Comparison of statistical machine learning models for rectal protocol compliance in prostate external beam radiation therapy.

PURPOSE: Limiting the dose to the rectum can be one of the most challenging aspects of creating a dosimetric external beam radiation therapy (EBRT) plan for prostate cancer treatment. Rectal sparing devices such as hydrogel spacers offer the prospect of increased space between the prostate and rectum, causing reduced rectal dose and potentially reduced injury. This study sought to help identify patients at higher risk of developing rectal injury based on estimated rectal dosimetry compliance prior to the EBRT simulation and planning procedure. Three statistical machine learning methods were compared for their ability to predict rectal dose outcomes with varied classification thresholds applied. METHODS: Prostate cancer patients treated with conventionally fractionated EBRT to a reference dose of 74-78 Gy were invited to participate in the study. The dose volume histogram data from each dosimetric plan was used to quantify planned rectal volume receiving 50%, 83% 96%, and 102% of the reference dose. Patients were classified into two groups for each of these dose levels: either meeting tolerance by having a rectal volume less than a clinically acceptable threshold for the dose level (Y) or violating the tolerance by having a rectal volume greater than the threshold for the dose level (N). Logistic regression, classification and regression tree, and random forest models were compared for their ability to discriminate between class outcomes. Performance metrics included area under the receiver operator characteristic curve (AUC), sensitivity, specificity, positive predictive value and negative predictive value. Finally, three classification threshold levels were evaluated for their impact on model performance. RESULTS: A total of 176 eligible participants were recruited. Variable importance differed between model methods. Area under the receiver operator characteristic curve performance varied greatly across the different rectal dose levels and between models. Logistic regression performed best at the 83% reference dose level with an AUC value of 0.844, while random forest demonstrated best discrimination at the 96% reference dose level with an AUC value of 0.733. In addition to the standard classification probability threshold of 50%, the clinically representative threshold of 10%, and the best threshold from each AUC plot was applied to compare metrics. This showed that using a 50% threshold and the best threshold from the AUC plots yields similar results. Conversely, applying the more conservative clinical threshold of 10% maximized the sensitivity at V83_RD and V96_RD for all model types. Based on the combination of the metrics, logistic regression would be the recommendation for rectal protocol compliance prediction at the 83% reference dose level, and random forest for the 96% reference dose level, particularly when using the clinical probability threshold of 10%. CONCLUSIONS: This study demonstrated the efficacy of statistical machine learning models on rectal protocol compliance prediction for prostate cancer EBRT dosimetric planning. Both logistic regression and random forest modeling approaches demonstrated good discriminative ability for predicting class outcomes in the upper dose levels. Application of a conservative clinical classification threshold maximized sensitivity and further confirmed the value of logistic regression and random forest models over classification and regression tree.

An image-guided radiotherapy decision support framework incorporating a Bayesian network and visualization tool.

PURPOSE: To describe a Bayesian network (BN) and complementary visualization tool that aim to support decision-making during online cone-beam computed tomography (CBCT)-based image-guided radiotherapy (IGRT) for prostate cancer patients. METHODS: The BN was created to represent relationships between observed prostate, proximal seminal vesicle (PSV), bladder and rectum volume variations, an image feature alignment score (FASTV_OAR ), delivered dose, and treatment plan compliance (TPC). Variables influencing tumor volume (TV) targeting accuracy such as intrafraction motion, and contouring and couch shift errors were also represented. A score of overall TPC (FASglobal ) and factors such as image quality were used to inform the BN output node providing advice about proceeding with treatment. The BN was quantified using conditional probabilities generated from published studies, FASTV_OAR/global modeling, and a survey of IGRT decision-making practices. A new IGRT visualization tool (IGRTREV ), in the form of Mollweide projection plots, was developed to provide a global summary of residual errors after online CBCT-planning CT registration. Sensitivity and scenario analyses were undertaken to evaluate the performance of the BN and the relative influence of the network variables on TPC and the decision to proceed with treatment. The IGRTREV plots were evaluated in conjunction with the BN scenario testing, using additional test data generated from retrospective CBCT-planning CT soft-tissue registrations for 13/36 patients whose data were used in the FASTV_OAR/global modeling. RESULTS: Modeling of the TV targeting errors resulted in a very low probability of corrected distances between the CBCT and planning CT prostate or PSV volumes being within their thresholds. Strength of influence evaluation with and without the BN TV targeting error nodes indicated that rectum- and bladder-related network variables had the highest relative importance. When the TV targeting error nodes were excluded from the BN, TPC was sensitive to observed PSV and rectum variations while the decision to treat was sensitive to observed prostate and PSV variations. When root nodes were set so the PSV and rectum variations exceeded thresholds, the probability of low TPC increased to 40%. Prostate and PSV variations exceeding thresholds increased the likelihood of repositioning or repeating patient preparation to 43%. Scenario testing using the test data from 13 patients, demonstrated two cases where the BN provided increased high TPC probabilities, despite some of the prostate and PSV volume variation metrics not being within tolerance. The IGRTREV tool was effective in highlighting and quantifying where TV and OAR variations occurred, supporting the BN recommendation to reposition the patient or repeat their bladder and bowel preparation. In another case, the IGRTREV tool was also effective in highlighting where PSV volume variation significantly exceeded tolerance when the BN had indicated to proceed with treatment. CONCLUSIONS: This study has demonstrated that both the BN and IGRTREV plots are effective tools for inclusion in a decision support system for online CBCT-based IGRT for prostate cancer patients. Alternate approaches to modeling TV targeting errors need to be explored as well as extension of the BN to support offline IGRT decisions related to adaptive radiotherapy.

A feature alignment score for online cone-beam CT-based image-guided radiotherapy for prostate cancer.

PURPOSE: To develop a method for scoring online cone-beam CT (CBCT)-to-planning CT image feature alignment to inform prostate image-guided radiotherapy (IGRT) decision-making. The feasibility of incorporating volume variation metric thresholds predictive of delivering planned dose into weighted functions, was investigated. METHODS: Radiation therapists and radiation oncologists participated in workshops where they reviewed prostate CBCT-IGRT case examples and completed a paper-based survey of image feature matching practices. For 36 prostate cancer patients, one daily CBCT was retrospectively contoured then registered with their plan to simulate delivered dose if (a) no online setup corrections and (b) online image alignment and setup corrections, were performed. Survey results were used to select variables for inclusion in classification and regression tree (CART) and boosted regression trees (BRT) modeling of volume variation metric thresholds predictive of delivering planned dose to the prostate, proximal seminal vesicles (PSV), bladder, and rectum. Weighted functions incorporating the CART and BRT results were used to calculate a score of individual tumor and organ at risk image feature alignment (FASTV_OAR ). Scaled and weighted FASTV_OAR were then used to calculate a score of overall treatment compliance (FASglobal ) for a given CBCT-planning CT registration. The FASTV_OAR were assessed for sensitivity, specificity, and predictive power. FASglobal thresholds indicative of high, medium, or low overall treatment plan compliance were determined using coefficients from multiple linear regression analysis. RESULTS: Thirty-two participants completed the prostate CBCT-IGRT survey. While responses demonstrated consensus of practice for preferential ranking of planning CT and CBCT match features in the presence of deformation and rotation, variation existed in the specified thresholds for observed volume differences requiring patient repositioning or repeat bladder and bowel preparation. The CART and BRT modeling indicated that for a given registration, a Dice similarity coefficient >0.80 and >0.60 for the prostate and PSV, respectively, and a maximum Hausdorff distance <8.0 mm for both structures were predictive of delivered dose ± 5% of planned dose. A normalized volume difference <1.0 and a CBCT anterior rectum wall >1.0 mm anterior to the planning CT anterior rectum wall were predictive of delivered dose >5% of planned rectum dose. A normalized volume difference <0.88, and a CBCT bladder wall >13.5 mm inferior and >5.0 mm posterior to the planning CT bladder were predictive of delivered dose >5% of planned bladder dose. A FASTV_OAR >0 is indicative of delivery of planned dose. For calculated FASTV_OAR for the prostate, PSV, bladder, and rectum using test data, sensitivity was 0.56, 0.75, 0.89, and 1.00, respectively; specificity 0.90, 0.94, 0.59, and 1.00, respectively; positive predictive power 0.90, 0.86, 0.53, and 1.00, respectively; and negative predictive power 0.56, 0.89, 0.91, and 1.00, respectively. Thresholds for the calculated FASglobal of were low <60, medium 60-80, and high >80, with a 27% misclassification rate for the test data. CONCLUSIONS: A FASglobal incorporating nested FASTV_OAR and volume variation metric thresholds predictive of treatment plan compliance was developed, offering an alternative to pretreatment dose calculations to assess treatment delivery accuracy.

Assessment of cone beam CT registration for prostate radiation therapy: fiducial marker and soft tissue methods.

INTRODUCTION: This investigation aimed to assess the consistency and accuracy of radiation therapists (RTs) performing cone beam computed tomography (CBCT) alignment to fiducial markers (FMs) (CBCTFM ) and the soft tissue prostate (CBCTST ). METHODS: Six patients receiving prostate radiation therapy underwent daily CBCTs. Manual alignment of CBCTFM and CBCTST was performed by three RTs. Inter-observer agreement was assessed using a modified Bland-Altman analysis for each alignment method. Clinically acceptable 95% limits of agreement with the mean (LoAmean ) were defined as ±2.0 mm for CBCTFM and ±3.0 mm for CBCTST . Differences between CBCTST alignment and the observer-averaged CBCTFM (AvCBCTFM ) alignment were analysed. Clinically acceptable 95% LoA were defined as ±3.0 mm for the comparison of CBCTST and AvCBCTFM . RESULTS: CBCTFM and CBCTST alignments were performed for 185 images. The CBCTFM 95% LoAmean were within ±2.0 mm in all planes. CBCTST 95% LoAmean were within ±3.0 mm in all planes. Comparison of CBCTST with AvCBCTFM resulted in 95% LoA of -4.9 to 2.6, -1.6 to 2.5 and -4.7 to 1.9 mm in the superior-inferior, left-right and anterior-posterior planes, respectively. CONCLUSIONS: Significant differences were found between soft tissue alignment and the predicted FM position. FMs are useful in reducing inter-observer variability compared with soft tissue alignment. Consideration needs to be given to margin design when using soft tissue matching due to increased inter-observer variability. This study highlights some of the complexities of soft tissue guidance for prostate radiation therapy.

Interobserver variability of radiation therapists aligning to fiducial markers for prostate radiation therapy.

INTRODUCTION: As the use of fiducial markers (FMs) for the localisation of the prostate during external beam radiation therapy (EBRT) has become part of routine practice, radiation therapists (RTs) have become increasingly responsible for online image interpretation. The aim of this investigation was to quantify the limits of agreement (LoA) between RTs when localising to FMs with orthogonal kilovoltage (kV) imaging. METHODS: Six patients receiving prostate EBRT utilising FMs were included in this study. Treatment localisation was performed using kV imaging prior to each fraction. Online stereoscopic assessment of FMs, performed by the treating RTs, was compared with the offline assessment by three RTs. Observer agreement was determined by pairwise Bland-Altman analysis. RESULTS: Stereoscopic analysis of 225 image pairs was performed online at the time of treatment, and offline by three RT observers. Eighteen pairwise Bland-Altman analyses were completed to assess the level of agreement between observers. Localisation by RTs was found to be within clinically acceptable 95% LoAs. CONCLUSIONS: Small differences between RTs, in both the online and offline setting, were found to be within clinically acceptable limits. RTs were able to make consistent and reliable judgements when matching FMs on planar kV imaging.

Quality Improvement Investigation for Head and Neck Stabilization in Radiotherapy Using Setup Tattoos.

PURPOSE: Highly complex planning techniques and delivery methods in the treatment of head and neck cancer require an advanced level of accuracy and reproducibility. AIM: To determine if the addition of tattoos placed on the chest inferior to the CIVCO Vac-Lok stabilization system improves accuracy and reproducibility of patient set up. METHODS: Eighteen patients with head and neck cancer were studied. Nine underwent radical treatment using the routine CIVCO stabilization system. The second group of nine used the same stabilization device but were positioned daily with the use of tattoos. Daily orthogonal kilovoltage setup images were used to calculate setup errors. Displacements in the left/right (Lt/Rt), superior/inferior (Sup/Inf), and anterior/posterior (Ant/Post) directions were determined as well as pitch and yaw rotational errors. RESULTS: Five hundred and twenty-three image pairs were analysed. Clinically significant differences were found in yaw error, Lt/Rt displacement, and Sup/Inf displacement in the tattooed patients. The median (interquartile range) absolute yaw error was larger for patients without tattoos: 1.4° (1.4° to 2.1°) compared to 0.8° (0.8° to 1.4°) for patients with tattoos. The percentage of both Sup/Inf and Lt/Rt errors >3 mm was also greater for patients without tattoos: 23.7% of Sup/Inf errors were >3 mm compared with 17.3% for patients with tattoos, and 22.3% of Lt/Rt errors were >3 mm compared with 10.0% for patients with tattoos. CONCLUSION: The addition of chest tattoos resulted in clinically relevant improvements in Lt/Rt and Sup/Inf translational displacements and variations in yaw for head and neck cancer patients.