Standardized beam bouquets for lung IMRT planning

Automatic planning for functional lung avoidance radiotherapy based on function-guided beam angle selection and plan optimization

Objective.This study aims to develop a fully automatic planning framework for functional lung avoidance radiotherapy (AP-FLART).Approach.The AP-FLART integrates a dosimetric score-based beam angle selection method and a meta-optimization-based plan optimization method, both of which incorporate lung function information to guide dose redirection from high functional lung (HFL) to low functional lung (LFL). It is applicable to both contour-based FLART (cFLART) and voxel-based FLART (vFLART) optimization options. A cohort of 18 lung cancer patient cases underwent planning-CT and SPECT perfusion scans were collected. AP-FLART was applied to generate conventional RT (ConvRT), cFLART, and vFLART plans for all cases. We compared automatic against manual ConvRT plans as well as automatic ConvRT against FLART plans, to evaluate the effectiveness of AP-FLART. Ablation studies were performed to evaluate the contribution of function-guided beam angle selection and plan optimization to dose redirection.Main results.Automatic ConvRT plans generated by AP-FLART exhibited similar quality compared to manual counterparts. Furthermore, compared to automatic ConvRT plans, HFL mean dose,V20, andV5were significantly reduced by 1.13 Gy (p< .001), 2.01% (p< .001), and 6.66% (p< .001) respectively for cFLART plans. Besides, vFLART plans showed a decrease in lung functionally weighted mean dose by 0.64 Gy (p< .01),fV20by 0.90% (p= 0.099), andfV5by 5.07% (p< .01) respectively. Though inferior conformity was observed, all dose constraints were well satisfied. The ablation study results indicated that both function-guided beam angle selection and plan optimization significantly contributed to dose redirection.Significance.AP-FLART can effectively redirect doses from HFL to LFL without severely degrading conventional dose metrics, producing high-quality FLART plans. It has the potential to advance the research and clinical application of FLART by providing labor-free, consistent, and high-quality plans.

A reinforcement learning application of a guided Monte Carlo Tree Search algorithm for beam orientation selection in radiation therapy

Current beam orientation optimization algorithms for radiotherapy, such as column generation (CG), are typically heuristic or greedy in nature because of the size of the combinatorial problem, which leads to suboptimal solutions. We propose a reinforcement learning strategy using a Monte Carlo Tree Search (MCTS) that can find a better beam orientation set in less time than CG. We utilize a reinforcement learning structure involving a supervised learning network to guide the MCTS and to explore the decision space of beam orientation selection problems. We previously trained a deep neural network (DNN) that takes in the patient anatomy, organ weights, and current beams, then approximates beam fitness values to indicate the next best beam to add. Here, we use this DNN to probabilistically guide the traversal of the branches of the Monte Carlo decision tree to add a new beam to the plan. To assess the feasibility of the algorithm, we used a test set of 13 prostate cancer patients, distinct from the 57 patients originally used to train and validate the DNN, to solve five-beam plans. To show the strength of the guided MCTS (GTS) compared to other search methods, we also provided the performances of Guided Search, Uniform Tree Search and Random Search algorithms. On average, GTS outperformed all the other methods. It found a better solution than CG in 237 s on average, compared to 360 s for CG, and outperformed all other methods in finding a solution with a lower objective function value in less than 1000 s. Using our GTS method, we could maintain planning target volume (PTV) coverage within 1% error similar to CG, while reducing the organ-at-risk mean dose for body, rectum, left and right femoral heads; the mean dose to bladder was 1% higher with GTS than with CG.

On the Importance of Individualized, Non-Coplanar Beam Configurations in Mediastinal Lymphoma Radiotherapy, Optimized With Automated Planning

Background and Purpose

Literature is non-conclusive regarding selection of beam configurations in radiotherapy for mediastinal lymphoma (ML) radiotherapy, and published studies are based on manual planning with its inherent limitations. In this study, coplanar and non-coplanar beam configurations were systematically compared, using a large number of automatically generated plans.

Material and Methods

An autoplanning workflow, including beam configuration optimization, was configured for young female ML patients. For each of 25 patients, 24 plans with different beam configurations were generated with autoplanning: 11 coplanar CP_x plans and 11 non-coplanar NCP_x plans with x = 5 to 15 IMRT beams with computer-optimized, patient-specific configurations, and the coplanar VMAT and non-coplanar Butterfly VMAT (B-VMAT) beam angle class solutions (600 plans in total).

Results

Autoplans compared favorably with manually generated, clinically delivered plans, ensuring that beam configuration comparisons were performed with high quality plans. There was no beam configuration approach that was best for all patients and all plan parameters. Overall there was a clear tendency towards higher plan quality with non-coplanar configurations (NCP_x≥12 and B-VMAT). NCP_x≥12 produced highly conformal plans with on average reduced high doses in lungs and patient and also a reduced heart Dmean, while B-VMAT resulted in reduced low-dose spread in lungs and left breast.

Conclusions

Non-coplanar beam configurations were favorable for young female mediastinal lymphoma patients, with patient-specific and plan-parameter-dependent dosimetric advantages of NCP_x≥12 and B-VMAT. Individualization of beam configuration approach, considering also the faster delivery of B-VMAT vs. NCP_x≥12, can importantly improve the treatments.

Knowledge Models as Teaching Aid for Training Intensity Modulated Radiation Therapy Planning: A Lung Cancer Case Study

Purpose: Artificial intelligence (AI) employs knowledge models that often behave as a black-box to the majority of users and are not designed to improve the skill level of users. In this study, we aim to demonstrate the feasibility that AI can serve as an effective teaching aid to train individuals to develop optimal intensity modulated radiation therapy (IMRT) plans.

Methods and Materials: The training program is composed of a host of training cases and a tutoring system that consists of a front-end visualization module powered by knowledge models and a scoring system. The current tutoring system includes a beam angle prediction model and a dose-volume histogram (DVH) prediction model. The scoring system consists of physician chosen criteria for clinical plan evaluation as well as specially designed criteria for learning guidance. The training program includes six lung/mediastinum IMRT patients: one benchmark case and five training cases. A plan for the benchmark case is completed by each trainee entirely independently pre- and post-training. Five training cases cover a wide spectrum of complexity from easy (2), intermediate (1) to hard (2). Five trainees completed the training program with the help of one trainer. Plans designed by the trainees were evaluated by both the scoring system and a radiation oncologist to quantify planning quality.

Results: For the benchmark case, trainees scored an average of 21.6% of the total max points pre-training and improved to an average of 51.8% post-training. In comparison, the benchmark case's clinical plans score an average of 54.1% of the total max points. Two of the five trainees' post-training plans on the benchmark case were rated as comparable to the clinically delivered plans by the physician and all five were noticeably improved by the physician's standards. The total training time for each trainee ranged between 9 and 12 h.

Conclusion: This first attempt at a knowledge model based training program brought unexperienced planners to a level close to experienced planners in fewer than 2 days. The proposed tutoring system can serve as an important component in an AI ecosystem that will enable clinical practitioners to effectively and confidently use KBP.

Automated Intensity Modulated Radiation Therapy Treatment Planning for Cervical Cancer Based on Convolution Neural Network

Purpose:

To develop and evaluate an automatic intensity-modulated radiation therapy (IMRT) program for cervical cancer, including a Convolution Neural Network (CNN)-based prediction model and an automated optimization strategy.

Methods:

A CNN deep learning model was trained to predict a patient-specify set of IMRT objectives based on overlap volume histograms (OVH) and high-quality plan of previous patients. A total of 140 cervical cancer patients were enrolled in this study, including 100 patients in the training set, 20 patients in the validation set and 20 patients in the testing set. The input of this model was OVH data and the output were values of IMRT plan objectives. For patients in the testing set, the set of planning objectives were predicted by the CNN model and used to automatically generate IMRT plans. Meanwhile, manual plans of these patients were generated by 1 beginner planner and 1 senior planner respectively. Finally, dose distribution, dosimetric parameters and planning time were analyzed. In addition, the 3 types of plans were blinded compared and ranked by an experienced oncologist.

Results:

There were almost no statistically differences among these 3 types of plans in target coverage and dose conformity. Dose homogeneity were slightly decreased while the average dose and parameters for most organs-at-risk (OARs) were decreased in automatic plans. Especially in comparison with manual plans by the beginner planner, V₄₀ of bladder and rectum decreased 6.3% and 12.3%, while mean dose of rectum and marrow were 1.1 Gy and 1.8 Gy lower with automatic plans (either P < 0.017). In the blinded comparison, automatic plans were chosen as best plan in 14 cases.

Conclusions:

For cervical cancer, automatic IMRT plans optimized from the CNN generated objectives have superior dose sparing without compromising of target dose. It significantly reduced the planning time.

Multicriterial CNN based beam generation for robotic radiosurgery of the prostate

Although robotic radiosurgery offers a flexible arrangement of treatment beams, generating treatment plans is computationally challenging and a time consuming process for the planner. Furthermore, different clinical goals have to be considered during planning and generally different sets of beams correspond to different clinical goals. Typically, candidate beams sampled from a randomized heuristic form the basis for treatment planning. We propose a new approach to generate candidate beams based on deep learning using radiological features as well as the desired constraints. We demonstrate that candidate beams generated for specific clinical goals can improve treatment plan quality. Furthermore, we compare two approaches to include information about constraints in the prediction. Our results show that CNN generated beams can improve treatment plan quality for different clinical goals, increasing coverage from 91.2 to 96.8% for 3,000 candidate beams on average. When including the clinical goal in the training, coverage is improved by 1.1% points.

Feasibility and analysis of CNN-based candidate beam generation for robotic radiosurgery

PURPOSE

Robotic radiosurgery offers the flexibility of a robotic arm to enable high conformity to the target and a steep dose gradient. However, treatment planning becomes a computationally challenging task as the search space for potential beam directions for dose delivery is arbitrarily large. We propose an approach based on deep learning to improve the search for treatment beams.

METHODS

In clinical practice, a set of candidate beams generated by a randomized heuristic forms the basis for treatment planning. We use a convolutional neural network to identify promising candidate beams. Using radiological features of the patient, we predict the influence of a candidate beam on the delivered dose individually and let this prediction guide the selection of candidate beams. Features are represented as projections of the organ structures which are relevant during planning. Solutions to the inverse planning problem are generated for random and CNN-predicted candidate beams.

RESULTS

The coverage increases from 95.35% to 97.67% for 6000 heuristically and CNN-generated candidate beams, respectively. Conversely, a similar coverage can be achieved for treatment plans with half the number of candidate beams. This results in a patient-dependent reduced averaged computation time of 20.28%-45.69%. The number of active treatment beams can be reduced by 11.35% on average, which reduces treatment time. Constraining the maximum number of candidate beams per beam node can further improve the average coverage by 0.75 percentage points for 6000 candidate beams.

CONCLUSIONS

We show that deep learning based on radiological features can substantially improve treatment plan quality, reduce computation runtime, and treatment time compared to the heuristic approach used in clinics.

Artificial Intelligence in Radiotherapy Treatment Planning: Present and Future

Treatment planning is an essential step of the radiotherapy workflow. It has become more sophisticated over the past couple of decades with the help of computer science, enabling planners to design highly complex radiotherapy plans to minimize the normal tissue damage while persevering sufficient tumor control. As a result, treatment planning has become more labor intensive, requiring hours or even days of planner effort to optimize an individual patient case in a trial-and-error fashion. More recently, artificial intelligence has been utilized to automate and improve various aspects of medical science. For radiotherapy treatment planning, many algorithms have been developed to better support planners. These algorithms focus on automating the planning process and/or optimizing dosimetric trade-offs, and they have already made great impact on improving treatment planning efficiency and plan quality consistency. In this review, the smart planning tools in current clinical use are summarized in 3 main categories: automated rule implementation and reasoning, modeling of prior knowledge in clinical practice, and multicriteria optimization. Novel artificial intelligence-based treatment planning applications, such as deep learning-based algorithms and emerging research directions, are also reviewed. Finally, the challenges of artificial intelligence-based treatment planning are discussed for future works.

Treatment planning optimization with beam motion modeling for dynamic arc delivery of SBRT using Cyberknife with multileaf collimation

PURPOSE

The use of dynamic arcs for delivery of stereotactic body radiation therapy (SBRT) on Cyberknife is investigated, with a view to improving treatment times. This study investigates the required modeling of robot and multileaf collimator (MLC) motion between control points in the trajectory and then uses this to develop an optimization method for treatment planning of a dynamic arc with Cyberknife. The resulting plans are compared in terms of dose-volume histograms and estimated treatment times with those produced by a conventional beam arrangement.

METHODS

Five SBRT patient cases (prostate A - conventional, prostate B - brachytherapy-type, lung, liver, and partial left breast) were retrospectively studied. A suitable arc trajectory with control points spaced at 5° was proposed and treatment plans were produced for typical clinical protocols. The optimization consisted of a fluence optimization, segmentation, and direct aperture optimization using a gradient descent method. Dose delivered by the moving MLC was either taken to be the dose delivered discretely at the control points or modeled using effective fluence delivered between control points. The accuracy of calculated dose was assessed by recalculating after optimization using five interpolated beams and 100 interpolated apertures between each optimization control point. The resulting plans were compared using dose-volume histograms and estimated treatment times with those for a conventional Cyberknife beam arrangement.

RESULTS

If optimization is performed based on discrete doses delivered at the arc control points, large differences of up to 40% of the prescribed dose are seen when recalculating with interpolation. When the effective fluence between control points is taken into account during optimization, dosimetric differences are <2% for most structures when the plans are recalculated using intermediate nodes, but there are differences of up to 15% peripherally. Treatment plan quality is comparable between the arc trajectory and conventional body path. All plans meet the relevant clinical goals, with the exception of specific structures which overlap with the planning target volume. Median estimated treatment time is 355 s (range 235-672 s) for arc delivery and 675 s (range 554-1025 s) for conventional delivery.

CONCLUSIONS

The method of using effective fluence to model MLC motion between control points is sufficiently accurate to provide for accurate inverse planning of dynamic arcs with Cyberknife. The proposed arcing method produces treatment plans with comparable quality to the body path, with reduced estimated treatment delivery time.

Knowledge-based planning for intensity-modulated radiation therapy: A review of data-driven approaches

Purpose

Intensity‐Modulated Radiation Therapy (IMRT), including its variations (including IMRT, Volumetric Arc Therapy (VMAT), and Tomotherapy), is a widely used and critically important technology for cancer treatment. It is a knowledge‐intensive technology due not only to its own technical complexity, but also to the inherently conflicting nature of maximizing tumor control while minimizing normal organ damage. As IMRT experience and especially the carefully designed clinical plan data are accumulated during the past two decades, a new set of methods commonly termed knowledge‐based planning (KBP) have been developed that aim to improve the quality and efficiency of IMRT planning by learning from the database of past clinical plans. Some of this development has led to commercial products recently that allowed the investigation of KBP in numerous clinical applications. In this literature review, we will attempt to present a summary of published methods of knowledge‐based approaches in IMRT and recent clinical validation results.

Methods

In March 2018, a literature search was conducted in the NIH Medline database using the PubMed interface to identify publications that describe methods and validations related to KBP in IMRT including variations such as VMAT and Tomotherapy. The search criteria were designed to have a broad scope to capture relevant results with high sensitivity. The authors filtered down the search results according to a predefined selection criteria by reviewing the titles and abstracts first and then by reviewing the full text. A few papers were added to the list based on the references of the reviewed papers. The final set of papers was reviewed and summarized here.

Results

The initial search yielded a total of 740 articles. A careful review of the titles, abstracts, and eventually the full text and then adding relevant articles from reviewing the references resulted in a final list of 73 articles published between 2011 and early 2018. These articles described methods for developing knowledge models for predicting such parameters as dosimetric and dose‐volume points, voxel‐level doses, and objective function weights that improve or automate IMRT planning for various cancer sites, addressing different clinical and quality assurance needs, and using a variety of machine learning approaches. A number of articles reported carefully designed clinical studies that assessed the performance of KBP models in realistic clinical applications. Overwhelming majority of the studies demonstrated the benefits of KBP in achieving comparable and often improved quality of IMRT planning while reducing planning time and plan quality variation.

Conclusions

The number of KBP‐related studies has been steadily increasing since 2011 indicating a growing interest in applying this approach to clinical applications. Validation studies have generally shown KBP to produce plans with quality comparable to expert planners while reducing the time and efforts to generate plans. However, current studies are mostly retrospective and leverage relatively small datasets. Larger datasets collected through multi‐institutional collaboration will enable the development of more advanced models to further improve the performance of KBP in complex clinical cases. Prospective studies will be an important next step toward widespread adoption of this exciting technology.

Beam selection for stereotactic ablative radiotherapy using Cyberknife with multileaf collimation

The Cyberknife system (Accuray Inc., Sunnyvale, CA) enables radiotherapy using stereotactic ablative body radiotherapy (SABR) with a large number of non-coplanar beam orientations. Recently, a multileaf collimator has also been available to allow flexibility in field shaping. This work aims to evaluate the quality of treatment plans obtainable with the multileaf collimator. Specifically, the aim is to find a subset of beam orientations from a predetermined set of candidate directions, such that the treatment quality is maintained but the treatment time is reduced. An evolutionary algorithm is used to successively refine a randomly selected starting set of beam orientations. By using an efficient computational framework, clinically useful solutions can be found in several hours. It is found that 15 beam orientations are able to provide treatment quality which approaches that of the candidate beam set of 110 beam orientations, but with approximately half of the estimated treatment time. Choice of an efficient subset of beam orientations offers the possibility to improve the patient experience and maximise the number of patients treated.

Incorporating Case-Based Reasoning for Radiation Therapy Knowledge Modeling: A Pelvic Case Study

Knowledge models in radiotherapy capture the relation between patient anatomy and dosimetry to provide treatment planning guidance. When treatment schemes evolve, existing models struggle to predict accurately. We propose a case-based reasoning framework designed to handle novel anatomies that are of same type but vary beyond original training samples. A total of 105 pelvic intensity-modulated radiotherapy cases were analyzed. Eighty cases were prostate cases while the other 25 were prostate-plus-lymph-node cases. We simulated 4 scenarios: Scarce scenario, Semiscarce scenario, Semiample scenario, and Ample scenario. For the Scarce scenario, a multiple stepwise regression model was trained using 85 cases (80 prostate, 5 prostate-plus-lymph-node). The proposed workflow started with evaluating the feature novelty of new cases against 5 training prostate-plus-lymph-node cases using leverage statistic. The case database was composed of a 5-case dose atlas. Case-based dose prediction was compared against the regression model prediction using sum of squared residual. Mean sum of squared residual of case-based and regression predictions for the bladder of 13 identified outliers were 0.174 ± 0.166 and 0.459 ± 0.508, respectively (P = .0326). For the rectum, the respective mean sum of squared residuals were 0.103 ± 0.120 and 0.150 ± 0.171 for case-based and regression prediction (P = .1972). By retaining novel cases, under the Ample scenario, significant statistical improvement was observed over the Scarce scenario (P = .0398) for the bladder model. We expect that the incorporation of case-based reasoning that judiciously applies appropriate predictive models could improve overall prediction accuracy and robustness in clinical practice.

A clustering Bayesian approach for multivariate non-ordered circular data

This article presents a Bayesian model for the clustering of non-ordered multivariate directional or circular data. The particular trait of our data is that each single observation is made up of [Formula: see text] non-ordered points on the circle. We introduce a hierarchical model that combines a symmetrization technique, projected normal distributions and a Dirichlet process. One parameter is introduced to model the non-ordered trait and another one to control the variability of the angles on the circle. An informative prior on the relative locations of the [Formula: see text] angles is also provided. The gain of the symmetrization is highlighted by a theoretical study. The parameters of the model are then inferred using a Metropolis–Hastings-within-Gibbs algorithm. Simulated datasets are analysed to study the sensitivity to hyperparameters. Then, the benefits of our approach are illustrated by clustering real data made up of the positions of five separate radiotherapy X-ray beams on a circle.

Lung IMRT planning with automatic determination of beam angle configurations

Beam angle configuration is a major planning decision in intensity modulated radiation treatment (IMRT) that has a significant impact on dose distributions and thus quality of treatment, especially in complex planning cases such as those for lung cancer treatment. We propose a novel method to automatically determine beam configurations that incorporates noncoplanar beams. We then present a completely automated IMRT planning algorithm that combines the proposed method with a previously reported OAR DVH prediction model. Finally, we validate this completely automatic planning algorithm using a set of challenging lung IMRT cases. A beam efficiency index map is constructed to guide the selection of beam angles. This index takes into account both the dose contributions from individual beams and the combined effect of multiple beams by introducing a beam-spread term. The effect of the beam-spread term on plan quality was studied systematically and the weight of the term to balance PTV dose conformity against OAR avoidance was determined. For validation, complex lung cases with clinical IMRT plans that required the use of one or more noncoplanar beams were re-planned with the proposed automatic planning algorithm. Important dose metrics for the PTV and OARs in the automatic plans were compared with those of the clinical plans. The results are very encouraging. The PTV dose conformity and homogeneity in the automatic plans improved significantly. And all the dose metrics of the automatic plans, except the lung V_{5 Gy}, were statistically better than or comparable with those of the clinical plans. In conclusion, the automatic planning algorithm can incorporate non-coplanar beam configurations in challenging lung cases and can generate plans efficiently with quality closely approximating that of clinical plans.

Creation of knowledge-based planning models intended for large scale distribution: Minimizing the effect of outlier plans

Knowledge‐based planning (KBP) can be used to estimate dose–volume histograms (DVHs) of organs at risk (OAR) using models. The task of model creation, however, can result in estimates with differing accuracy; particularly when outlier plans are not properly addressed. This work used RapidPlan^™ to create models for the prostate and head and neck intended for large‐scale distribution. Potential outlier plans were identified by means of regression analysis scatter plots, Cook's distance, coefficient of determination, and the chi‐squared test. Outlier plans were identified as falling into three categories: geometric, dosimetric, and over‐fitting outliers. The models were validated by comparing DVHs estimated by the model with those from a separate and independent set of clinical plans. The estimated DVHs were also used as optimization objectives during inverse planning. The analysis tools lead us to identify as many as 7 geometric, 8 dosimetric, and 20 over‐fitting outliers in the raw models. Geometric and over‐fitting outliers were removed while the dosimetric outliers were replaced after re‐planning. Model validation was done by comparing the DVHs at 50%, 85%, and 99% of the maximum dose for each OAR (denoted as V50, V85, and V99) and agreed within −2% to 4% for the three metrics for the final prostate model. In terms of the head and neck model, the estimated DVHs agreed from −2.0% to 5.1% at V50, 0.1% to 7.1% at V85, and 0.1% to 7.6% at V99. The process used to create these models improved the accuracy for the pharyngeal constrictor DVH estimation where one plan was originally over‐estimated by more than twice. In conclusion, our results demonstrate that KBP models should be carefully created since their accuracy could be negatively affected by outlier plans. Outlier plans can be addressed by removing them from the model and re‐planning.

Automated inverse optimization facilitates lower doses to normal tissue in pancreatic stereotactic body radiotherapy

Purpose

Inverse planning is trial-and-error iterative process. This work introduces a fully automated inverse optimization approach, where the treatment plan is closely tailored to the unique patient anatomy. The auto-optimization is applied to pancreatic stereotactic body radiotherapy (SBRT).

Materials and methods

The automation is based on stepwise reduction of dose-volume histograms (DVHs). Five uniformly spaced points, from 1% to 70% of the organ at risk (OAR) volumes, are used. Doses to those DVH points are iteratively decreased through multiple optimization runs. With each optimization run the doses to the OARs are decreased, while the dose homogeneity over the target is increased. The iterative process is terminated when a pre-specified dose heterogeneity over the target is reached. Twelve pancreatic cases were retrospectively studied. Doses to the target, maximum doses to duodenum, bowel, stomach, and spinal cord were evaluated. In addition, mean doses to liver and kidneys were tallied. The auto-optimized plans were compared to the actual treatment plans, which are based on national protocols.

Results

The prescription dose to 95% of the planning target volume (PTV) is the same for the treatment and the auto-optimized plans. The average difference for maximum doses to duodenum, bowel, stomach, and spinal cord are -4.6 Gy, -1.8 Gy, -1.6 Gy, and -2.4 Gy respectively. The negative sign indicates lower doses with the auto-optimization. The average differences in the mean doses to liver and kidneys are -0.6 Gy, and -1.1 Gy to -1.5 Gy respectively.

Conclusions

Automated inverse optimization holds great potential for personalization and tailoring of radiotherapy to particular patient anatomies. It can be utilized for normal tissue sparing or for an isotoxic dose escalation.

Outlier identification in radiation therapy knowledge-based planning: A study of pelvic cases

PURPOSE

The purpose of this study was to apply statistical metrics to identify outliers and to investigate the impact of outliers on knowledge-based planning in radiation therapy of pelvic cases. We also aimed to develop a systematic workflow for identifying and analyzing geometric and dosimetric outliers.

METHODS

Four groups (G1-G4) of pelvic plans were sampled in this study. These include the following three groups of clinical IMRT cases: G1 (37 prostate cases), G2 (37 prostate plus lymph node cases) and G3 (37 prostate bed cases). Cases in G4 were planned in accordance with dynamic-arc radiation therapy procedure and include 10 prostate cases in addition to those from G1. The workflow was separated into two parts: 1. identifying geometric outliers, assessing outlier impact, and outlier cleaning; 2. identifying dosimetric outliers, assessing outlier impact, and outlier cleaning. G2 and G3 were used to analyze the effects of geometric outliers (first experiment outlined below) while G1 and G4 were used to analyze the effects of dosimetric outliers (second experiment outlined below). A baseline model was trained by regarding all G2 cases as inliers. G3 cases were then individually added to the baseline model as geometric outliers. The impact on the model was assessed by comparing leverages of inliers (G2) and outliers (G3). A receiver-operating-characteristic (ROC) analysis was performed to determine the optimal threshold. The experiment was repeated by training the baseline model with all G3 cases as inliers and perturbing the model with G2 cases as outliers. A separate baseline model was trained with 32 G1 cases. Each G4 case (dosimetric outlier) was subsequently added to perturb the model. Predictions of dose-volume histograms (DVHs) were made using these perturbed models for the remaining 5 G1 cases. A Weighted Sum of Absolute Residuals (WSAR) was used to evaluate the impact of the dosimetric outliers.

RESULTS

The leverage of inliers and outliers was significantly different. The Area-Under-Curve (AUC) for differentiating G2 (outliers) from G3 (inliers) was 0.98 (threshold: 0.27) for the bladder and 0.81 (threshold: 0.11) for the rectum. For differentiating G3 (outlier) from G2 (inlier), the AUC (threshold) was 0.86 (0.11) for the bladder and 0.71 (0.11) for the rectum. Significant increase in WSAR was observed in the model with 3 dosimetric outliers for the bladder (P < 0.005 with Bonferroni correction), and in the model with only 1 dosimetric outlier for the rectum (P < 0.005).

CONCLUSIONS

We established a systematic workflow for identifying and analyzing geometric and dosimetric outliers, and investigated statistical metrics for outlier detection. Results validated the necessity for outlier detection and clean-up to enhance model quality in clinical practice.