NRG Oncology Assessment of Artificial Intelligence Deep Learning-Based Auto-segmentation for Radiation Therapy: Current Developments, Clinical Considerations, and Future Directions.

Deep learning neural networks (DLNN) in Artificial intelligence (AI) have been extensively explored for automatic segmentation in radiotherapy (RT). In contrast to traditional model-based methods, data-driven AI-based models for auto-segmentation have shown high accuracy in early studies in research settings and controlled environment (single institution). Vendor-provided commercial AI models are made available as part of the integrated treatment planning system (TPS) or as a stand-alone tool that provides streamlined workflow interacting with the main TPS. These commercial tools have drawn clinics' attention thanks to their significant benefit in reducing the workload from manual contouring and shortening the duration of treatment planning. However, challenges occur when applying these commercial AI-based segmentation models to diverse clinical scenarios, particularly in uncontrolled environments. Contouring nomenclature and guideline standardization has been the main task undertaken by the NRG Oncology. AI auto-segmentation holds the potential clinical trial participants to reduce interobserver variations, nomenclature non-compliance, and contouring guideline deviations. Meanwhile, trial reviewers could use AI tools to verify contour accuracy and compliance of those submitted datasets. In recognizing the growing clinical utilization and potential of these commercial AI auto-segmentation tools, NRG Oncology has formed a working group to evaluate the clinical utilization and potential of commercial AI auto-segmentation tools. The group will assess in-house and commercially available AI models, evaluation metrics, clinical challenges, and limitations, as well as future developments in addressing these challenges. General recommendations are made in terms of the implementation of these commercial AI models, as well as precautions in recognizing the challenges and limitations.

Intra-operative Radiation Therapy for Colorectal or Anal Cancer at Risk for Margin-Positive Resection: Initial Results of a Single-Institution Registry.

PURPOSE: Pelvic recurrence of rectal or anal cancers is associated with considerable morbidity and mortality. We report our initial experience with an aggressive intra-operative radiotherapy (IORT) program. METHODS: Patients with locally advanced or recurrent rectal or anal cancers considered to have a high likelihood of R1 or R2 resection after multi-disciplinary review underwent surgical excision and IORT using a high-dose-rate afterloader (Ir-192) and HAM applicator. Endpoints included local or distant recurrence, and acute and late toxicity graded using the American College of Surgeons (ACS) NSQIP and the LENT-SOMA scale. RESULTS: Twenty-one patients, largely with prior history of both pelvic external beam radiotherapy (EBRT, median 50.4 Gy) and surgical resection, underwent excision with IORT (median dose 12.5 Gy, range 10-15). Median follow-up was 20 months. Twelve (57%) patients had failure at the IORT site. Freedom from failure (FFF) within the IORT field was associated with resection status (FFF at 1 year 75% for R0 vs 15% for R1/2, p = 0.0065) but not re-irradiation EBRT or IORT dose (p > 0.05). Twelve, 5, and 13 patients experienced local, regional, and distant failure, respectively; 3 (14%) patients were disease-free at last follow-up. The most frequent acute toxicity was sepsis/abscess (24%). One patient (5%) required a ureteral stent; no patients developed neuropathy attributable to IORT. CONCLUSIONS: In patients treated with excision and IORT for locally recurrent cancer, R0 resection is a critical determinant of local control. For patients with R1/2 resection, poor disease-free outcomes warrant consideration of a different treatment strategy.

Toward understanding deep learning classification of anatomic sites: lessons from the development of a CBCT projection classifier.

Purpose: Deep learning (DL) applications strongly depend on the training dataset and convolutional neural network architecture; however, it is unclear how to objectively select such parameters. We investigate the classification performance of different DL models and training schemes for the anatomic classification of cone-beam computed tomography (CBCT) projections. Approach: CBCT scans from 1055 patients were collected and manually classified into five anatomic classes and used to develop DL models to predict the anatomic class from single x-ray projections. VGG-16, Xception, and Inception v3 architectures were trained with 75% of the data, and the remaining 25% was used for testing and evaluation. To study the dependence of the classification performance on dataset size, training data was downsampled to various dataset sizes. Gradient-weighted class activation maps (grad-CAM) were generated using the model with highest classification performance, to identify regions with strong influence on CNN decisions. Results: The highest precision and recall values were achieved with VGG-16. One of the best performing combinations was the VGG-16 trained with 90 deg projections (mean class precision = 0.87). The training dataset size could be reduced to ∼ 50 % of its initial size, without compromising the classification performance. For correctly classified cases, Grad-CAM were more heavily weighted for anatomically relevant regions. Conclusions: It was possible to determine those dependencies with a higher influence on the classification performance of DL models for the studied task. Grad-CAM enabled the identification of possible sources of class confusion.

AAPM task group report 302: Surface-guided radiotherapy.

The clinical use of surface imaging has increased dramatically, with demonstrated utility for initial patient positioning, real-time motion monitoring, and beam gating in a variety of anatomical sites. The Therapy Physics Subcommittee and the Imaging for Treatment Verification Working Group of the American Association of Physicists in Medicine commissioned Task Group 302 to review the current clinical uses of surface imaging and emerging clinical applications. The specific charge of this task group was to provide technical guidelines for clinical indications of use for general positioning, breast deep-inspiration breath hold treatment, and frameless stereotactic radiosurgery. Additionally, the task group was charged with providing commissioning and on-going quality assurance (QA) requirements for surface-guided radiation therapy (SGRT) as part of a comprehensive QA program including risk assessment. Workflow considerations for other anatomic sites and for computed tomography simulation, including motion management, are also discussed. Finally, developing clinical applications, such as stereotactic body radiotherapy (SBRT) or proton radiotherapy, are presented. The recommendations made in this report, which are summarized at the end of the report, are applicable to all video-based SGRT systems available at the time of writing.

A Phase 1 Trial of Concurrent or Sequential Ipilimumab, Nivolumab, and Stereotactic Body Radiotherapy in Patients With Stage IV NSCLC Study.

INTRODUCTION: Previous studies have evaluated stereotactic body radiotherapy (SBRT) in oligometastatic patients with NSCLC, including multimodality treatment with anti-programmed cell death protein-1 monotherapy. Questions remain regarding the timing of SBRT and immunotherapy, safety with dual checkpoint blockade, and the utility in widely metastatic patients. This randomized phase 1 trial combined nivolumab and ipilimumab with sequential or concurrent multisite SBRT in patients with stage IV NSCLC to evaluate safety and obtain preliminary activity data. METHODS: Treatment-naive patients with metastatic NSCLC were randomized to concurrent (SBRT with immunotherapy) or sequential (SBRT followed by immunotherapy) treatment. A maximum of four treatment fields received SBRT. Nivolumab and ipilimumab were continued until clinical progression, development of toxicity, or after 2 years. Dose-limiting toxicity was defined as greater than or equal to grade 3 toxicity to the relevant organ system attributed to SBRT and immunotherapy occuring within 3 months. RESULTS: A total of 37 patients were assessable. No dose-limiting toxicity occurred in the concurrent cohort (n = 18). The sequential cohort required a dose reduction in the central lung group owing to two grade 4 pneumonitis events (2 of 19). Overall best response was as follows: 5.4% (2 of 37) complete response, 40.5% (15 of 37) partial response, 16.2% (6 of 37) stable disease, and 37.8% (14 of 37) progressive disease. Median progression-free survival was 5.8 months (95% confidence interval: 3.6-11.4 mo), with median follow-up of 17.0 months. Median overall survival was not reached. CONCLUSIONS: Concurrent nivolumab, ipilimumab, and SBRT were not more toxic than sequential therapy, and multisite SBRT was well tolerated in widely metastatic patients. Multimodality therapy resulted in durable metastasis control and encouraging early overall survival.

Evaluation of Dose Distribution to Organs-at-Risk in a Prospective Phase 1 Trial of Pembrolizumab and Multisite Stereotactic Body Radiation Therapy (SBRT).

PURPOSE: Our purpose was to characterize the radiation doses to organs-at-risk (OAR) in the phase I trial (NCT02608385) that established safety/efficacy of stereotactic body radiation therapy (SBRT) using NRG-BR001 dose constraints combined with programmed cell death protein 1 blockade for metastatic disease. METHODS AND MATERIALS: Between January 2016 and May 2018, 73 patients with advanced solid tumors were treated with SBRT followed by pembrolizumab. Tumor volumes (gross tumor volume/internal tumor volume) were delineated for each metastasis, with planning target volume contraction to limit OAR dose per protocol (n = 54) or when gross tumor volume/internal tumor volume > 65 cm3 (n = 19). For 20 OAR, doses were compared with NRG-BR001 constraints. Protocol constraints were considered challenged when the minimum of the highest dose received by ≥6 patients without dose-limiting toxicities (DLTs) (Dmax6th) was ≥70% of the protocol constraint. RESULTS: A total of 151 metastases were irradiated including 32 peripheral lung, 23 central lung, 13 mediastinal/cervical, 24 liver, 28 abdominal-pelvic, 16 osseous, and 15 spinal metastases. A median of 2 metastases (range, 2-4) with mean volumes of 33.5 cm3 (range, 0.4-391 cm3) were treated using average planning target volumes of 50.7 cm3 (range, 3.2-161 cm3). At least 1 dose constraint from NRG-BR001 was exceeded in 38 of 73 (52%) patients. OAR constraints were challenged in 10 serial organs (gastrointestinal, cardio-pulmonary, musculoskeletal, and nervous systems) and 1 parallel OAR (lung). Grade 3 DLTs occurred in 6 patients, including pneumonitis (n = 3), colitis (n = 2), and hepatic failure (n = 1). In 4 patients, the toxicity could be directly attributed to the planned dose to OAR (ie, pneumonitis due to high lung dose or colitis due to high bowel dose). CONCLUSIONS: Multisite SBRT in combination with programmed cell death protein 1 blockade was safely tolerated when treating critical central, abdominal-pelvic, and peripheral OAR nearing NRG-BR001 constraints with clinically acceptable toxicity in the corresponding organ systems. The observed relationship between dose and DLTs in 4 of 6 patients indicates that NRG-BR001 dose constraints should be respected in subsequent trials to maintain clinical safety.

A framework for automated and streamlined kV cone beam computed tomography image quality assurance: a multi-institutional study.

The purpose of this study was to develop and evaluate a framework to support automated standardized testing and analysis of Cone Beam Computed Tomography (CBCT) image quality QA across multiple institutions. A survey was conducted among the participating institutions to understand the variability of the CBCT QA practices. A commercial, automated software platform was validated by seven institutions participating in a consortium dedicated to automated quality assurance. The CBCT image analysis framework was used to compare periodic QA results among 23 linear accelerators (linacs) from seven institutions. The CBCT image quality metrics (geometric distortion, spatial resolution, contrast, HU constancy, uniformity and noise) data are plotted as a function of means with the upper and lower control limits compared to the linac acceptance criteria and AAPM recommendations. For example, mean geometric distortion and HU constancy metrics were found to be 0.13 mm (TG142 recommendation: ≤2 mm) and 13.4 respectively (manufacturer acceptance specification: ≤±50).Image upload and analysis process was fully automated using a MATLAB-based platform. This analysis enabled a quantitative, longitudinal assessment of the performance of quality metrics which were also compared across 23 linacs. For key CBCT parameters such as uniformity, contrast, and HU constancy, all seven institutions used stricter goals than what would be recommended based on the analysis of the upper and lower control limits. These institutional goals were also found to be stricter than that found in AAPM published guidance. This work provides a reference that could be used to machine-specific optimized tolerance of CBCT image maintenance via control charts to monitor performance we well as the sensitivity of different tests in support of a broader quality assurance program. To ensure the daily image quality needed for patient care, the optimized statistical QA metrics recommended to using along with risk-based QA.

The role of surface-guided radiation therapy for improving patient safety.

Emerging data indicates SGRT could improve safety and quality by preventing errors in its capacity as an independent system in the treatment room. The aim of this work is to investigate the utility of SGRT in the context of safety and quality. Three incident learning systems (ILS) were reviewed to categorize and quantify errors that could have been prevented with SGRT: SAFRON (International Atomic Energy Agency), UW-ILS (University of Washington) and AvIC (Skåne University Hospital). A total of 849/9737 events occurred during the pre-treatment review/verification and treatment stages. Of these, 179 (21%) events were predicted to have been preventable with SGRT. The most common preventable events were wrong isocentre (43%) and incorrect accessories (34%), which appeared at comparable rates among SAFRON and UW-ILS. The proportion of events due to wrong accessories was much smaller in the AvIC ILS, which may be attributable to the mandatory use of SGRT in Sweden. Several case scenarios are presented to demonstrate that SGRT operates as a valuable complement to other quality-improvement tools routinely used in radiotherapy. Cases are noted in which SGRT itself caused incidents. These were mostly related to workflow issues and were of low severity. Severity data indicated that events with the potential to be mitigated by SGRT were of higher severity for all categories except wrong accessories. Improved vendor integration of SGRT systems within the overall workflow could further enhance its clinical utility. SGRT is a valuable tool with the potential to increase patient safety and treatment quality in radiotherapy.

Development of a neonate X-ray phantom for 2D imaging applications using single-tone inkjet printing.

PURPOSE: Inkjet printers can be used to fabricate anthropomorphic phantoms by the use of iodine-doped ink. However, challenges persist in implementing this technique. The calibration from grayscale to ink density is complex and time-consuming. The purpose of this work is to develop a printing methodology that requires a simpler calibration and is less dependent on printer characteristics to produce the desired range of x-ray attenuation values. METHODS: Conventional grayscale printing was substituted by single-tone printing; that is, the superposition of pure black layers of iodinated ink. Printing was performed with a consumer-grade inkjet printer using ink made of potassium-iodide (KI) dissolved in water at 1 g/ml. A calibration for the attenuation of ink was measured using a commercial x-ray system at 70 kVp. A neonate radiograph obtained at 70 kVp served as an anatomical model. The attenuation map of the neonate radiograph was processed into a series of single-tone images. Single-tone images were printed, stacked, and imaged at 70 kVp. The phantom was evaluated by comparing attenuation values between the printed phantom and the original radiograph; attenuation maps were compared using the structural similarity index measure (SSIM), while attenuation histograms were compared using the Kullback-Leibler (KL) divergence. A region of interest (ROI)-based analysis was also performed, where the attenuation distribution within given ROIs was compared between phantom and patient. The phantom sharpness was evaluated in terms of modulation transfer function (MTF) estimates and signal spread profiles of high spatial resolution features in the image. RESULTS: The printed phantom required 36 pages. The printing queue was automated and it took about 2 h to print the phantom. The radiograph of the printed phantom demonstrated a close resemblance to the original neonate radiograph. The SSIM of the phantom with respect to that of the patient was 0.53. Both patient and phantom attenuation histograms followed similar distributions, and the KL divergence between such histograms was 0.20. The ROI-based analysis showed that the largest deviations from patient attenuation values were observed at the higher and lower ends of the attenuation range. The limiting resolution of the proposed methodology was about 1 mm. CONCLUSION: A methodology to generate a neonate phantom for 2D imaging applications, using single-tone printing, was developed. This method only requires a single-value calibration and required less than 2 h to print a complete phantom.

AAPM medical physics practice guideline 3.b.: Levels of supervision for medical physicists in clinical training.

The American Association of Physicists in Medicine (AAPM) is a nonprofit professional society whose primary purposes are to advance the science, education and professional practice of medical physics. The AAPM has more than 8,000 members and is the principal organization of medical physicists in the United States. The AAPM will periodically define new practice guidelines for medical physics practice to help advance the science of medical physics and to improve the quality of service to patients throughout the United States. Existing medical physics practice guidelines will be reviewed for the purpose of revision or renewal, as appropriate, on their fifth anniversary or sooner. Each medical physics practice guideline represents a policy statement by the AAPM, has undergone a thorough consensus process in which it has been subjected to extensive review, and requires the approval of the Professional Council. The medical physics practice guidelines recognize that the safe and effective use of diagnostic and therapeutic radiology requires specific training, skills, and techniques, as described in each document. Reproduction or modification of the published practice guidelines and technical standards by those entities not providing these services is not authorized. The following terms are used in the AAPM practice guidelines: (1) Must and Must Not: Used to indicate that adherence to the recommendation is considered necessary to conform to this practice guideline. (2) Should and Should Not: Used to indicate a prudent practice to which exceptions may occasionally be made in appropriate circumstances.

Evaluation of Safety of Stereotactic Body Radiotherapy for the Treatment of Patients With Multiple Metastases: Findings From the NRG-BR001 Phase 1 Trial.

IMPORTANCE: Stereotactic body radiotherapy (SBRT) for oligometastases is hypothesized to improve survival and is increasingly used. Little evidence supports its safe use to treat patients with multiple metastases. OBJECTIVE: To establish safety of SBRT dose schedules in patients with 3 to 4 metastases or 2 metastases in close proximity to each other. DESIGN, SETTING, AND PARTICIPANTS: This phase 1 trial opened on August 4, 2014, and closed to accrual on March 20, 2018. Metastases to 7 anatomic locations were included: bone/osseous (BO), spinal/paraspinal (SP), peripheral lung (PL), central lung (CL), abdominal-pelvic (AP), mediastinal/cervical lymph node (MC), and liver (L). Six patients could be enrolled per anatomic site. The setting was a consortium of North American academic and community practice cancer centers participating in NRG Oncology trials. Patients with breast, prostate, or non-small cell lung cancer with 3 to 4 metastases or 2 metastases in close proximity (≤5 cm) amenable to SBRT were eligible for this phase 1 study. Statistical analyses were performed from December 31, 2017, to September 19, 2019. INTERVENTIONS: The starting dose was 50 Gy in 5 fractions (CL, MC), 45 Gy in 3 fractions (PL, AP, L), and 30 Gy in 3 fractions (BO, SP). MAIN OUTCOMES AND MEASURES: The primary end point was dose-limiting toxicity (DLT) defined by the Common Terminology Criteria for Adverse Events, version 4.0, as specific adverse events (AEs) of grades 3 to 5 (definite or probable per the protocol DLT definition) related to SBRT within 180 days of treatment. Dose levels were considered safe if DLTs were observed in no more than 1 of 6 patients per location; otherwise, the dose at that location would be de-escalated. RESULTS: A total of 42 patients enrolled, 39 were eligible, and 35 (mean [SD] age, 63.1 [14.2] years; 20 men [57.1%]; 30 White patients [85.7%]) were evaluable for DLT. Twelve patients (34.3%) had breast cancer, 10 (28.6%) had non-small cell lung cancer, and 13 (37.1%) had prostate cancer; there was a median of 3 metastases treated per patient. Median survival was not reached. No protocol-defined DLTs were observed. When examining all AEs, 8 instances of grade 3 AEs, most likely related to protocol therapy, occurred approximately 125 to 556 days from SBRT initiation in 7 patients. CONCLUSIONS AND RELEVANCE: This phase 1 trial demonstrated the safety of SBRT for patients with 3 to 4 metastases or 2 metastases in close proximity. There were no treatment-related deaths. Late grade 3 AEs demonstrate the need for extended follow-up in long-surviving patients with oligometastatic disease. Treatment with SBRT for multiple metastases has been expanded into multiple ongoing randomized phase 2/3 National Cancer Institute-sponsored trials (NRG-BR002, NRG-LU002). TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT02206334.

Harmonization of radiomic feature variability resulting from differences in CT image acquisition and reconstruction: assessment in a cadaveric liver.

Studies investigating the effects of computed tomography (CT) image acquisition and reconstruction parameters have mostly been limited to non-human phantoms to limit exposure to patients. This study investigates these variations using a cadaveric liver and determines harmonization methods to mitigate these variations. A reference CT scan of a cadaveric liver was acquired along with 16 modified scans. Modified scans were obtained with altered image acquisition and reconstruction parameters. In each slice, the liver was segmented and used to calculate 142 features. Student's t-tests assessed differences between reference and modified scans for each feature after correcting for multiple comparisons. Features were harmonized between reference and modified scans using histogram normalization, pixel resampling, Butterworth filtering, resampling and filtering combined, and ComBat harmonization. The number of features reflecting significant differences before and after harmonization were compared across imaging parameters. Reducing the field-of-view (FOV) and using coronal instead of axial scans resulted in the greatest number of features reflecting significant differences (67.6%, and 35.9%, respectively) and resulted in the greatest median relative change in feature values (25.4% and 18.2%, respectively). Changes in tube voltage, pitch, and slice interval resulted in the smallest number of features reflecting significance (0.7%) with median relative changes in feature <2%. Histogram normalization reduced or maintained the number of significantly different features for all scans, while ComBat reduced the number of significantly different features to zero for all scans. The remaining harmonization methods had mixed effects: resampling reduced the number of features reflecting significant differences for half of the imaging parameters, while filtering alone and filtering combined with resampling both reduced the number of features reflecting significance for 10 of the 16 parameters. The dependence of radiomic features on image acquisition and reconstruction parameters varies in a cadaveric liver; however, various harmonization methods have shown promise in mitigating these dependencies, particularly ComBat.

A study of the dosimetric impact of daily setup variations measured with cone-beam CT on three-dimensional conformal radiotherapy for early-stage breast cancer delivered in the prone position.

PURPOSE: To evaluate the dosimetric impact of daily positioning variations measured with cone-beam computed tomography (CBCT) on whole-breast radiotherapy patients treated in the prone position. METHODS: Daily CBCT was prospectively acquired for 30 consecutive patients positioned prone. Treatment for early-stage (≤II) breast cancer was prescribed with standard dose (50 Gy/25 fractions) or hypofractionation (42.56 Gy/16 fractions) for 13 and 17 patients, respectively. Systematic and random errors were calculated from the translational CBCT shifts and used to determine population-based setup margins. Mean translations (±one standard deviation) for each patient were used to simulate the dosimetric impact on targets (PTV_eval and lumpectomy cavity), heart, and lung. Paired Student's t tests at α = 0.01 were used to compare dose metrics after correction for multiple testing (P < 0.002). Significant correlation coefficients were used to identify associations (P < 0.01). RESULTS: Of 597 total fractions, 20 ± 13% required patient rotation. Mean translations were 0.29 ± 0.27 cm, 0.41 ± 0.34 cm, and 0.48 ± 0.33 cm in the anterior-posterior, superior-inferior, and lateral directions leading to calculated setup margins of 0.63, 0.88, and 1.10 cm, respectively. Average three-dimensional (3D) shifts correlated with the maximum distance of breast tissue from the sternum (r = 0.62) but not with body-mass index. Simulated shifts showed significant, but minor, changes in dose metrics for PTV_eval, lung, and heart. For left-sided treatments (n = 18), mean heart dose increased from 109 ± 75 cGy to 148 ± 115 cGy. Shifts from the original plan caused PTV_eval hotspots (V105%) to increase by 5.2% ± 3.8%, which correlated with the total MU of wedged fields (r = 0.59). No significant change in V95% to the cavity was found. CONCLUSIONS: Large translational variations that occur when positioning prone breast patients had small but significant dosimetric effects on 3DCRT plans. Daily CBCT may still be necessary to correct for rotational variations that occur in 20% of treatments. To maintain planned dose metrics, unintended beam shifts toward the heart and the contribution of wedged fields should be minimized.