P.E. Lindsay

Dose response explorer: an integrated open-source tool for exploring and modelling radiotherapy dose–volume outcome relationships

Radiotherapy treatment outcome models are a complicated function of treatment, clinical and biological factors. Our objective is to provide clinicians and scientists with an accurate, flexible and user-friendly software tool to explore radiotherapy outcomes data and build statistical tumour control or normal tissue complications models. The software tool, called the dose response explorer system (DREES), is based on Matlab, and uses a named-field structure array data type. DREES/Matlab in combination with another open-source tool (CERR) provides an environment for analysing treatment outcomes. DREES provides many radiotherapy outcome modelling features, including (1) fitting of analytical normal tissue complication probability (NTCP) and tumour control probability (TCP) models, (2) combined modelling of multiple dose-volume variables (e.g., mean dose, max dose, etc) and clinical factors (age, gender, stage, etc) using multi-term regression modelling, (3) manual or automated selection of logistic or actuarial model variables using bootstrap statistical resampling, (4) estimation of uncertainty in model parameters, (5) performance assessment of univariate and multivariate analyses using Spearmans rank correlation and chi-square statistics, boxplots, nomograms, Kaplan-Meier survival plots, and receiver operating characteristics curves, and (6) graphical capabilities to visualize NTCP or TCP prediction versus selected variable models using various plots. DREES provides clinical researchers with a tool customized for radiotherapy outcome modelling. DREES is freely distributed. We expect to continue developing DREES based on user feedback.

Retrospective Monte Carlo dose calculations with limited beam weight information

An important unresolved issue in outcomes analysis for lung complications is the effect of poor or completely lacking heterogeneity corrections in previously archived treatment plans. To estimate this effect, we developed a novel method based on Monte Carlo (MC) dose calculations which can be applied retrospectively to RTOG/AAPM-style archived treatment plans (ATP). We applied this method to 218 archived nonsmall cell lung cancer lung treatment plans that were originally calculated either without heterogeneity corrections or with primitive corrections. To retrospectively specify beam weights and wedges, beams were broken into Monte Carlo-generated beamlets, simulated using the VMC++ code, and mathematical optimization was used to match the archived water-based dose distributions. The derived beam weights (and any wedge effects) were then applied to Monte Carlo beamlets regenerated based on the patient computed tomography densities. Validation of the process was performed against five comparable lung treatment plans generated using a commercial convolution/superposition implementation. For the application here (normal lung, esophagus, and planning target volume dose distributions), the agreement was very good. Resulting MC and convolution/superposition values were similar when dose distributions without heterogeneity corrections or dose distributions with corrections were compared. When applied to the archived plans (218), the average absolute percent difference between water-based MC and water-based ATPs, for doses above 2.5% of the maximum dose was 1.8+/-0.6%. The average absolute percent difference between heterogeneity-corrected MC and water-based ATPs increased to 3.1+/-0.9%. The average absolute percent difference between the MC heterogeneity-corrected and the ATP heterogeneity-corrected dose distributions was 3.8+/-1.6% (available in 132/218 archives). The entire dose-volume-histograms for lung, tumor, and esophagus from the different calculation methods, as well as specific dose metrics, were compared. The average difference in maximum lung dose between water-based ATPs and heterogeneity-corrected MC dose distributions was -1.0+/-2.1 Gy. Potential errors in relying on primitive heterogeneity corrections are most evident from a comparison of maximum lung doses, for which the average MC heterogeneity-corrected values were 5.3+/-2.8 Gy less than the ATP heterogeneity-corrected values. We have demonstrated that recalculation of archived dose distributions, without explicit information about beam weights or wedges, is feasible using beamlet-based optimization methods. The method provides heterogeneity-corrected dose data consistent with convolution-superposition calculations and is one feasible approach for improving dosimetric data for outcomes analyses.

TU‐FF‐A1‐06: Monte Carlo Based Retrospective Dose Calculations for Outcomes Modeling

Purpose: To improve the dosimetric accuracy of archived lung treatment plans, we use a novel Monte Carlo recalculation method based on pencil beam optimization methods. The impact of the dose corrections on outcome modeling of pneumonitis was assessed. Method and Materials: For 189 archived non‐small cell lungcancer plans, dose distributions were re‐calculated using the VMC++ Monte Carlo code (I.Kawrakow). Nominal input spectra for 6 or 18 MV photons were used; only radiation transport through the patient was modeled, using each patients pre‐treatment CT scan. We derived approximate beam weights and wedge effects with a novel method based on optimization of MC‐derived pencil beams: MC and treatment planning results were matched for the water‐based (non‐heterogeneity corrected) results. Heterogeneity‐corrected plans were then produced using Monte Carlo with the derived beam profiles and weights. Results: The method showed good agreement when compared against a small series of treatment plans using a convolution‐superposition dose calculation. For the lung plans, the average absolute differences in metrics of interest (V20, maximum lungdose, and mean GTV dose) between water‐based TPS and water‐based MC data were 0.5%, 0.9 Gy, and 0.8 Gy; for water‐based TPS versus heterogeneity‐corrected MC data the absolute differences were greater: 2.0%, 1.8 Gy, and 2.5 Gy (typically heterogeneity corrected dose distributions produced higher dose values). The correlations between V20 and occurrence of pneumonitis for water‐based TPS, water‐based MC, and heterogeneity corrected MC data were (using Spearmans rank correlation coefficient) 0.13, 0.13, and 0.14 (respectively). For maximum lungdose, the correlations were 0.15, 0.14, and 0.09. Conclusion: The differences in some metrics (e.g., maximum lungdose) between water‐based and heterogeneity corrected data may have a significant impact on modeling treatment outcome. This method could be applied to any multi‐institutional data sets for which RTOG format plan archives are available.

MO‐D‐T‐6E‐02: Dose‐Response Explorer: An Open‐Source‐Code Matlab‐Based Tool for Modeling Treatment Outcome as a Function of Predictive Factors

Purpose:Radiotherapy treatment outcome models are a complicated function of treatment parameters and/or clinical factors. Our objective is to provide clinicians and scientists with an accurate, flexible, and user‐friendly tool to explore radiotherapy outcome models with different factors leading to tumorcontrol or normal tissue complications. We refer to this tool as the dose response explorer (DREX). Method and Materials: DREX, based on Matlab named‐field structures, provides tools for multi‐term logistic regression modeling, correlation calculations, and graphical comparisons between model predictions and observations. A GUI‐driven interface was constructed using Matlab tools. Named‐field structures in Matlab support development of very human‐readable databasesResults: The DREX tool provides the NTCP or TCP analyst with multiple features which include: (1) Combination of multiple dose‐volume variables (mean dose, max dose, Vx (percentage volume receiving × Gy), Dx (dose to × percentage volume), EUD (equivalent uniform dose), etc) and clinical factors (age, gender, ethnicity, etc), (2) Modelanalysis using logistic regression, (3) Performance assessment using Spearmans rank correlation and receiver operating characteristics (ROC) curves, and (4) Graphical capability to visualize NTCP or TCP prediction versus selected variable model using contour and histogram plots. DREX has been in constant use in our research group for the last nine months. Conclusion: We developed user‐friendly software to explore and modelradiotherapydose‐response correlations. DREX facilitates convenient study of different treatment and clinical factors which may correlate with complication or control. We believe that the DREX software combined with CERR archiving would provide the clinical researcher with convenient tools to accrue and modelradiotherapy outcomes data. DREX will be freely distributed via the web. We expect to continue developing DREX, including adding methods to automatically select model terms, find the optimal model size, and estimate parameter uncertainties.

SU‐FF‐T‐376: Multi‐Variable Modeling of Radiotherapy Outcomes: Determining Optimal Model Size

Purpose: The probability of a specified radiotherapy outcome (e.g., a normal tissue complication or tumor eradication) is typically a complex, non‐linear, unknown function of dose distribution characteristics and clinical factors (such as chemotherapy, age, gender, diabetes, etc.). However, current outcome models are usually over‐simplified, and standard model fitting methods give little guidance as to how to best add choose from many complicated, alternative models. We discuss methods for building multivariable response models within the framework of logistic regression. We study in detail the issue of how to select model complexity to reach the goal of maximizing predictive power. Method and Materials:Analyses of esophagitis and xerostomia datasets are used as examples. We describe techniques for approximating the unknown dose‐volume‐responsefunction as a linear combination of multiple candidate dosimetric variables and clinical factors. In order to guard against under‐ and over‐fitting, we compare several methods for selecting optimal model size, including: fitting against bootstrap training and testing datasets, Akaike information criteria, and leave‐one‐out cross validation. Results: Leave‐one‐out cross validation produced the most unambiguous guidance for optimal model size. Optimal esophagitis model size was five variables (concurrent chemotherapy, A55, A30, A45, A85). Although the xerostomia model could be improved using clinical factors, the improvement over using the single dose‐volume model term was small, and therefore judged not worth the added complexity. Conclusions. Treatment response models, including dose‐volume effects, can be made more predictive by mixing clinical and multiple dose‐volume factors into a single model. Over‐simplified treatment response models are only justified in those cases where more complicated models cannot be supported by the data. Leave‐one‐out cross correlation model testing combined with Spearmans correlation coefficient often provided the least ambiguous method to study the tradeoff between prediction improvements and model size and to choose optimal model size.

SU-FF-T-360: Software Tools for 4-D and Adaptive Treatment Planning Data Visualization and Manipulation (CERR Version 3)

Purpose: Open‐source tools are needed to facilitate the use of multiple imaging datasets for adaptive and 4‐D treatment planning. Commercial systems do not yet provide effective solutions for reviewing, manipulating, and comparing multiple scan datasets taken at different times of with different imaging modalities. Our research treatment planning system, CERR (“computational environment for radiotherapy research,” pronounced ‘sir’), provides a convenient and powerful basis for constructing adaptive and 4‐D treatment planning tools. Method and Materials: The flexible Matlab‐based system CERR was modified and extended. A fast algorithm for image registration was developed and integrated with CERR. Results: We have added support in CERR for: (1) multiple patient image sets, which can be combinations of CT, MRI, or PET scans, and corresponding anatomical structure datasets, (2) rapid image registration (by hand as well as by using an automated method), (3) visualization tools appropriate to review anatomic structure changes with different scan sets, and (4) linked‐storage of multiple scan sets, eliminating memory limitations. In addition, many smaller features have been added to CERR, such as interactive profile plots of dose and/or image values. This version of CERR also provides support for IMRT treatment planning simulations. Local data storage in a sub‐directory or network scan data storage we integrated. Flexible reporting tools allow for structures defined on one dataset to be combined with dose distributions computed for another scan, and resulting dose‐volume‐histograms can be easily derived. The latest release can be downloaded from radium.wustl.edu/cerr. Conclusion: CERR version 3 provides foundational tools for research in adaptive and 4‐ D treatment planning. This framework provides a powerful basis for experimenting with deformable imaging methods, as well as other adaptive radiotherapy challenges, such as re‐optimization research.