Augmenting Quality Assurance Measures in Treatment Review with Machine Learning in Radiation Oncology.

Purpose: Pretreatment quality assurance (QA) of treatment plans often requires a high cognitive workload and considerable time expenditure. This study explores the use of machine learning to classify pretreatment chart check QA for a given radiation plan as difficult or less difficult, thereby alerting the physicists to increase scrutiny on difficult plans. Methods and Materials: Pretreatment QA data were collected for 973 cases between July 2018 and October 2020. The outcome variable, a degree of difficulty, was collected as a subjective rating by physicists who performed the pretreatment chart checks. Potential features were identified based on clinical relevance, contribution to plan complexity, and QA metrics. Five machine learning models were developed: support vector machine, random forest classifier, adaboost classifier, decision tree classifier, and neural network. These were incorporated into a voting classifier, where at least 2 algorithms needed to predict a case as difficult for it to be classified as such. Sensitivity analyses were conducted to evaluate feature importance. Results: The voting classifier achieved an overall accuracy of 77.4% on the test set, with 76.5% accuracy on difficult cases and 78.4% accuracy on less difficult cases. Sensitivity analysis showed features associated with plan complexity (number of fractions, dose per monitor unit, number of planning structures, and number of image sets) and clinical relevance (patient age) were sensitive across at least 3 algorithms. Conclusions: This approach can be used to equitably allocate plans to physicists rather than randomly allocate them, potentially improving pretreatment chart check effectiveness by reducing errors propagating downstream.

Multi-Institutional Stereotactic Body Radiation Therapy Incident Learning: Evaluation of Safety Barriers Using a Human Factors Analysis and Classification System.

OBJECTIVES: Stereotactic body radiation therapy (SBRT) can improve therapeutic ratios and patient convenience, but delivering higher doses per fraction increases the potential for patient harm. Incident learning systems (ILSs) are being increasingly adopted in radiation oncology to analyze reported events. This study used an ILS coupled with a Human Factor Analysis and Classification System (HFACS) and barriers management to investigate the origin and detection of SBRT events and to elucidate how safeguards can fail allowing errors to propagate through the treatment process. METHODS: Reported SBRT events were reviewed using an in-house ILS at 4 institutions over 2014-2019. Each institution used a customized care path describing their SBRT processes, including designated safeguards to prevent error propagation. Incidents were assigned a severity score based on the American Association of Physicists in Medicine Task Group Report 275. An HFACS system analyzed failing safeguards. RESULTS: One hundred sixty events were analyzed with 106 near misses (66.2%) and 54 incidents (33.8%). Fifty incidents were designated as low severity, with 4 considered medium severity. Incidents most often originated in the treatment planning stage (38.1%) and were caught during the pretreatment review and verification stage (37.5%) and treatment delivery stage (31.2%). An HFACS revealed that safeguard failures were attributed to human error (95.2%), routine violation (4.2%), and exceptional violation (0.5%) and driven by personnel factors 32.1% of the time, and operator condition also 32.1% of the time. CONCLUSIONS: Improving communication and documentation, reducing time pressures, distractions, and high workload should guide proposed improvements to safeguards in radiation oncology.

Common Error Pathways in CyberKnife™ Radiation Therapy.

Purpose/Objectives: Stereotactic radiosurgery (SRS) and stereotactic body radiation therapy (SBRT) may be considered "high risk" due to the high doses per fraction. We analyzed CyberKnife™ (CK) SRS and SBRT-related incidents that were prospectively reported to our in-house incident learning system (ILS) in order to identify severity, contributing factors, and common error pathways. Material and Methods: From 2012 to 2019, 221 reported incidents related to the 4,569 CK fractions delivered (5.8%) were prospectively analyzed by our multi-professional Quality and Safety Committee with regard to severity, contributing factors, as well as the location where the incident occurred (tripped), where it was discovered (caught), and the safety barriers that were traversed (crossed) on the CK process map. Based on the particular step in the process map that incidents tripped, we categorized incidents into general error pathways. Results: There were 205 severity grade 1-2 (did not reach patient or no clinical impact), 11 grade 3 (clinical impact unlikely), 5 grade 4 (altered the intended treatment), and 0 grade 5-6 (life-threatening or death) incidents, with human performance being the most common contributing factor (79% of incidents). Incidents most commonly tripped near the time when the practitioner requested CK simulation (e.g., pre-CK simulation fiducial marker placement) and most commonly caught during the physics pre-treatment checklist. The four general error pathways included pre-authorization, billing, and scheduling issues (n= 119); plan quality (n= 30); administration of IV contrast during simulation or pre-medications during treatment (n= 22); and image guidance (n= 12). Conclusion: Most CK incidents led to little or no patient harm and most were related to billing and scheduling issues. Suboptimal human performance appeared to be the most common contributing factor to CK incidents. Additional study is warranted to develop and share best practices to reduce incidents to further improve patient safety.

Human Error Bowtie Analysis to Enhance Patient Safety in Radiation Oncology.

PURPOSE: Ensuring safety within RT is of paramount importance. To further support and augment patient safety efforts, the purpose of this research was to test and refine a robust methodology for analyzing human errors that defeat individual controls within RT quality assurance (QA) programs. METHODS: The method proposed for performing Bowtie Analysis (BTA) was based on training and recommendations from practitioners in the field of Human Factors and Ergonomics practice. Multidisciplinary meetings to iteratively develop BTA focused on incorrect site setup instructions was conducted. RESULTS: From November 2015 to February 2017, we had 12 reported incidents related to site setup notes that could have led to site setup errors. Based on this data, we conducted five BTA analyses related to incorrect site setup instructions. None of the individual controls within our QA program designed to check for potential errors with site setup instructions met the level of robustness to be classified as key safeguards or barriers. CONCLUSIONS: The relatively low number of incidents causing patient harm has led us to typically assume that we have sufficient and effective controls in place to prevent serious human errors from leading to severe patient consequences. Based on our BTA, we question how well we truly understand the details of our individual controls. To meet the level of safety achieved by high reliability organizations (HROs), we need to better ensure that our controls are as reliable and robust as we assume.

Task Group 174 Report: Utilization of [18 F]Fluorodeoxyglucose Positron Emission Tomography ([18 F]FDG-PET) in Radiation Therapy.

The use of positron emission tomography (PET) in radiation therapy (RT) is rapidly increasing in the areas of staging, segmentation, treatment planning, and response assessment. The most common radiotracer is 18 F-fluorodeoxyglucose ([18 F]FDG), a glucose analog with demonstrated efficacy in cancer diagnosis and staging. However, diagnosis and RT planning are different endeavors with unique requirements, and very little literature is available for guiding physicists and clinicians in the utilization of [18 F]FDG-PET in RT. The two goals of this report are to educate and provide recommendations. The report provides background and education on current PET imaging systems, PET tracers, intensity quantification, and current utilization in RT (staging, segmentation, image registration, treatment planning, and therapy response assessment). Recommendations are provided on acceptance testing, annual and monthly quality assurance, scanning protocols to ensure consistency between interpatient scans and intrapatient longitudinal scans, reporting of patient and scan parameters in literature, requirements for incorporation of [18 F]FDG-PET in treatment planning systems, and image registration. The recommendations provided here are minimum requirements and are not meant to cover all aspects of the use of [18 F]FDG-PET for RT.

Combining multiple FDG-PET radiotherapy target segmentation methods to reduce the effect of variable performance of individual segmentation methods.

PURPOSE: Many approaches have been proposed to segment high uptake objects in 18F-fluoro-deoxy-glucose positron emission tomography images but none provides consistent performance across the large variety of imaging situations. This study investigates the use of two methods of combining individual segmentation methods to reduce the impact of inconsistent performance of the individual methods: simple majority voting and probabilistic estimation. METHODS: The National Electrical Manufacturers Association image quality phantom containing five glass spheres with diameters 13-37 mm and two irregularly shaped volumes (16 and 32 cc) formed by deforming high-density polyethylene bottles in a hot water bath were filled with 18-fluoro-deoxyglucose and iodine contrast agent. Repeated 5-min positron emission tomography (PET) images were acquired at 4:1 and 8:1 object-to-background contrasts for spherical objects and 4.5:1 and 9:1 for irregular objects. Five individual methods were used to segment each object: 40% thresholding, adaptive thresholding, k-means clustering, seeded region-growing, and a gradient based method. Volumes were combined using a majority vote (MJV) or Simultaneous Truth And Performance Level Estimate (STAPLE) method. Accuracy of segmentations relative to CT ground truth volumes were assessed using the Dice similarity coefficient (DSC) and the symmetric mean absolute surface distances (SMASDs). RESULTS: MJV had median DSC values of 0.886 and 0.875; and SMASD of 0.52 and 0.71 mm for spheres and irregular shapes, respectively. STAPLE provided similar results with median DSC of 0.886 and 0.871; and median SMASD of 0.50 and 0.72 mm for spheres and irregular shapes, respectively. STAPLE had significantly higher DSC and lower SMASD values than MJV for spheres (DSC, p < 0.0001; SMASD, p = 0.0101) but MJV had significantly higher DSC and lower SMASD values compared to STAPLE for irregular shapes (DSC, p < 0.0001; SMASD, p = 0.0027). DSC was not significantly different between 128 × 128 and 256 × 256 grid sizes for either method (MJV, p = 0.0519; STAPLE, p = 0.5672) but was for SMASD values (MJV, p < 0.0001; STAPLE, p = 0.0164). The best individual method varied depending on object characteristics. However, both MJV and STAPLE provided essentially equivalent accuracy to using the best independent method in every situation, with mean differences in DSC of 0.01-0.03, and 0.05-0.12 mm for SMASD. CONCLUSIONS: Combining segmentations offers a robust approach to object segmentation in PET. Both MJV and STAPLE improved accuracy and were robust against the widely varying performance of individual segmentation methods. Differences between MJV and STAPLE are such that either offers good performance when combining volumes. Neither method requires a training dataset but MJV is simpler to interpret, easy to implement and fast.

Influence of filter choice on 18F-FDG PET segmentation accuracy determined using generalized estimating equations.

This study aims to quantify how filter choice affects several fluoro-deoxy-glucose (FDG)-positron emission tomography (PET) segmentation methods and present the use of model fitting via generalized estimating equations (GEEs) to appropriately account for the properties of a common segmentation quality metric (Dice similarity coefficient). Spherical and irregularly shaped 'hot' objects filled with 18F-FDG were placed in a medium with background activity and imaged for 1, 2 and 5 min durations at low and high contrasts. Images were filtered with Gaussian and bilateral filters of 5 and 7 mm full-width half maximum (FWHM), with and without 3 mm FWHM Gaussian pre-smoothing. Four segmentation methods were used: 40% thresholding, adaptive thresholding, k-means clustering and seeded region-growing. Segmentation accuracy was quantified by overlap (using Dice similarity coefficient (DSC)) and distance between surfaces (using symmetric-mean-absolute-surface-distance (SMASD)) of the ground truth and segmented volumes. All segmentation methods showed mean DSC values between 0.71-0.87 and mean SMASD values between 0.72-2.10 mm across filters. The bilateral filter with 3 mm FWHM Gaussian pre-smoothing had mean DSC 0.80 ± 0.17 and mean SMASD 1.17 ± 1.51 mm displaying approximately equal performance to a 5 mm Gaussian filter with mean DSC 0.79 ± 0.18 and mean SMASD 1.27 ± 1.52 mm. Results from models fit using GEE with a binomial distribution and exchangeable correlation structure estimated the correlation between DSC values as 0.118 and 0.290 for spheres and irregular objects, respectively. The GEE approach accounts for several factors specific to the DSC metric that simpler statistical approaches do not, providing more accurate estimations of experimental effects commonly associated with nuclear medicine segmentation studies.

Development and dosimetry of a small animal lung irradiation platform.

Advances in large scale screening of medical countermeasures for radiation-induced normal tissue toxicity are currently hampered by animal irradiation paradigms that are both inefficient and highly variable among institutions. Here, a novel high-throughput small animal irradiation platform is introduced for use in orthovoltage small animal irradiators. Radiochromic film and metal oxide semiconductor field effect transistor detectors were used to examine several parameters, including 2D field uniformity, dose rate consistency, and shielding transmission. The authors posit that this setup will improve efficiency of drug screens by allowing for simultaneous targeted irradiation of multiple animals to improve efficiency within a single institution. Additionally, they suggest that measurement of the described parameters in all centers conducting countermeasure studies will improve the translatability of findings among institutions. The use of tissue equivalent phantoms in performing dosimetry measurements for small animal irradiation experiments was also investigated. Though these phantoms are commonly used in dosimetry, the authors recorded a significant difference in both the entrance and target tissue dose rates between euthanized rats and mice with implanted detectors and the corresponding phantom measurement. This suggests that measurements using these phantoms may not provide accurate dosimetry for in vivo experiments. Based on these measurements, the authors propose that this small animal irradiation platform can increase the capacity of animal studies by allowing for more efficient animal irradiation. They also suggest that researchers fully characterize the parameters of whatever radiation setup is in use in order to facilitate better comparison among institutions.

A preclinical rodent model of radiation-induced lung injury for medical countermeasure screening in accordance with the FDA animal rule.

The purpose of preclinical murine model development is to establish that the pathophysiological outcome of the rodent model of radiation-induced lung injury is sufficiently representative of the anticipated pulmonary response in the human population. This objective is based on concerns that the C57BL/6J strain may not be the most appropriate preclinical model of lethal radiation lung injury in humans. In this study, the authors assessed this issue by evaluating the relationship between morbidity (pulmonary function, histopathologic damage) and mortality among three strains of mice: C57BL/6J, CBA/J, and C57L/J. These different strains display variations in latency and phenotypic expression of radiation-induced lung damage. By comparing the response of each strain to the human pulmonary response, an appropriate animal model(s) of human radiation-induced pulmonary injury was established. Observations in the C57L/J and CBA/J murine models can be extrapolated to the human lung for evaluation of the mechanisms of action of radiation as well as future efficacy testing and approving agents that fall under the "Animal Rule" of the U.S. Food and Drug Administration (FDA) (21 CFR Parts 314 and 601).

Extension of the NCAT phantom for the investigation of intra-fraction respiratory motion in IMRT using 4D Monte Carlo.

The purpose of this work was to create a computational platform for studying motion in intensity modulated radiotherapy (IMRT). Specifically, the non-uniform rational B-spline (NURB) cardiac and torso (NCAT) phantom was modified for use in a four-dimensional Monte Carlo (4D-MC) simulation system to investigate the effect of respiratory-induced intra-fraction organ motion on IMRT dose distributions as a function of diaphragm motion, lesion size and lung density. Treatment plans for four clinical scenarios were designed: diaphragm peak-to-peak amplitude of 1 cm and 3 cm, and two lesion sizes--2 cm and 4 cm diameter placed in the lower lobe of the right lung. Lung density was changed for each phase using a conservation of mass calculation. Further, a new heterogeneous lung model was implemented and tested. Each lesion had an internal target volume (ITV) subsequently expanded by 15 mm isotropically to give the planning target volume (PTV). The PTV was prescribed to receive 72 Gy in 40 fractions. The MLC leaf sequence defined by the planning system for each patient was exported and used as input into the MC system. MC simulations using the dose planning method (DPM) code together with deformable image registration based on the NCAT deformation field were used to find a composite dose distribution for each phantom. These composite distributions were subsequently analyzed using information from the dose volume histograms (DVH). Lesion motion amplitude has the largest effect on the dose distribution. Tumor size was found to have a smaller effect and can be mitigated by ensuring the planning constraints are optimized for the tumor size. The use of a dynamic or heterogeneous lung density model over a respiratory cycle does not appear to be an important factor with a