96%, respectively. CONCLUSIONS: The probabilistic BN method was able to perform physician order error detection at a higher accuracy than previously reported in a variety of complex prescription instances, thus warranting further development in incorporating BNs into clinical error detection tools to assist manual physician order checks.

Accuracy and stability of deep inspiration breath hold in gated breast radiotherapy - A comparison of two tracking and guidance systems.

PURPOSE: To characterize reproducibility of patient breath-hold positioning and compare tracking system performance for Deep Inspiration Breath Hold (DIBH) gated left breast radiotherapy. METHODS: 29 consecutive left breast DIBH patients (655 fractions) were treated under the guidance of Calypso surface beacons with audio-feedback and 35 consecutive patients (631 fractions) were treated using C-RAD Catalyst HD surface imaging with audiovisual feedback. The Calypso system tracks a centroid determined by two radio-frequency transponders, with a manually enforced institutional tolerance, while the surface image based CatalystHD system utilizes real-time biometric feedback to track a pre-selected point with an institutional tolerance enforced by the Elekta Response gating interface. DIBH motion data from Calypso was extracted to obtain the displacement of breath hold marker in ant/post direction from a set-zero reference point. Ant/post point displacement data from CatalystHD was interpreted by computing the difference between raw tracking points and the center of individual gating windows. Mean overall errors were compared using Welsh's unequal variance t-test. Wilcoxon rank sum test were used for statistical analysis with P < 0.05 considered significant. RESULTS: Mean overall error for Calypso and CatalystHD were 0.33 ±â€¯1.17 mm and 0.22 ±â€¯0.43 mm, respectively, with t-test comparison P-value < 0.034. Absolute errors for Calypso and CatalystHD were 0.95 ±â€¯0.75 mm and 0.38 ±â€¯0.30 mm, respectively, with Wilcoxon rank sum test P-value <2×10-16. Average standard deviation per fraction was found to be 0.74 ±â€¯0.44 mm for Calypso patients versus 0.54 ±â€¯0.22 mm for CatalystHD. CONCLUSION: Reduced error distribution widths in overall positioning, deviation of position, and per fraction deviation suggest that the use of functionalities available in CatalystHD such as audiovisual biofeedback and patient surface matching improves accuracy and stability during DIBH gated left breast radiotherapy.

Characterization of a Bayesian network-based radiotherapy plan verification model.

PURPOSE: The current process for radiotherapy treatment plan quality assurance relies on human inspection of treatment plans, which is time-consuming, error prone and oft reliant on inconsistently applied professional judgments. A previous proof-of-principle paper describes the use of a Bayesian network (BN) to aid in this process. This work studied how such a BN could be expanded and trained to better represent clinical practice. METHODS: We obtained 51 540 unique radiotherapy cases including diagnostic, prescription, plan/beam, and therapy setup factors from a de-identified Elekta oncology information system from the years 2010-2017 from a single institution. Using a knowledge base derived from clinical experience, factors were coordinated into a 29-node, 40-edge BN representing dependencies among the variables. Conditional probabilities were machine learned using expectation maximization module using all data except a subset of 500 patient cases withheld for testing. Different classes of errors that were obtained from incident learning systems were introduced to the testing set of cases which were withheld from the dataset used for building the BN. Different sizes of datasets were used to train the network. In addition, the BN was trained using data from different length epochs as well as different eras. Its performance under these different conditions was evaluated by means of Areas Under the receiver operating characteristic Curve (AUC). RESULTS: Our performance analysis found AUCs of 0.82, 0.85, 0.89, and 0.88 in networks trained with 2-yr, 3-yr 4-yr and 5-yr windows. With a 4-yr sliding window, we found AUC reduction of 3% per year when moving the window back in time in 1-yr steps. Compared to the 4-yr window moved back by 4 yrs (2010-2013 vs 2014-2017), the largest component of overall reduction in AUC over time was from the loss of detection performance in plan/beam error types. CONCLUSIONS: The expanded BN method demonstrates the ability to detect classes of errors commonly encountered in radiotherapy planning. The results suggest that a 4-yr training dataset optimizes the performance of the network in this institutional dataset, and that yearly updates are sufficient to capture the evolution of clinical practice and maintain fidelity.

Dosimetric effects of bolus and lens shielding in treating ocular lymphomas with low-energy electrons.

Radiation therapy is an effective treatment for primary orbital lymphomas. Lens shielding with electrons can reduce the risk of high-grade cataracts in patients undergoing treatment for superficial tumors. This work evaluates the dosimetric effects of a suspended eye shield, placement of bolus, and varying electron energies. Film (GafChromic EBT3) dosimetry and relative output factors were measured for 6, 8, and 10 MeV electron energies. A customized 5-cm diameter circle electron orbital cutout was constructed for a 6 × 6-cm applicator with a suspended lens shield (8-mm diameter Cerrobend cylinder, 2.2-cm length). Point doses were measured using a scanning electron diode in a solid water phantom at depths representative of the anterior and posterior lens. Depth dose profiles were compared for 0-mm, 3-mm, and 5-mm bolus thicknesses. At 5 mm (the approximate distance of the anterior lens from the surface of the cornea), the percent depth dose under the suspended lens shield was reduced to 15%, 15%, and 14% for electron energies 6, 8, and 10 MeV, respectively. Applying bolus reduced the benefit of lens shielding by increasing the estimated doses under the block to 27% for 3-mm and 44% for 5-mm bolus for a 6 MeV incident electron beam. This effect is minimized with 8 MeV electron beams where the corresponding values were 15.5% and 18% for 3-mm and 5-mm bolus. Introduction of a 7-mm hole in 5-mm bolus to stabilize eye motion during treatment altered lens doses by about 1%. Careful selection of electron energy and consideration of bolus effects are needed to account for electron scatter under a lens shield.

Developing Bayesian networks from a dependency-layered ontology: A proof-of-concept in radiation oncology.

PURPOSE: Bayesian networks (BNs) are graphical representations of probabilistic knowledge that offer normative reasoning under uncertainty and are well suited for use in medical domains. Traditional knowledge-based network development of BN topology requires that modeling experts establish relevant dependency links between domain concepts by searching and translating published literature, querying domain experts, or applying machine learning algorithms on data. For initial development these methods are time-intensive and this cost hinders the growth of BN applications in medical decision making. Further, this approach fails to utilize knowledge representation in medical fields to automate network development. Our research alleviates the challenges surrounding BN modeling in radiation oncology by leveraging an ontology based hub and spoke system for BN construction. METHODS: We implement a hub and spoke system by developing (a) an ontology of knowledge in radiation oncology (the hub) which includes dependency semantics similar to BN relations and (b) a software tool that operates on ontological semantics using deductive reasoning to create BN topologies (the spokes). We demonstrate that network topologies built using the software are terminologically consistent and form networks that are topologically compatible with existing ones. We do this first by merging two different BN models for prostate cancer radiotherapy prediction which contain domain cross terms. We then use the logic to perform discovery of new causal chains between radiation oncology concepts. RESULTS: From the radiation oncology (RO) ontology we successfully reconstructed a previously published prostate cancer radiotherapy Bayes net using up-to-date domain knowledge. Merging this model with another similar prostate cancer model in the RO domain produced a larger, highly interconnected model representing the expanded scope of knowledge available regarding prostate cancer therapy parameters, complications, and outcomes. The causal discovery resulted in an automatically-built causal network model of all ontologized radiotherapy concepts between a 'Mucositis' complication and anatomic tumor location. CONCLUSIONS: The proposed model building approach lowers barriers to developing probabilistic models relevant to real-world clinical decision making, and offers a solution to the consistency and compatibility problems. Further, the knowledge representation in this work demonstrates potential for broader radiation oncology applications outside of Bayes nets.

Dosimetric comparison of single-beam multi-arc and 2-beam multi-arc VMAT optimization in the Monaco treatment planning system.

The purpose of this study was to evaluate the dosimetric and practical effects of the Monaco treatment planning system "max arcs-per-beam" optimization parameter in pelvic radiotherapy treatments. We selected for this study a total of 17 previously treated patients with a range of pelvic disease sites including prostate (9), bladder (1), uterus (3), rectum (3), and cervix (1). For each patient, 2 plans were generated, one using an arc-per-beam setting of "1" and another with an arc-per-beam setting of "2" using the volumes and constraints established from the initial clinical treatments. All constraints and dose coverage objects were kept the same between plans, and all plans were normalized to 99.7% to ensure 100% of the planning target volume (PTV) received 95% of the prescription dose. Plans were evaluated for PTV conformity, homogeneity, number of monitor units, number of control points, and overall plan acceptability. Treatment delivery time, patient-specific quality assurance procedures, and the impact on clinical workflow were also assessed. We found that for complex-shaped target volumes (small central volumes with extending arms to cover nodal regions), the use of 2 arc-per-beam (2APB) parameter setting achieved significantly lower average dose-volume histogram values for the rectum V20 (p = 0.0012) and bladder V30 (p = 0.0036) while meeting the high dose target constraints. For simple PTV shapes, we found reduced monitor units (13.47%, p = 0.0009) and control points (8.77%, p = 0.0004) using 2APB planning. In addition, we found a beam delivery time reduction of approximately 25%. In summary, the dosimetric benefit, although moderate, was improved over a 1APB setting for complex PTV, and equivalent in other cases. The overall reduced delivery time suggests that the use of mulitple arcs per beam could lead to reduced patient-on-table time, increased clinical throughput, and reduced medical physics quality assurance effort.

Electron beam energy QA - a note on measurement tolerances.

Monthly QA is recommended to verify the constancy of high-energy electron beams generated for clinical use by linear accelerators. The tolerances are defined as 2%/2 mm in beam penetration according to AAPM task group report 142. The practical implementation is typically achieved by measuring the ratio of readings at two different depths, preferably near the depth of maximum dose and at the depth corresponding to half the dose maximum. Based on beam commissioning data, we show that the relationship between the ranges of energy ratios for different electron energies is highly nonlinear. We provide a formalism that translates measurement deviations in the reference ratios into change in beam penetration for electron energies for six Elekta (6-18 MeV) and eight Varian (6-22 MeV) electron beams. Experimental checks were conducted for each Elekta energy to compare calculated values with measurements, and it was shown that they are in agreement. For example, for a 6 MeV beam a deviation in the measured ionization ratio of ± 15% might still be acceptable (i.e., be within ± 2 mm), whereas for an 18 MeV beam the corresponding tolerance might be ± 6%. These values strongly depend on the initial ratio chosen. In summary, the relationship between differences of the ionization ratio and the corresponding beam energy are derived. The findings can be translated into acceptable tolerance values for monthly QA of electron beam energies.

Bayesian network models for error detection in radiotherapy plans.

The purpose of this study is to design and develop a probabilistic network for detecting errors in radiotherapy plans for use at the time of initial plan verification. Our group has initiated a multi-pronged approach to reduce these errors. We report on our development of Bayesian models of radiotherapy plans. Bayesian networks consist of joint probability distributions that define the probability of one event, given some set of other known information. Using the networks, we find the probability of obtaining certain radiotherapy parameters, given a set of initial clinical information. A low probability in a propagated network then corresponds to potential errors to be flagged for investigation. To build our networks we first interviewed medical physicists and other domain experts to identify the relevant radiotherapy concepts and their associated interdependencies and to construct a network topology. Next, to populate the network's conditional probability tables, we used the Hugin Expert software to learn parameter distributions from a subset of de-identified data derived from a radiation oncology based clinical information database system. These data represent 4990 unique prescription cases over a 5 year period. Under test case scenarios with approximately 1.5% introduced error rates, network performance produced areas under the ROC curve of 0.88, 0.98, and 0.89 for the lung, brain and female breast cancer error detection networks, respectively. Comparison of the brain network to human experts performance (AUC of 0.90 ± 0.01) shows the Bayes network model performs better than domain experts under the same test conditions. Our results demonstrate the feasibility and effectiveness of comprehensive probabilistic models as part of decision support systems for improved detection of errors in initial radiotherapy plan verification procedures.

Validation of the Pinnacle³ photon convolution-superposition algorithm applied to fast neutron beams.

We evaluate a photon convolution-superposition algorithm used to model a fast neutron therapy beam in a commercial treatment planning system (TPS). The neutron beam modeled was the Clinical Neutron Therapy System (CNTS) fast neutron beam produced by 50 MeV protons on a Be target at our facility, and we implemented the Pinnacle3 dose calculation model for computing neutron doses. Measured neutron data were acquired by an IC30 ion chamber flowing 5 cc/min of tissue equivalent gas. Output factors and profile scans for open and wedged fields were measured according to the Pinnacle physics reference guide recommendations for photon beams in a Wellhofer water tank scanning system. Following the construction of a neutron beam model, computed doses were then generated using 100 monitor units (MUs) beams incident on a water-equivalent phantom for open and wedged square fields, as well as multileaf collimator (MLC)-shaped irregular fields. We compared Pinnacle dose profiles, central axis doses, and off-axis doses (in irregular fields) with 1) doses computed using the Prism treatment planning system, and 2) doses measured in a water phantom and having matching geometry to the computation setup. We found that the Pinnacle photon model may be used to model most of the important dosimetric features of the CNTS fast neutron beam. Pinnacle-calculated dose points among open and wedged square fields exhibit dose differences within 3.9 cGy of both Prism and measured doses along the central axis, and within 5 cGy difference of measurement in the penumbra region. Pinnacle dose point calculations using irregular treatment type fields showed a dose difference up to 9 cGy from measured dose points, although most points of comparison were below 5 cGy. Comparisons of dose points that were chosen from cases planned in both Pinnacle and Prism show an average dose difference less than 0.6%, except in certain fields which incorporate both wedges and heavy blocking of the central axis. All clinical cases planned in both Prism and Pinnacle were found to be comparable in terms of dose-volume histograms and spatial dose distribution following review by the treating clinicians. Variations were considered minor and within clinically acceptable limits by the treating clinicians. The Pinnacle TPS has sufficient computational modeling ability to adequately produce a viable neutron model for clinical use in treatment planning.