Evaluating automatically generated normal tissue contours for safe use in head and neck and cervical cancer treatment planning.

PURPOSE: Volumetric-modulated arc therapy (VMAT) is a widely accepted treatment method for head and neck (HN) and cervical cancers; however, creating contours and plan optimization for VMAT plans is a time-consuming process. Our group has created an automated treatment planning tool, the Radiation Planning Assistant (RPA), that uses deep learning models to generate organs at risk (OARs), planning structures and automates plan optimization. This study quantitatively evaluates the quality of contours generated by the RPA tool. METHODS: For patients with HN (54) and cervical (39) cancers, we retrospectively generated autoplans using the RPA. Autoplans were generated using deep-learning and RapidPlan models developed in-house. The autoplans were, then, applied to the original, physician-drawn contours, which were used as a ground truth (GT) to compare with the autocontours (RPA). Using a "two one-sided tests" (TOST) procedure, we evaluated whether the autocontour normal tissue dose was equivalent to that of the ground truth by a margin, δ, that we determined based on clinical judgement. We also calculated the number of plans that met established clinically accepted dosimetric criteria. RESULTS: For HN plans, 91.8% and 91.7% of structures met dosimetric criteria for automatic and manual contours, respectively; for cervical plans, 95.6% and 95.7% of structures met dosimetric criteria for automatic and manual contours, respectively. Autocontours were equivalent to the ground truth for 71% and 75% of common DVH metrics for the HN and cervix, respectively. CONCLUSIONS: This study shows that dosimetrically equivalent normal tissue contours can be created for HN and cervical cancers using deep learning techniques. In general, differences between the contours did not affect the passing or failing of clinical dose tolerances.

Automatic end-to-end VMAT treatment planning for rectal cancers.

BACKGROUND: The treatment planning process from segmentation to producing a deliverable plan is time-consuming and labor-intensive. Existing solutions automate the segmentation and planning processes individually. The feasibility of combining auto-segmentation and auto-planning for volumetric modulated arc therapy (VMAT) for rectal cancers in an end-to-end process is not clear. PURPOSE: To create and clinically evaluate a complete end-to-end process for auto-segmentation and auto-planning of VMAT for rectal cancer requiring only the gross tumor volume contour and a CT scan as inputs. METHODS: Patient scans and data were retrospectively selected from our institutional records for patients treated for malignant neoplasm of the rectum. We trained, validated, and tested deep learning auto-segmentation models using nnU-Net architecture for clinical target volume (CTV), bowel bag, large bowel, small bowel, total bowel, femurs, bladder, bone marrow, and female and male genitalia. For the CTV, we identified 174 patients with clinically drawn CTVs. We used data for 18 patients for all structures other than the CTV. The structures were contoured under the guidance of and reviewed by a gastrointestinal (GI) radiation oncologist. The predicted results for CTV in 35 patients and organs at risk (OAR) in six patients were scored by the GI radiation oncologist using a five-point Likert scale. For auto-planning, a RapidPlan knowledge-based planning solution was modeled for VMAT delivery with a prescription of 25 Gy in five fractions. The model was trained and tested on 20 and 34 patients, respectively. The resulting plans were scored by two GI radiation oncologists using a five-point Likert scale. Finally, the end-to-end pipeline was evaluated on 16 patients, and the resulting plans were scored by two GI radiation oncologists. RESULTS: In 31 of 35 patients, CTV contours were clinically acceptable without necessary modifications. The CTV achieved a Dice similarity coefficient of 0.85 (±0.05) and 95% Hausdorff distance of 15.25 (±5.59) mm. All OAR contours were clinically acceptable without edits, except for large and small bowel which were challenging to differentiate. However, contours for total, large, and small bowel were clinically acceptable. The two physicians accepted 100% and 91% of the auto-plans. For the end-to-end pipeline, the two physicians accepted 88% and 62% of the auto-plans. CONCLUSIONS: This study demonstrated that the VMAT treatment planning technique for rectal cancer can be automated to generate clinically acceptable and safe plans with minimal human interventions.

Automating the treatment planning process for 3D-conformal pediatric craniospinal irradiation therapy.

PURPOSE: Pediatric patients with medulloblastoma in low- and middle-income countries (LMICs) are most treated with 3D-conformal photon craniospinal irradiation (CSI), a time-consuming, complex treatment to plan, especially in resource-constrained settings. Therefore, we developed and tested a 3D-conformal CSI autoplanning tool for varying patient lengths. METHODS AND MATERIALS: Autocontours were generated with a deep learning model trained:tested (80:20 ratio) on 143 pediatric medulloblastoma CT scans (patient ages: 2-19 years, median = 7 years). Using the verified autocontours, the autoplanning tool generated two lateral brain fields matched to a single spine field, an extended single spine field, or two matched spine fields. Additional spine subfields were added to optimize the corresponding dose distribution. Feathering was implemented (yielding nine to 12 fields) to give a composite plan. Each planning approach was tested on six patients (ages 3-10 years). A pediatric radiation oncologist assessed clinical acceptability of each autoplan. RESULTS: The autocontoured structures' average Dice similarity coefficient ranged from .65 to .98. The average V95 for the brain/spinal canal for single, extended, and multi-field spine configurations was 99.9% ± 0.06%/99.9% ± 0.10%, 99.9% ± 0.07%/99.4% ± 0.30%, and 99.9% ± 0.06%/99.4% ± 0.40%, respectively. The average maximum dose across all field configurations to the brainstem, eyes (L/R), lenses (L/R), and spinal cord were 23.7 ± 0.08, 24.1 ± 0.28, 13.3 ± 5.27, and 25.5 ± 0.34 Gy, respectively (prescription = 23.4 Gy/13 fractions). Of the 18 plans tested, all were scored as clinically acceptable as-is or clinically acceptable with minor, time-efficient edits preferred or required. No plans were scored as clinically unacceptable. CONCLUSION: The autoplanning tool successfully generated pediatric CSI plans for varying patient lengths in 3.50 ± 0.4 minutes on average, indicating potential for an efficient planning aid in a resource-constrained settings.

Automated field-in-field whole brain radiotherapy planning.

PURPOSE: We developed and tested an automatic field-in-field (FIF) solution for whole-brain radiotherapy (WBRT) planning that creates a homogeneous dose distribution by minimizing hotspots, resulting in clinically acceptable plans. METHODS: A configurable auto-planning algorithm was developed to automatically generate FIF WBRT plans independent of the treatment planning system. Configurable parameters include the definition of hotspots, target volume, maximum number of subfields, and minimum number of monitor units per field. This algorithm iteratively identifies a hotspot, creates two opposing subfields, calculates the dose, and optimizes the beam weight based on user-configured constraints of dose-volume histogram coverage and least-squared cost functions. The algorithm was retrospectively tested on 17 whole-brain patients. First, an in-house landmark-based automated beam aperture technique was used to generate the treatment fields and initial plans. Second, the FIF algorithm was employed to optimize the plans using physician-defined goals of 99.9% of the brain volume receiving 100% of the prescription dose (30 Gy in 10 fractions) and a target hotspot definition of 107% of the prescription dose. The final auto-optimized plans were assessed for clinical acceptability by an experienced radiation oncologist using a five-point scale. RESULTS: The FIF algorithm reduced the mean (± SD) plan hotspot percentage dose from 35.0 Gy (116.6%) ± 0.6 Gy (2.0%) to 32.6 Gy (108.8%) ± 0.4 Gy (1.2%). Also, it decreased the mean (± SD) hotspot V107% [cm3 ] from 959 ± 498 cm3 to 145 ± 224 cm3 . On average, plans were produced in 16 min without any user intervention. Furthermore, 76.5% of the auto-plans were clinically acceptable (needing no or minor stylistic edits), and all of them were clinically acceptable after minor clinically necessary edits. CONCLUSIONS: This algorithm successfully produced high-quality WBRT plans and can improve treatment planning efficiency when incorporated into an automatic planning workflow.

Parametric delineation uncertainties contouring (PDUC) modeling on CT scans of prostate cancer patients.

PURPOSE: Variability in contouring contributes to large variations in radiation therapy planning and treatment outcomes. The development and testing of tools to automatically detect contouring errors require a source of contours that includes well-understood and realistic errors. The purpose of this work was to develop a simulation algorithm that intentionally injects errors of varying magnitudes into clinically accepted contours and produces realistic contours with different levels of variability. METHODS: We used a dataset of CT scans from 14 prostate cancer patients with clinician-drawn contours of the regions of interest (ROI) of the prostate, bladder, and rectum. Using our newly developed Parametric Delineation Uncertainties Contouring (PDUC) model, we automatically generated alternative, realistic contours. The PDUC model consists of the contrast-based DU generator and a 3D smoothing layer. The DU generator transforms contours (deformation, contraction, and/or expansion) as a function of image contrast. The generated contours undergo 3D smoothing to obtain a realistic look. After model building, the first batch of auto-generated contours was reviewed. Editing feedback from the reviews was then used in a filtering model for the auto-selection of clinically acceptable (minor-editing) DU contours. RESULTS: Overall, C values of 5 and 50 consistently produced high proportions of minor-editing contours across all ROI compared to the other C values (0.936 ± $ \pm \;$ 0.111 and 0.552 ± $ \pm \;$ 0.228, respectively). The model performed best on the bladder, which had the highest proportion of minor-editing contours (0.606) of the three ROI. In addition, the classification AUC for the filtering model across all three ROI is 0.724 ± $ \pm \;$ 0.109. DISCUSSION: The proposed methodology and subsequent results are promising and could have a great impact on treatment planning by generating mathematically simulated alternative structures that are clinically relevant and realistic enough (i.e., similar to clinician-drawn contours) to be used in quality control of radiation therapy.

Hazard testing to reduce risk in the development of automated planning tools.

PURPOSE: Hazard scenarios were created to assess and reduce the risk of planning errors in automated planning processes. This was accomplished through iterative testing and improvement of examined user interfaces. METHODS: Automated planning requires three user inputs: a computed tomography (CT), a prescription document, known as the service request, and contours. We investigated the ability of users to catch errors that were intentionally introduced into each of these three stages, according to an FMEA analysis. Five radiation therapists each reviewed 15 patient CTs, containing three errors: inappropriate field of view, incorrect superior border, and incorrect identification of isocenter. Four radiation oncology residents reviewed 10 service requests, containing two errors: incorrect prescription and treatment site. Four physicists reviewed 10 contour sets, containing two errors: missing contour slices and inaccurate target contour. Reviewers underwent video training prior to reviewing and providing feedback for various mock plans. RESULTS: Initially, 75% of hazard scenarios were detected in the service request approval. The visual display of prescription information was then updated to improve the detectability of errors based on user feedback. The change was then validated with five new radiation oncology residents who detected 100% of errors present. 83% of the hazard scenarios were detected in the CT approval portion of the workflow. For the contour approval portion of the workflow none of the errors were detected by physicists, indicating this step will not be used for quality assurance of contours. To mitigate the risk from errors that could occur at this step, radiation oncologists must perform a thorough review of contour quality prior to final plan approval. CONCLUSIONS: Hazard testing was used to pinpoint the weaknesses of an automated planning tool and as a result, subsequent improvements were made. This study identified that not all workflow steps should be used for quality assurance and demonstrated the importance of performing hazard testing to identify points of risk in automated planning tools.

Customizable landmark-based field aperture design for automated whole-brain radiotherapy treatment planning.

PURPOSE: To develop and evaluate an automated whole-brain radiotherapy (WBRT) treatment planning pipeline with a deep learning-based auto-contouring and customizable landmark-based field aperture design. METHODS: The pipeline consisted of the following steps: (1) Auto-contour normal structures on computed tomography scans and digitally reconstructed radiographs using deep learning techniques, (2) locate the landmark structures using the beam's-eye-view, (3) generate field apertures based on eight different landmark rules addressing different clinical purposes and physician preferences. Two parallel approaches for generating field apertures were developed for quality control. The performance of the generated field shapes and dose distributions were compared with the original clinical plans. The clinical acceptability of the plans was assessed by five radiation oncologists from four hospitals. RESULTS: The performance of the generated field apertures was evaluated by the Hausdorff distance (HD) and mean surface distance (MSD) from 182 patients' field apertures used in the clinic. The average HD and MSD for the generated field apertures were 16 ± 7 and 7 ± 3 mm for the first approach, respectively, and 17 ± 7 and 7 ± 3 mm, respectively, for the second approach. The differences regarding HD and MSD between the first and the second approaches were 1 ± 2 and 1 ± 3 mm, respectively. A clinical review of the field aperture design, conducted using 30 patients, achieved a 100% acceptance rate for both the first and second approaches, and the plan review achieved a 100% acceptance rate for the first approach and a 93% acceptance rate for the second approach. The average acceptance rate for meeting lens dosimetric recommendations was 80% (left lens) and 77% (right lens) for the first approach, and 70% (both left and right lenses) for the second approach, compared with 50% (left lens) and 53% (right lens) for the clinical plans. CONCLUSION: This study provided an automated pipeline with two field aperture generation approaches to automatically generate WBRT treatment plans. Both quantitative and qualitative evaluations demonstrated that our novel pipeline was comparable with the original clinical plans.

Resection cavity auto-contouring for patients with pediatric medulloblastoma using only CT information.

PURPOSE: Target delineation for radiation therapy is a time-consuming and complex task. Autocontouring gross tumor volumes (GTVs) has been shown to increase efficiency. However, there is limited literature on post-operative target delineation, particularly for CT-based studies. To this end, we trained a CT-based autocontouring model to contour the post-operative GTV of pediatric patients with medulloblastoma. METHODS: One hundred four retrospective pediatric CT scans were used to train a GTV auto-contouring model. Eighty patients were then preselected for contour visibility, continuity, and location to train an additional model. Each GTV was manually annotated with a visibility score based on the number of slices with a visible GTV (1 = < 25%, 2 = 25-50%, 3 = > 50-75%, and 4 = > 75-100%). Contrast and the contrast-to-noise ratio (CNR) were calculated for the GTV contour with respect to a cropped background image. Both models were tested on the original and pre-selected testing sets. The resulting surface and overlap metrics were calculated comparing the clinical and autocontoured GTVs and the corresponding clinical target volumes (CTVs). RESULTS: Eighty patients were pre-selected to have a continuous GTV within the posterior fossa. Of these, 7, 41, 21, and 11 were visibly scored as 4, 3, 2, and 1, respectively. The contrast and CNR removed an additional 11 and 20 patients from the dataset, respectively. The Dice similarity coefficients (DSC) were 0.61 ± 0.29 and 0.67 ± 0.22 on the models without pre-selected training data and 0.55 ± 13.01 and 0.83 ± 0.17 on the models with pre-selected data, respectively. The DSC on the CTV expansions were 0.90 ± 0.13. CONCLUSION: We successfully automatically contoured continuous GTVs within the posterior fossa on scans that had contrast > ± 10 HU. CT-Based auto-contouring algorithms have potential to positively impact centers with limited MRI access.

Identifying the optimal deep learning architecture and parameters for automatic beam aperture definition in 3D radiotherapy.

PURPOSE: Two-dimensional radiotherapy is often used to treat cervical cancer in low- and middle-income countries, but treatment planning can be challenging and time-consuming. Neural networks offer the potential to greatly decrease planning time through automation, but the impact of the wide range of hyperparameters to be set during training on model accuracy has not been exhaustively investigated. In the current study, we evaluated the effect of several convolutional neural network architectures and hyperparameters on 2D radiotherapy treatment field delineation. METHODS: Six commonly used deep learning architectures were trained to delineate four-field box apertures on digitally reconstructed radiographs for cervical cancer radiotherapy. A comprehensive search of optimal hyperparameters for all models was conducted by varying the initial learning rate, image normalization methods, and (when appropriate) convolutional kernel size, the number of learnable parameters via network depth and the number of feature maps per convolution, and nonlinear activation functions. This yielded over 1700 unique models, which were all trained until performance converged and then tested on a separate dataset. RESULTS: Of all hyperparameters, the choice of initial learning rate was most consistently significant for improved performance on the test set, with all top-performing models using learning rates of 0.0001. The optimal image normalization was not consistent across architectures. High overlap (mean Dice similarity coefficient = 0.98) and surface distance agreement (mean surface distance < 2 mm) were achieved between the treatment field apertures for all architectures using the identified best hyperparameters. Overlap Dice similarity coefficient (DSC) and distance metrics (mean surface distance and Hausdorff distance) indicated that DeepLabv3+ and D-LinkNet architectures were least sensitive to initial hyperparameter selection. CONCLUSION: DeepLabv3+ and D-LinkNet are most robust to initial hyperparameter selection. Learning rate, nonlinear activation function, and kernel size are also important hyperparameters for improving performance.

Assessing the practicality of using a single knowledge-based planning model for multiple linac vendors.

PURPOSE: Knowledge-based planning (KBP) has been shown to be an effective tool in quality control for intensity-modulated radiation therapy treatment planning and generating high-quality plans. Previous studies have evaluated its ability to create consistent plans across institutions and between planners within the same institution as well as its use as teaching tool for inexperienced planners. This study evaluates whether planning quality is consistent when using a KBP model to plan across different treatment machines. MATERIALS AND METHODS: This study used a RapidPlan model (Varian Medical Systems) provided by the vendor, to which we added additional planning objectives, maximum dose limits, and planning structures, such that a clinically acceptable plan is achieved in a single optimization. This model was used to generate and optimize volumetric-modulated arc therapy plans for a cohort of 50 patients treated for head-neck cancer. Plans were generated using the following treatment machines: Varian 2100, Elekta Versa HD, and Varian Halcyon. A noninferiority testing methodology was used to evaluate the hypothesis that normal and target metrics in our autoplans were no worse than a set of clinically-acceptable baseline plans by a margin of 1.8 Gy or 3% dose-volume. The quality of these plans were also compared through the use of common clinical dose-volume histogram criteria. RESULTS: The Versa HD met our noninferiority criteria for 23 of 34 normal and target metrics; while the Halcyon and Varian 2100 machines met our criteria for 24 of 34 and 26 of 34 metrics, respectively. The experimental plans tended to have less volume coverage for prescription dose planning target volume and larger hotspot volumes. However, comparable plans were generated across different treatment machines. CONCLUSIONS: These results support the use of a head-neck RapidPlan models in centralized planning workflows that support clinics with different linac models/vendors, although some fine-tuning for targets may be necessary.

Statistical process control to monitor use of a web-based autoplanning tool.

PURPOSE: To investigate the use of statistical process control (SPC) for quality assurance of an integrated web-based autoplanning tool, Radiation Planning Assistant (RPA). METHODS: Automatically generated plans were downloaded and imported into two treatment planning systems (TPSs), RayStation and Eclipse, in which they were recalculated using fixed monitor units. The recalculated plans were then uploaded back to the RPA, and the mean dose differences for each contour between the original RPA and the TPSs plans were calculated. SPC was used to characterize the RPA plans in terms of two comparisons: RayStation TPS versus RPA and Eclipse TPS versus RPA for three anatomical sites, and variations in the machine parameters dosimetric leaf gap (DLG) and multileaf collimator transmission factor (MLC-TF) for two algorithms (Analytical Anisotropic Algorithm [AAA]) and Acuros in the Eclipse TPS. Overall, SPC was used to monitor the process of the RPA, while clinics would still perform their routine patient-specific QA. RESULTS: For RayStation, the average mean percent dose differences across all contours were 0.65% ± 1.05%, -2.09% ± 0.56%, and 0.28% ± 0.98% and average control limit ranges were 1.89% ± 1.32%, 2.16% ± 1.31%, and 2.65% ± 1.89% for the head and neck, cervix, and chest wall, respectively. In contrast, Eclipse's average mean percent dose differences across all contours were -0.62% ± 0.34%, 0.32% ± 0.23%, and -0.91% ± 0.98%, while average control limit ranges were 1.09% ± 0.77%, 3.69% ± 2.67%, 2.73% ± 1.86%, respectively. Averaging all contours and removing outliers, a 0% dose difference corresponded with a DLG value of 0.202 ± 0.019 cm and MLC-TF value of 0.020 ± 0.001 for Acuros and a DLG value of 0.135 ± 0.031 cm and MLC-TF value of 0.015 ± 0.001 for AAA. CONCLUSIONS: Differences in mean dose and control limits between RPA and two separately commissioned TPSs were determined. With varying control limits and means, SPC provides a flexible and useful process quality assurance tool for monitoring a complex automated system such as the RPA.

Development and validation of a checklist for use with automatically generated radiotherapy plans.

PURPOSE: To develop a checklist that improves the rate of error detection during the plan review of automatically generated radiotherapy plans. METHODS: A custom checklist was developed using guidance from American Association of Physicists in Medicine task groups 275 and 315 and the results of a failure modes and effects analysis of the Radiation Planning Assistant (RPA), an automated contouring and treatment planning tool. The preliminary checklist contained 90 review items for each automatically generated plan. In the first study, eight physicists were recruited from our institution who were familiar with the RPA. Each physicist reviewed 10 artificial intelligence-generated resident treatment plans from the RPA for safety and plan quality, five of which contained errors. Physicists performed plan checks, recorded errors, and rated each plan's clinical acceptability. Following a 2-week break, physicists reviewed 10 additional plans with a similar distribution of errors using our customized checklist. Participants then provided feedback on the usability of the checklist and it was modified accordingly. In a second study, this process was repeated with 14 senior medical physics residents who were randomly assigned to checklist or no checklist for their reviews. Each reviewed 10 plans, five of which contained errors, and completed the corresponding survey. RESULTS: In the first study, the checklist significantly improved the rate of error detection from 3.4 ± 1.1 to 4.4 ± 0.74 errors per participant without and with the checklist, respectively (p = 0.02). Error detection increased by 20% when the custom checklist was utilized. In the second study, 2.9 ± 0.84 and 3.5 ± 0.84 errors per participant were detected without and with the revised checklist, respectively (p = 0.08). Despite the lack of statistical significance for this cohort, error detection increased by 18% when the checklist was utilized. CONCLUSION: Our results indicate that the use of a customized checklist when reviewing automated treatment plans will result in improved patient safety.

Knowledge-based planning for the radiation therapy treatment plan quality assurance for patients with head and neck cancer.

This study aimed to investigate the feasibility of using a knowledge-based planning technique to detect poor quality VMAT plans for patients with head and neck cancer. We created two dose-volume histogram (DVH) prediction models using a commercial knowledge-based planning system (RapidPlan, Varian Medical Systems, Palo Alto, CA) from plans generated by manual planning (MP) and automated planning (AP) approaches. DVHs were predicted for evaluation cohort 1 (EC1) of 25 patients and compared with achieved DVHs of MP and AP plans to evaluate prediction accuracy. Additionally, we predicted DVHs for evaluation cohort 2 (EC2) of 25 patients for which we intentionally generated plans with suboptimal normal tissue sparing while satisfying dose-volume limits of standard practice. Three radiation oncologists reviewed these plans without seeing the DVH predictions. We found that predicted DVH ranges (upper-lower predictions) were consistently wider for the MP model than for the AP model for all normal structures. The average ranges of mean dose predictions among all structures was 9.7 Gy (MP model) and 3.4 Gy (AP model) for EC1 patients. RapidPlan models identified 7 MP plans as outliers according to mean dose or D1% for at least one structure, while none of AP plans were flagged. For EC2 patients, 22 suboptimal plans were identified by prediction. While re-generated AP plans validated that these suboptimal plans could be improved, 40 out of 45 structures with predicted poor sparing were also identified by oncologist reviews as requiring additional planning to improve sparing in the clinical setting. Our study shows that knowledge-based DVH prediction models can be sufficiently accurate for plan quality assurance purposes. A prediction model built by a small cohort automatically-generated plans was effective in detecting suboptimal plans. Such tools have potential to assist the plan quality assurance workflow for individual patients in the clinic.

Automation of radiation treatment planning for rectal cancer.

PURPOSE: To develop an automated workflow for rectal cancer three-dimensional conformal radiotherapy (3DCRT) treatment planning that combines deep learning (DL) aperture predictions and forward-planning algorithms. METHODS: We designed an algorithm to automate the clinical workflow for 3DCRT planning with field aperture creations and field-in-field (FIF) planning. DL models (DeepLabV3+ architecture) were trained, validated, and tested on 555 patients to automatically generate aperture shapes for primary (posterior-anterior [PA] and opposed laterals) and boost fields. Network inputs were digitally reconstructed radiographs, gross tumor volume (GTV), and nodal GTV. A physician scored each aperture for 20 patients on a 5-point scale (>3 is acceptable). A planning algorithm was then developed to create a homogeneous dose using a combination of wedges and subfields. The algorithm iteratively identifies a hotspot volume, creates a subfield, calculates dose, and optimizes beam weight all without user intervention. The algorithm was tested on 20 patients using clinical apertures with varying wedge angles and definitions of hotspots, and the resulting plans were scored by a physician. The end-to-end workflow was tested and scored by a physician on another 39 patients. RESULTS: The predicted apertures had Dice scores of 0.95, 0.94, and 0.90 for PA, laterals, and boost fields, respectively. Overall, 100%, 95%, and 87.5% of the PA, laterals, and boost apertures were scored as clinically acceptable, respectively. At least one auto-plan was clinically acceptable for all patients. Wedged and non-wedged plans were clinically acceptable for 85% and 50% of patients, respectively. The hotspot dose percentage was reduced from 121% (σ = 14%) to 109% (σ = 5%) of prescription dose for all plans. The integrated end-to-end workflow of automatically generated apertures and optimized FIF planning gave clinically acceptable plans for 38/39 (97%) of patients. CONCLUSION: We have successfully automated the clinical workflow for generating radiotherapy plans for rectal cancer for our institution.

Automatic contouring QA method using a deep learning-based autocontouring system.

PURPOSE: To determine the most accurate similarity metric when using an independent system to verify automatically generated contours. METHODS: A reference autocontouring system (primary system to create clinical contours) and a verification autocontouring system (secondary system to test the primary contours) were used to generate a pair of 6 female pelvic structures (UteroCervix [uterus + cervix], CTVn [nodal clinical target volume (CTV)], PAN [para-aortic lymph nodes], bladder, rectum, and kidneys) on 49 CT scans from our institution and 38 from other institutions. Additionally, clinically acceptable and unacceptable contours were manually generated using the 49 internal CT scans. Eleven similarity metrics (volumetric Dice similarity coefficient (DSC), Hausdorff distance, 95% Hausdorff distance, mean surface distance, and surface DSC with tolerances from 1 to 10 mm) were calculated between the reference and the verification autocontours, and between the manually generated and the verification autocontours. A support vector machine (SVM) was used to determine the threshold that separates clinically acceptable and unacceptable contours for each structure. The 11 metrics were investigated individually and in certain combinations. Linear, radial basis function, sigmoid, and polynomial kernels were tested using the combinations of metrics as inputs for the SVM. RESULTS: The highest contouring error detection accuracies were 0.91 for the UteroCervix, 0.90 for the CTVn, 0.89 for the PAN, 0.92 for the bladder, 0.95 for the rectum, and 0.97 for the kidneys and were achieved using surface DSCs with a thickness of 1, 2, or 3 mm. The linear kernel was the most accurate and consistent when a combination of metrics was used as an input for the SVM. However, the best model accuracy from the combinations of metrics was not better than the best model accuracy from a surface DSC as an input. CONCLUSIONS: We distinguished clinically acceptable contours from clinically unacceptable contours with an accuracy higher than 0.9 for the targets and critical structures in patients with cervical cancer; the most accurate similarity metric was surface DSC with a thickness of 1, 2, or 3 mm.

The Emergence of Artificial Intelligence within Radiation Oncology Treatment Planning.

BACKGROUND: The future of artificial intelligence (AI) heralds unprecedented change for the field of radiation oncology. Commercial vendors and academic institutions have created AI tools for radiation oncology, but such tools have not yet been widely adopted into clinical practice. In addition, numerous discussions have prompted careful thoughts about AI's impact upon the future landscape of radiation oncology: How can we preserve innovation, creativity, and patient safety? When will AI-based tools be widely adopted into the clinic? Will the need for clinical staff be reduced? How will these devices and tools be developed and regulated? SUMMARY: In this work, we examine how deep learning, a rapidly emerging subset of AI, fits into the broader historical context of advancements made in radiation oncology and medical physics. In addition, we examine a representative set of deep learning-based tools that are being made available for use in external beam radiotherapy treatment planning and how these deep learning-based tools and other AI-based tools will impact members of the radiation treatment planning team. Key Messages: Compared to past transformative innovations explored in this article, such as the Monte Carlo method or intensity-modulated radiotherapy, the development and adoption of deep learning-based tools is occurring at faster rates and promises to transform practices of the radiation treatment planning team. However, accessibility to these tools will be determined by each clinic's access to the internet, web-based solutions, or high-performance computing hardware. As seen by the trends exhibited by many technologies, high dependence on new technology can result in harm should the product fail in an unexpected manner, be misused by the operator, or if the mitigation to an expected failure is not adequate. Thus, the need for developers and researchers to rigorously validate deep learning-based tools, for users to understand how to operate tools appropriately, and for professional bodies to develop guidelines for their use and maintenance is essential. Given that members of the radiation treatment planning team perform many tasks that are automatable, the use of deep learning-based tools, in combination with other automated treatment planning tools, may refocus tasks performed by the treatment planning team and may potentially reduce resource-related burdens for clinics with limited resources.

Impact of slice thickness, pixel size, and CT dose on the performance of automatic contouring algorithms.

PURPOSE: To investigate the impact of computed tomography (CT) image acquisition and reconstruction parameters, including slice thickness, pixel size, and dose, on automatic contouring algorithms. METHODS: Eleven scans from patients with head-and-neck cancer were reconstructed with varying slice thicknesses and pixel sizes. CT dose was varied by adding noise using low-dose simulation software. The impact of these imaging parameters on two in-house auto-contouring algorithms, one convolutional neural network (CNN)-based and one multiatlas-based system (MACS) was investigated for 183 reconstructed scans. For each algorithm, auto-contours for organs-at-risk were compared with auto-contours from scans with 3 mm slice thickness, 0.977 mm pixel size, and 100% CT dose using Dice similarity coefficient (DSC), Hausdorff distance (HD), and mean surface distance (MSD). RESULTS: Increasing the slice thickness from baseline value of 3 mm gave a progressive reduction in DSC and an increase in HD and MSD on average for all structures. Reducing the CT dose only had a relatively minimal effect on DSC and HD. The rate of change with respect to dose for both auto-contouring methods is approximately 0. Changes in pixel size had a small effect on DSC and HD for CNN-based auto-contouring with differences in DSC being within 0.07. Small structures had larger deviations from the baseline values than large structures for DSC. The relative differences in HD and MSD between the large and small structures were small. CONCLUSIONS: Auto-contours can deviate substantially with changes in CT acquisition and reconstruction parameters, especially slice thickness and pixel size. The CNN was less sensitive to changes in pixel size, and dose levels than the MACS. The results contraindicated more restrictive values for the parameters should be used than a typical imaging protocol for head-and-neck.

Beam energy metrics for the acceptance and quality assurance of Halcyon linear accelerator.

PURPOSE: Establish and compare two metrics for monitoring beam energy changes in the Halcyon platform and evaluate the accuracy of these metrics across multiple Halcyon linacs. METHOD: The first energy metric is derived from the diagonal normalized flatness (FDN ), which is defined as the ratio of the average measurements at a fixed off-axis equal distance along the open profiles in two diagonals to the measurement at the central axis with an ionization chamber array (ICA). The second energy metric comes from the area ratio (AR) of the quad wedge (QW) profiles measured with the QW on the top of the ICA. Beam energy is changed by adjusting the magnetron current in a non-clinical Halcyon. With D10cm measured in water at each beam energy, the relationships between FDN or AR energy metrics to D10cm in water is established with linear regression across six energy settings. The coefficients from these regressions allow D10cm (FDN ) calculation from FDN using open profiles and D10cm (QW) calculation from AR using QW profiles. RESULTS: Five Halcyon linacs from five institutions were used to evaluate the accuracy of the D10cm (FDN ) and the D10cm (QW) energy metrics by comparing to the D10cm values computed from the treatment planning system (TPS) and D10cm measured in water. For the five linacs, the D10cm (FDN ) reported by the ICA based on FDN from open profiles agreed with that calculated by TPS within -0.29 ± 0.23% and 0.61% maximum discrepancy; the D10cm (QW) reported by the QW profiles agreed with that calculated by TPS within -0.82 ± 1.27% and -2.43% maximum discrepancy. CONCLUSION: The FDN -based energy metric D10cm (FDN ) can be used for acceptance testing of beam energy, and also for the verification of energy in periodic quality assurance (QA) processes.

Clinical implementation of automated treatment planning for whole-brain radiotherapy.

The purpose of the study was to develop and clinically deploy an automated, deep learning-based approach to treatment planning for whole-brain radiotherapy (WBRT). We collected CT images and radiotherapy treatment plans to automate a beam aperture definition from 520 patients who received WBRT. These patients were split into training (n = 312), cross-validation (n = 104), and test (n = 104) sets which were used to train and evaluate a deep learning model. The DeepLabV3+ architecture was trained to automatically define the beam apertures on lateral-opposed fields using digitally reconstructed radiographs (DRRs). For the beam aperture evaluation, 1st quantitative analysis was completed using a test set before clinical deployment and 2nd quantitative analysis was conducted 90 days after clinical deployment. The mean surface distance and the Hausdorff distances were compared in the anterior-inferior edge between the clinically used and the predicted fields. Clinically used plans and deep-learning generated plans were evaluated by various dose-volume histogram metrics of brain, cribriform plate, and lens. The 1st quantitative analysis showed that the average mean surface distance and Hausdorff distance were 7.1 mm (±3.8 mm) and 11.2 mm (±5.2 mm), respectively, in the anterior-inferior edge of the field. The retrospective dosimetric comparison showed that brain dose coverage (D99%, D95%, D1%) of the automatically generated plans was 29.7, 30.3, and 32.5 Gy, respectively, and the average dose of both lenses was up to 19.0% lower when compared to the clinically used plans. Following the clinical deployment, the 2nd quantitative analysis showed that the average mean surface distance and Hausdorff distance between the predicted and clinically used fields were 2.6 mm (±3.2 mm) and 4.5 mm (±5.6 mm), respectively. In conclusion, the automated patient-specific treatment planning solution for WBRT was implemented in our clinic. The predicted fields appeared consistent with clinically used fields and the predicted plans were dosimetrically comparable.

Development and application of an elastic net logistic regression model to investigate the impact of cardiac substructure dose on radiation-induced pericardial effusion in patients with NSCLC.

BACKGROUND: Typically, cardiac substructures are neither delineated nor analyzed during radiation treatment planning. Therefore, we developed a novel machine learning model to evaluate the impact of cardiac substructure dose for predicting radiation-induced pericardial effusion (PCE). MATERIALS AND METHODS: One-hundred and forty-one stage III NSCLC patients, who received radiation therapy in a prospective clinical trial, were included in this analysis. The impact of dose-volume histogram (DVH) metrics (mean and max dose, V5Gy[%]-V70Gy[%]) for the whole heart, left and right atrium, and left and right ventricle, on pericardial effusion toxicity (≥grade 2, CTCAE v4.0 grading) were examined. Elastic net logistic regression, using repeat cross-validation (n = 100 iterations, 75%/25% training/test set data split), was conducted with cardiac-based DVH metrics as covariates. The following model types were constructed and analyzed: (i) standard model type, which only included whole-heart DVH metrics; and (ii) a model type trained with both whole-heart and substructure DVH metrics. Model performance was analyzed on the test set using area under the curve (AUC), accuracy, calibration slope and calibration intercept. A final fitted model, based on the optimal model type, was developed from the entire study population for future comparisons. RESULTS: Grade 2 PCE incidence was 49.6% (n = 70). Models using whole heart and substructure dose had the highest performance (median values: AUC = 0.820; calibration slope/intercept = 1.356/-0.235; accuracy = 0.743) and outperformed the standard whole-heart only model type (median values: AUC = 0.799; calibration slope/intercept = 2.456/-0.729; accuracy = 0.713). The final fitted elastic net model showed high performance in predicting PCE (median values: AUC = 0.879; calibration slope/intercept = 1.352/-0.174; accuracy = 0.801). CONCLUSIONS: We developed and evaluated elastic net regression toxicity models of radiation-induced PCE. We found the model type that included cardiac substructure dose had superior predictive performance. A final toxicity model that included cardiac substructure dose metrics was developed and reported for comparison with external datasets.