Methodological issues in radiation dose-volume outcome analyses: summary of a joint AAPM/NIH workshop

Development and validation of survival prognostic models for head and neck cancer patients using machine learning and dosiomics and CT radiomics features: a multicentric study

BACKGROUND

This study aimed to investigate the value of clinical, radiomic features extracted from gross tumor volumes (GTVs) delineated on CT images, dose distributions (Dosiomics), and fusion of CT and dose distributions to predict outcomes in head and neck cancer (HNC) patients.

METHODS

A cohort of 240 HNC patients from five different centers was obtained from The Cancer Imaging Archive. Seven strategies, including four non-fusion (Clinical, CT, Dose, DualCT-Dose), and three fusion algorithms (latent low-rank representation referred (LLRR),Wavelet, weighted least square (WLS)) were applied. The fusion algorithms were used to fuse the pre-treatment CT images and 3-dimensional dose maps. Overall, 215 radiomics and Dosiomics features were extracted from the GTVs, alongside with seven clinical features incorporated. Five feature selection (FS) methods in combination with six machine learning (ML) models were implemented. The performance of the models was quantified using the concordance index (CI) in one-center-leave-out 5-fold cross-validation for overall survival (OS) prediction considering the time-to-event.

RESULTS

The mean CI and Kaplan-Meier curves were used for further comparisons. The CoxBoost ML model using the Minimal Depth (MD) FS method and the glmnet model using the Variable hunting (VH) FS method showed the best performance with CI = 0.73 ± 0.15 for features extracted from LLRR fused images. In addition, both glmnet-Cindex and Coxph-Cindex classifiers achieved a CI of 0.72 ± 0.14 by employing the dose images (+ incorporated clinical features) only.

CONCLUSION

Our results demonstrated that clinical features, Dosiomics and fusion of dose and CT images by specific ML-FS models could predict the overall survival of HNC patients with acceptable accuracy. Besides, the performance of ML methods among the three different strategies was almost comparable.

Fractionation versus Adaptation for Compensation of Target Volume Changes during Online Adaptive Radiotherapy for Bladder Cancer: Answers from a Prospective Registry

Online adaptive radiotherapy (ART) allows adaptation of the dose distribution to the anatomy captured by with pre-adaptation imaging. ART is time-consuming, and thus intra-fractional deformations can occur. This prospective registry study analyzed the effects of intra-fraction deformations of clinical target volume (CTV) on the equivalent uniform dose (EUDCTV) of focal bladder cancer radiotherapy. Using margins of 5-10 mm around CTV on pre-adaptation imaging, intra-fraction CTV-deformations found in a second imaging study reduced the 10th percentile of EUDCTV values per fraction from 101.1% to 63.2% of the prescribed dose. Dose accumulation across fractions of a series was determined with deformable-image registration and worst-case dose accumulation that maximizes the correlation of cold spots. A strong fractionation effect was demonstrated-the EUDCTV was above 95% and 92.5% as determined by the two abovementioned accumulation methods, respectively, for all series of dose fractions. A comparison of both methods showed that the fractionation effect caused the EUDCTV of a series to be insensitive to EUDCTV-declines per dose fraction, and this could be explained by the small size and spatial variations of cold spots. Therefore, ART for each dose fraction is unnecessary, and selective ART for fractions with large inter-fractional deformations alone is sufficient for maintaining a high EUDCTV for a radiotherapy series.

Generalized approach for radiotherapy treatment planning by optimizing projected health outcome: preliminary results for prostate radiotherapy patients

Research in cancer care increasingly focuses on survivorship issues, e.g. managing disease- and treatment-related morbidity and mortality occurring during and after treatment. This necessitates innovative approaches that consider treatment side effects in addition to tumor cure. Current treatment-planning methods rely on constrained iterative optimization of dose distributions as a surrogate for health outcomes. The goal of this study was to develop a generally applicable method to directly optimize projected health outcomes. We developed an outcome-based objective function to guide selection of the number, angle, and relative fluence weight of photon and proton radiotherapy beams in a sample of ten prostate-cancer patients by optimizing the projected health outcome. We tested whether outcome-optimized radiotherapy (OORT) improved the projected longitudinal outcome compared to dose-optimized radiotherapy (DORT) first for a statistically significant majority of patients, then for each individual patient. We assessed whether the results were influenced by the selection of treatment modality, late-risk model, or host factors. The results of this study revealed that OORT was superior to DORT. Namely, OORT maintained or improved the projected health outcome of photon- and proton-therapy treatment plans for all ten patients compared to DORT. Furthermore, the results were qualitatively similar across three treatment modalities, six late-risk models, and 10 patients. The major finding of this work was that it is feasible to directly optimize the longitudinal (i.e. long- and short-term) health outcomes associated with the total (i.e. therapeutic and stray) absorbed dose in all of the tissues (i.e. healthy and diseased) in individual patients. This approach enables consideration of arbitrary treatment factors, host factors, health endpoints, and times of relevance to cancer survivorship. It also provides a simpler, more direct approach to realizing the full beneficial potential of cancer radiotherapy.

Spatial and dosimetric evaluation of residual distortions of prostate and seminal vesicle bed after image-guided definitive and postoperative radiotherapy of prostate cancer with endorectal balloon

Purpose

To quantify daily residual deviations from the planned geometry after image‐guided prostate radiotherapy with endorectal balloon and to evaluate their effect on the delivered dose distribution.

Methods

Daily kV‐CBCT imaging was used for online setup‐correction in six degrees of freedom (6‐dof) for 24 patients receiving definitive (12 RT_def patients) or postoperative (12 RT_postop patients) radiotherapy with endorectal balloon (overall 739 CBCTs). Residual deviations were evaluated using several spatial and dosimetric variables, including: (a) posterior Hausdorff distance HD^post (=maximum distance between planned and daily CTV contour), (b) point P_worst with largest HD^post over all fractions, (c) equivalent uniform dose using a cell survival model (EUD_SF) and the generalized EUD concept (gEUD_a with parameter a = −7 and a = −20). EUD values were determined for planned (EUDSFplan), daily (EUDSFind), and delivered dose distributions (EUDSFaccum) for plans with 6 mm (=clinical plans) and 2 mm CTV‐to‐PTV margin. Time series analyses of interfractional spatial and dosimetric deviations were conducted.

Results

Large HD^post values ≥ 12.5 mm (≥15 mm) were observed in 20/739 (5/739) fractions distributed across 7 (3) patients. Points P_worst were predominantly located at the posterior CTV boundary in the seminal vesicle region (16/24 patients, 6/7 patients with HD^post ≥ 12.5 mm). Time series analyses revealed a stationary white noise characteristic of HD^post and relative dose at P_worst. The EUD_SF difference between planned and accumulated dose distributions was < 5.4% for all 6‐mm plans. Evaluating 2‐mm plans, EUD_SF deteriorated by < 10% (<5%) in 75% (58.5%) of the patients. EUDSFaccum was well described by the median value of the EUDSFind distribution. PTV margin calculation at P_worst yielded 8.8 mm.

Conclusions

Accumulated dose distributions in prostate radiotherapy with endorectal balloon are forgiving of considerable residual distortions after 6‐dof patient setup if they are observed in a minority of fractions and the median value of EUDSFind determined per fraction stays within 95% of prescribed dose. Common PTV margin calculations are overly conservative because after online correction of translational and rotational errors only residual deformations need to be included. These results provide guidelines regarding online navigation, margin optimization, and treatment adaptation strategies.

Multi-view radiomics and dosiomics analysis with machine learning for predicting acute-phase weight loss in lung cancer patients treated with radiotherapy

We propose a multi-view data analysis approach using radiomics and dosiomics (R&D) texture features for predicting acute-phase weight loss (WL) in lung cancer radiotherapy. Baseline weight of 388 patients who underwent intensity modulated radiation therapy (IMRT) was measured between one month prior to and one week after the start of IMRT. Weight change between one week and two months after the commencement of IMRT was analyzed, and dichotomized at 5% WL. Each patient had a planning CT and contours of gross tumor volume (GTV) and esophagus (ESO). A total of 355 features including clinical parameter (CP), GTV and ESO (GTV&ESO) dose-volume histogram (DVH), GTV radiomics, and GTV&ESO dosiomics features were extracted. R&D features were categorized as first- (L1), second- (L2), higher-order (L3) statistics, and three combined groups, L1 + L2, L2 + L3 and L1 + L2 + L3. Multi-view texture analysis was performed to identify optimal R&D input features. In the training set (194 earlier patients), feature selection was performed using Boruta algorithm followed by collinearity removal based on variance inflation factor. Machine-learning models were developed using Laplacian kernel support vector machine (lpSVM), deep neural network (DNN) and their averaged ensemble classifiers. Prediction performance was tested on an independent test set (194 more recent patients), and compared among seven different input conditions: CP-only, DVH-only, R&D-only, DVH + CP, R&D + CP, R&D + DVH and R&D + DVH + CP. Combined GTV L1 + L2 + L3 radiomics and GTV&ESO L3 dosiomics were identified as optimal input features, which achieved the best performance with an ensemble classifier (AUC = 0.710), having statistically significantly higher predictability compared with DVH and/or CP features (p < 0.05). When this performance was compared to that with full R&D-only features which reflect traditional single-view data, there was a statistically significant difference (p < 0.05). Using optimized multi-view R&D input features is beneficial for predicting early WL in lung cancer radiotherapy, leading to improved performance compared to using conventional DVH and/or CP features.

An extended dose-volume model in high dose-rate brachytherapy - Using mean-tail-dose to reduce tumor underdosage

Purpose

High dose–rate brachytherapy is a method of radiotherapy for cancer treatment in which the radiation source is placed within the body. In addition to give a high enough dose to a tumor, it is also important to spare nearby healthy organs [organs at risk (OAR)]. Dose plans are commonly evaluated using the so‐called dosimetric indices; for the tumor, the portion of the structure that receives a sufficiently high dose is calculated, while for OAR it is instead the portion of the structure that receives a sufficiently low dose that is of interest. Models that include dosimetric indices are referred to as dose–volume models (DVMs) and have received much interest recently. Such models do not take the dose to the coldest (least irradiated) volume of the tumor into account, which is a distinct weakness since research indicates that the treatment effect can be largely impaired by tumor underdosage even to small volumes. Therefore, our aim is to extend a DVM to also consider the dose to the coldest volume.

Methods

An improved DVM for dose planning is proposed. In addition to optimizing with respect to dosimetric indices, this model also takes mean dose to the coldest volume of the tumor into account.

Results

Our extended model has been evaluated against a standard DVM in ten prostate geometries. Our results show that the dose to the coldest volume could be increased, while also computing times for the dose planning were improved.

Conclusion

While the proposed model yields dose plans similar to other models in most aspects, it fulfils its purpose of increasing the dose to cold tumor volumes. An additional benefit is shorter solution times, and especially for clinically relevant times (of minutes) we show major improvements in tumour dosimetric indices.

Expert system classifier for adaptive radiation therapy in prostate cancer

A classifier-based expert system was developed to compare delivered and planned radiation therapy in prostate cancer patients. Its aim is to automatically identify patients that can benefit from an adaptive treatment strategy. The study predominantly addresses dosimetric uncertainties and critical issues caused by motion of hollow organs. 1200 MVCT images of 38 prostate adenocarcinoma cases were analyzed. An automatic daily re-contouring of structures (i.e. rectum, bladder and femoral heads), rigid/deformable registration and dose warping was carried out to simulate dose and volume variations during therapy. Support vector machine, K-means clustering algorithms and similarity index analysis were used to create an unsupervised predictive tool to detect incorrect setup and/or morphological changes as a consequence of inadequate patient preparation due to stochastic physiological changes, supporting clinical decision-making. After training on a dataset that was considered sufficiently dosimetrically stable, the system identified two equally sized macro clusters with distinctly different volumetric and dosimetric baseline properties and defined thresholds for these two clusters. Application to the test cohort resulted in 25% of the patients located outside the two macro clusters thresholds and which were therefore suspected to be dosimetrically unstable. In these patients, over the treatment course, mean volumetric changes of 30 and 40% for rectum and bladder were detected which possibly represents values justifying adjustment of patient preparation, frequent re-planning or a plan-of-the-day strategy. Based on our research, by combining daily IGRT images with rigid/deformable registration and dose warping, it is possible to apply a machine learning approach to the clinical setting obtaining useful information for a decision regarding an individualized adaptive strategy. Especially for treatments influenced by the movement of hollow organs, this could reduce inadequate treatments and possibly reduce toxicity, thereby increasing overall RT efficacy.

Monte Carlo role in radiobiological modelling of radiotherapy outcomes

Radiobiological models are essential components of modern radiotherapy. They are increasingly applied to optimize and evaluate the quality of different treatment planning modalities. They are frequently used in designing new radiotherapy clinical trials by estimating the expected therapeutic ratio of new protocols. In radiobiology, the therapeutic ratio is estimated from the expected gain in tumour control probability (TCP) to the risk of normal tissue complication probability (NTCP). However, estimates of TCP/NTCP are currently based on the deterministic and simplistic linear-quadratic formalism with limited prediction power when applied prospectively. Given the complex and stochastic nature of the physical, chemical and biological interactions associated with spatial and temporal radiation induced effects in living tissues, it is conjectured that methods based on Monte Carlo (MC) analysis may provide better estimates of TCP/NTCP for radiotherapy treatment planning and trial design. Indeed, over the past few decades, methods based on MC have demonstrated superior performance for accurate simulation of radiation transport, tumour growth and particle track structures; however, successful application of modelling radiobiological response and outcomes in radiotherapy is still hampered with several challenges. In this review, we provide an overview of some of the main techniques used in radiobiological modelling for radiotherapy, with focus on the MC role as a promising computational vehicle. We highlight the current challenges, issues and future potentials of the MC approach towards a comprehensive systems-based framework in radiobiological modelling for radiotherapy.

A Bayesian network approach for modeling local failure in lung cancer

Locally advanced non-small cell lung cancer (NSCLC) patients suffer from a high local failure rate following radiotherapy. Despite many efforts to develop new dose-volume models for early detection of tumor local failure, there was no reported significant improvement in their application prospectively. Based on recent studies of biomarker proteins' role in hypoxia and inflammation in predicting tumor response to radiotherapy, we hypothesize that combining physical and biological factors with a suitable framework could improve the overall prediction. To test this hypothesis, we propose a graphical Bayesian network framework for predicting local failure in lung cancer. The proposed approach was tested using two different datasets of locally advanced NSCLC patients treated with radiotherapy. The first dataset was collected retrospectively, which comprises clinical and dosimetric variables only. The second dataset was collected prospectively in which in addition to clinical and dosimetric information, blood was drawn from the patients at various time points to extract candidate biomarkers as well. Our preliminary results show that the proposed method can be used as an efficient method to develop predictive models of local failure in these patients and to interpret relationships among the different variables in the models. We also demonstrate the potential use of heterogeneous physical and biological variables to improve the model prediction. With the first dataset, we achieved better performance compared with competing Bayesian-based classifiers. With the second dataset, the combined model had a slightly higher performance compared to individual physical and biological models, with the biological variables making the largest contribution. Our preliminary results highlight the potential of the proposed integrated approach for predicting post-radiotherapy local failure in NSCLC patients.

A bioinformatics approach for biomarker identification in radiation-induced lung inflammation from limited proteomics data

Many efforts have been made to discover novel bio-markers for early disease detection in oncology. However, the lack of efficient computational strategies impedes the discovery of disease-specific biomarkers for better understanding and management of treatment outcomes. In this study, we propose a novel graph-based scoring function to rank and identify the most robust biomarkers from limited proteomics data. The proposed method measures the proximity between candidate proteins identified by mass spectrometry (MS) analysis utilizing prior reported knowledge in the literature. Recent advances in mass spectrometry provide new opportunities to identify unique biomarkers from peripheral blood samples in complex treatment modalities such as radiation therapy (radiotherapy), which enables early disease detection, disease progression monitoring, and targeted intervention. Specifically, the dose-limiting role of radiation-induced lung injury known as radiation pneumonitis (RP) in lung cancer patients receiving radiotherapy motivates the search for robust predictive biomarkers. In this case study, plasma from 26 locally advanced non-small cell lung cancer (NSCLC) patients treated with radiotherapy in a longitudinal 3 × 3 matched-control cohort was fractionated using in-line, sequential multiaffinity chromatography. The complex peptide mixtures from endoprotease digestions were analyzed using comparative, high-resolution liquid chromatography (LC)-MS to identify and quantify differential peptide signals. Through analysis of survey mass spectra and annotations of peptides from the tandem spectra, we found candidate proteins that appear to be associated with RP. On the basis of the proposed methodology, α-2-macroglobulin (α2M) was unambiguously ranked as the top candidate protein. As independent validation of this candidate protein, enzyme-linked immunosorbent assay (ELISA) experiments were performed on independent cohort of 20 patients' samples resulting in early significant discrimination between RP and non-RP patients (p = 0.002). These results suggest that the proposed methodology based on longitudinal proteomics analysis and a novel bioinformatics ranking algorithm is a potentially promising approach for the challenging problem of identifying relevant biomarkers in sample-limited clinical applications.

Datamining approaches for modeling tumor control probability

BACKGROUND

Tumor control probability (TCP) to radiotherapy is determined by complex interactions between tumor biology, tumor microenvironment, radiation dosimetry, and patient-related variables. The complexity of these heterogeneous variable interactions constitutes a challenge for building predictive models for routine clinical practice. We describe a datamining framework that can unravel the higher order relationships among dosimetric dose-volume prognostic variables, interrogate various radiobiological processes, and generalize to unseen data before when applied prospectively.

MATERIAL AND METHODS

Several datamining approaches are discussed that include dose-volume metrics, equivalent uniform dose, mechanistic Poisson model, and model building methods using statistical regression and machine learning techniques. Institutional datasets of non-small cell lung cancer (NSCLC) patients are used to demonstrate these methods. The performance of the different methods was evaluated using bivariate Spearman rank correlations (rs). Over-fitting was controlled via resampling methods.

RESULTS

Using a dataset of 56 patients with primary NCSLC tumors and 23 candidate variables, we estimated GTV volume and V75 to be the best model parameters for predicting TCP using statistical resampling and a logistic model. Using these variables, the support vector machine (SVM) kernel method provided superior performance for TCP prediction with an rs=0.68 on leave-one-out testing compared to logistic regression (rs=0.4), Poisson-based TCP (rs=0.33), and cell kill equivalent uniform dose model (rs=0.17).

CONCLUSIONS

The prediction of treatment response can be improved by utilizing datamining approaches, which are able to unravel important non-linear complex interactions among model variables and have the capacity to predict on unseen data for prospective clinical applications.

Quantitative Analyses of Normal Tissue Effects in the Clinic (QUANTEC): an introduction to the scientific issues

Advances in dose-volume/outcome (or normal tissue complication probability, NTCP) modeling since the seminal Emami paper from 1991 are reviewed. There has been some progress with an increasing number of studies on large patient samples with three-dimensional dosimetry. Nevertheless, NTCP models are not ideal. Issues related to the grading of side effects, selection of appropriate statistical methods, testing of internal and external model validity, and quantification of predictive power and statistical uncertainty, all limit the usefulness of much of the published literature. Synthesis (meta-analysis) of data from multiple studies is often impossible because of suboptimal primary analysis, insufficient reporting and variations in the models and predictors analyzed. Clinical limitations to the current knowledge base include the need for more data on the effect of patient-related cofactors, interactions between dose distribution and cytotoxic or molecular targeted agents, and the effect of dose fractions and overall treatment time in relation to nonuniform dose distributions. Research priorities for the next 5-10 years are proposed.

Use of normal tissue complication probability models in the clinic

The Quantitative Analysis of Normal Tissue Effects in the Clinic (QUANTEC) review summarizes the currently available three-dimensional dose/volume/outcome data to update and refine the normal tissue dose/volume tolerance guidelines provided by the classic Emami et al. paper published in 1991. A "clinician's view" on using the QUANTEC information in a responsible manner is presented along with a description of the most commonly used normal tissue complication probability (NTCP) models. A summary of organ-specific dose/volume/outcome data, based on the QUANTEC reviews, is included.

NTCP modeling of subacute/late laryngeal edema scored by fiberoptic examination

PURPOSE

Finding best-fit parameters of normal tissue complication probability (NTCP) models for laryngeal edema after radiotherapy for head and neck cancer.

METHODS AND MATERIALS

Forty-eight patients were considered for this study who met the following criteria: (1) grossly uninvolved larynx, (2) no prior major surgery except for neck dissection and tonsillectomy, (3) at least one fiberoptic examination of the larynx within 2 years from radiotherapy, (4) minimum follow-up of 15 months. Larynx dose-volume histograms (DVHs) were corrected into a linear quadratic equivalent one at 2 Gy/fr with alpha/beta = 3 Gy. Subacute/late edema was prospectively scored at each follow-up examination according to the Radiation Therapy Oncology Group scale. G2-G3 edema within 15 months from RT was considered as our endpoint. Two NTCP models were considered: (1) the Lyman model with DVH reduced to the equivalent uniform dose (EUD; LEUD) and (2) the Logit model with DVH reduced to the EUD (LOGEUD). The parameters for the models were fit to patient data using a maximum likelihood analysis.

RESULTS

All patients had a minimum of 15 months follow-up (only 8/48 received concurrent chemotherapy): 25/48 (52.1%) experienced G2-G3 edema. Both NTCP models fit well the clinical data: with LOGEUD the relationship between EUD and NTCP can be described with TD50 = 46.7 +/- 2.1 Gy, n = 1.41 +/- 0.8 and a steepness parameter k = 7.2 +/- 2.5 Gy. Best fit parameters for LEUD are n = 1.17 +/- 0.6, m = 0.23 +/- 0.07 and TD50 = 47.3 +/- 2.1 Gy.

CONCLUSIONS

A clear volume effect was found for edema, consistent with a parallel architecture of the larynx for this endpoint. On the basis of our findings, an EUD <30-35 Gy should drastically reduce the risk of G2-G3 edema.

Volumetric arc intensity-modulated therapy for spine body radiotherapy: comparison with static intensity-modulated treatment

PURPOSE

This clinical study evaluates the feasibility of using volumetric arc-modulated treatment (VMAT) for spine stereotactic body radiotherapy (SBRT) to achieve highly conformal dose distributions that spare adjacent organs at risk (OAR) with reduced treatment time.

METHODS AND MATERIALS

Ten spine SBRT patients were studied retrospectively. The intensity-modulated radiotherapy (IMRT) and VMAT plans were generated using either one or two arcs. Planning target volume (PTV) dose coverage, OAR dose sparing, and normal tissue integral dose were measured and compared. Differences in treatment delivery were also analyzed.

RESULTS

The PTV DVHs were comparable between VMAT and IMRT plans in the shoulder (D(99%)-D(90%)), slope (D(90%)-D(10%)), and tail (D(10%)-D(1%)) regions. Only VMAT(2arc) had a better conformity index than IMRT (1.09 vs. 1.15, p = 0.007). For cord sparing, IMRT was the best, and VMAT(1arc) was the worst. Use of IMRT achieved greater than 10% more D(1%) sparing for six of 10 cases and 7% to 15% more D(10%) sparing over the VAMT(1arc). The differences between IMRT and VAMT(2arc) were smaller and statistically nonsignificant at all dose levels. The differences were also small and statistically nonsignificant for other OAR sparing. The mean monitor units (MUs) were 8711, 7730, and 6317 for IMRT, VMAT(1arc), and VMAT(2arc) plans, respectively, with a 26% reduction from IMRT to VMAT(2arc). The mean treatment time was 15.86, 8.56, and 7.88 min for IMRT, VMAT(1arc,) and VMAT(2arc). The difference in integral dose was statistically nonsignificant.

CONCLUSIONS

Although VMAT provided comparable PTV coverage for spine SBRT, 1arc showed significantly worse spinal cord sparing compared with IMRT, whereas 2arc was comparable to IMRT. Treatment efficiency is substantially improved with the VMAT.

Predicting radiotherapy outcomes using statistical learning techniques

Radiotherapy outcomes are determined by complex interactions between treatment, anatomical and patient-related variables. A common obstacle to building maximally predictive outcome models for clinical practice is the failure to capture potential complexity of heterogeneous variable interactions and applicability beyond institutional data. We describe a statistical learning methodology that can automatically screen for nonlinear relations among prognostic variables and generalize to unseen data before. In this work, several types of linear and nonlinear kernels to generate interaction terms and approximate the treatment-response function are evaluated. Examples of institutional datasets of esophagitis, pneumonitis and xerostomia endpoints were used. Furthermore, an independent RTOG dataset was used for 'generalizabilty' validation. We formulated the discrimination between risk groups as a supervised learning problem. The distribution of patient groups was initially analyzed using principle components analysis (PCA) to uncover potential nonlinear behavior. The performance of the different methods was evaluated using bivariate correlations and actuarial analysis. Over-fitting was controlled via cross-validation resampling. Our results suggest that a modified support vector machine (SVM) kernel method provided superior performance on leave-one-out testing compared to logistic regression and neural networks in cases where the data exhibited nonlinear behavior on PCA. For instance, in prediction of esophagitis and pneumonitis endpoints, which exhibited nonlinear behavior on PCA, the method provided 21% and 60% improvements, respectively. Furthermore, evaluation on the independent pneumonitis RTOG dataset demonstrated good generalizabilty beyond institutional data in contrast with other models. This indicates that the prediction of treatment response can be improved by utilizing nonlinear kernel methods for discovering important nonlinear interactions among model variables. These models have the capacity to predict on unseen data.

Bioinformatics methods for learning radiation-induced lung inflammation from heterogeneous retrospective and prospective data

Radiotherapy outcomes are determined by complex interactions between physical and biological factors, reflecting both treatment conditions and underlying genetics. Recent advances in radiotherapy and biotechnology provide new opportunities and challenges for predicting radiation-induced toxicities, particularly radiation pneumonitis (RP), in lung cancer patients. In this work, we utilize datamining methods based on machine learning to build a predictive model of lung injury by retrospective analysis of treatment planning archives. In addition, biomarkers for this model are extracted from a prospective clinical trial that collects blood serum samples at multiple time points. We utilize a 3-way proteomics methodology to screen for differentially expressed proteins that are related to RP. Our preliminary results demonstrate that kernel methods can capture nonlinear dose-volume interactions, but fail to address missing biological factors. Our proteomics strategy yielded promising protein candidates, but their role in RP as well as their interactions with dose-volume metrics remain to be determined.

Analyzing adjuvant radiotherapy suggests a non monotonic radio-sensitivity over tumor volumes

BACKGROUND

Adjuvant Radiotherapy (RT) after surgical removal of tumors proved beneficial in long-term tumor control and treatment planning. For many years, it has been well concluded that radio-sensitivities of tumors upon radiotherapy decrease according to the sizes of tumors and RT models based on Poisson statistics have been used extensively to validate clinical data.

RESULTS

We found that Poisson statistics on RT is actually derived from bacterial cells despite of many validations from clinical data. However cancerous cells do have abnormal cellular communications and use chemical messengers to signal both surrounding normal and cancerous cells to develop new blood vessels and to invade, to metastasis and to overcome intercellular spatial confinements in general. We therefore investigated the cell killing effects on adjuvant RT and found that radio-sensitivity is actually not a monotonic function of volume as it was believed before. We present detailed analysis and explanation to justify above statement. Based on EUD, we present an equivalent radio-sensitivity model.

CONCLUSION

We conclude that radio sensitivity is a sophisticated function over tumor volumes, since tumor responses upon radio therapy also depend on cellular communications.

Physical models and simpler dosimetric descriptors of radiation late toxicity

Predicting radiation damage to specific organs is becoming ever more challenging with the use of intensity-modulated beams, nonuniform dose distributions, partial organ irradiation, and interpatient and even intraorgan variations in radiation sensitivity. Data-based physical models can be of use in summarizing complicated dose-volume data to help describe clinical outcomes and ultimately aid in the prediction of clinical toxicity. This article attempts to provide a brief overview of the use of normal tissue complication probability (NTCP) models and other simple dose/volume metrics to describe a few clinically significant complications (either frequent or serious) associated with radiation therapy of the head and neck, thorax, and abdominal-pelvic regions. Specifically, it reviews the application of these methods for late toxicities of the parotid, lung, heart, spinal cord, liver, and rectum. It focuses on organ-specific NTCP parameters as well as simple dosimetric descriptors that might be used to help treatment plan evaluation in clinical practice.

Relationships between rectal wall dose-volume constraints and radiobiologic indices of toxicity for patients with prostate cancer

PURPOSE

The purpose of this article was to investigate how exceeding specified rectal wall dose-volume constraints impacts on the risk of late rectal bleeding by using radiobiologic calculations.

METHODS AND MATERIALS

Dose-volume histograms (DVH) of the rectal wall of 250 patients with prostate cancer were analyzed. All patients were treated by three-dimensional conformal radiation therapy, receiving mean target doses of 80 Gy. To study the main features of the patient population, the average and the standard deviation of the distribution of DVHs were generated. The mean dose <D>, generalized equivalent uniform dose formulation (gEUD), modified equivalent uniform dose formulation (mEUD)(0), and normal tissue complication probability (NTCP) distributions were also produced. The DVHs set was then binned into eight classes on the basis of the exceeding or the fulfilling of three dose-volume constraints: V(40) = 60%, V(50) = 50%, and V(70) = 25%. Comparisons were made between them by <D>, gEUD, mEUD(0), and NTCP.

RESULTS

The radiobiologic calculations suggest that late rectal toxicity is mostly influenced by V(70). The gEUD and mEUD(0) are risk factors of toxicity always concordant with NTCP, inside each DVH class. The mean dose, although a reliable index, may be misleading in critical situations.

CONCLUSIONS

Both in three-dimensional conformal radiation therapy and particularly in intensity-modulated radiation therapy, it should be known what the relative importance of each specified dose-volume constraint is for each organ at risk. This requires a greater awareness of radiobiologic properties of tissues and radiobiologic indices may help to gradually become aware of this issue.

Early clinical and radiological pulmonary complications following breast cancer radiation therapy: NTCP fit with four different models

OBJECTIVE

To fit four different NTCP (Normal Tissue Complication Probability) models to prospectively collected data on short-term pulmonary complications following breast cancer radiotherapy (RT).

MATERIALS/METHODS

Four hundred and seventy-five breast cancer patients, referred to the Radiotherapy Department at Stockholm Söder Hospital (1994-1998) for adjuvant post-operative RT were prospectively followed for pulmonary complications 1, 4 and 7 months after the completion of RT. Eighty-seven patients with complete dose-volume histogram (DVH) of the ipsilateral lung were selected for the present analysis. Mean dose to the ipsilateral lateral lung ranged from 2.5 to 18Gy (median 12Gy). Three different endpoints were considered: (1) clinical pneumonitis scored according to CTC-NCIC criteria: asymptomatic (grade 0) vs grade 1 and grade 2; (2) radiological changes assessed with diagnostic chest X-ray: no/slight radiological changes vs moderate/severe; (3) radiological changes assessed with CT: no/slight vs moderate/severe. Four NTCP models were used: the Lyman model with DVH reduced to the equivalent uniform dose (LEUD), the Logit model with DVH reduced to EUD, the Mean Lung Dose (MLD) model and the Relative Seriality (RS) model. The data fitting procedure was done using the maximum likelihood analysis. The analysis was done on the entire population (n=87) and on a subgroup of patients treated with loco-regional RT (n=44).

RESULTS

24/87 patients (28%) developed clinical pneumonitis; 28/81 patients (35%) had radiological side effects on chest X-rays and 11/75 patients (15%) showed radiological density changes on Computed Tomography (CT). The analysis showed that the risk of clinical pneumonitis was a smooth function of EUD (calculated from DVH using n=0.86+/-0.10, best fit result). With LEUD, the relationship between EUD and NTCP could be described with a D(50) of 16.4Gy+/-1.1Gy and a steepness parameter m of 0.36+/-0.7. The results found in the overall population were substantially confirmed in the subgroup of patients treated with loco-regional RT.

CONCLUSIONS

A large group of prospective patient data (87 pts), including grade 1 pneumonitis, were analysed. The four NTCP models fit quite accurately the considered endpoints. EUD or the mean lung dose are robust and simple parameters correlated with the risk of pneumonitis. For all endpoints the D(50) values ranged in an interval between 10 and 20Gy.

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 Spearman's 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.

Gamma histograms for radiotherapy plan evaluation

BACKGROUND AND PURPOSE

The technique known as the 'gamma evaluation method' incorporates pass-fail criteria for both distance-to-agreement and dose difference analysis of 3D dose distributions and provides a numerical index (gamma) as a measure of the agreement between two datasets. As the gamma evaluation index is being adopted in more centres as part of treatment plan verification procedures for 2D and 3D dose maps, the development of methods capable of encapsulating the information provided by this technique is recommended.

PATIENTS AND METHODS

In this work the concept of gamma index was extended to create gamma histograms (GH) in order to provide a measure of the agreement between two datasets in two or three dimensions. Gamma area histogram (GAH) and gamma volume histogram (GVH) graphs were produced using one or more 2D gamma maps generated for each slice of the irradiated volume. GHs were calculated for IMRT plans, evaluating the 3D dose distribution from a commercial treatment planning system (TPS) compared to a Monte Carlo (MC) calculation used as reference dataset.

RESULTS

The extent of local anatomical inhomogenities in the plans under consideration was strongly correlated with the level of difference between reference and evaluated calculations. GHs provided an immediate visual representation of the proportion of the treated volume that fulfilled the gamma criterion and offered a concise method for comparative numerical evaluation of dose distributions.

CONCLUSIONS

We have introduced the concept of GHs and investigated its applications to the evaluation and verification of IMRT plans. The gamma histogram concept set out in this paper can provide a valuable technique for quantitative comparison of dose distributions and could be applied as a tool for the quality assurance of treatment planning systems.

Volumetric Uncertainty in Radiotherapy

The technologies available to identify anatomical structures (including radiotherapy target and normal tissue 'volumes'), and to deliver dose accurately to these volumes, have improved significantly in the past decade. However, the ability of clinicians to identify volumes accurately and consistently in patients still suffers from uncertainties that arise from human error, inadequate training, lack of consensus on the derivation of volumes and inadequate characterisation of the accuracy and specificity of imaging technologies. Inadequate volume definition of a target can result in treatment failure and, consequently, disease progression; excessive volume may also lead to unnecessary patient injury. This is a serious problem in routine clinical care. In the context of large multi-centre clinical trials, uncertainty and inconsistency in tissue-volume reporting will be carried through to the analysis of treatment effect on outcome, which will subsequently influence the treatment of future patients. Strategies need to be set in place to ensure that the abilities and consistency of clinicians in defining volumes are aligned with the ability of new technologies to present volumetric information. This review seeks to define the concept of volumetric uncertainty and propose a conceptual model that has these errors evaluated and responded to separately. Specifically, we will explore the major causes, consequences of, and possible remediation of volumetric uncertainty, from the point of view of a multidisciplinary radiotherapy clinical environment.

Clinical knowledge-based inverse treatment planning

Clinical IMRT treatment plans are currently made using dose-based optimization algorithms, which do not consider the nonlinear dose-volume effects for tumours and normal structures. The choice of structure specific importance factors represents an additional degree of freedom of the system and makes rigorous optimization intractable. The purpose of this work is to circumvent the two problems by developing a biologically more sensible yet clinically practical inverse planning framework. To implement this, the dose-volume status of a structure was characterized by using the effective volume in the voxel domain. A new objective function was constructed with the incorporation of the volumetric information of the system so that the figure of merit of a given IMRT plan depends not only on the dose deviation from the desired distribution but also the dose-volume status of the involved organs. The conventional importance factor of an organ was written into a product of two components: (i) a generic importance that parametrizes the relative importance of the organs in the ideal situation when the goals for all the organs are met; (ii) a dose-dependent factor that quantifies our level of clinical/dosimetric satisfaction for a given plan. The generic importance can be determined a priori, and in most circumstances, does not need adjustment, whereas the second one, which is responsible for the intractable behaviour of the trade-off seen in conventional inverse planning, was determined automatically. An inverse planning module based on the proposed formalism was implemented and applied to a prostate case and a head-neck case. A comparison with the conventional inverse planning technique indicated that, for the same target dose coverage, the critical structure sparing was substantially improved for both cases. The incorporation of clinical knowledge allows us to obtain better IMRT plans and makes it possible to auto-select the importance factors, greatly facilitating the inverse planning process. The new formalism proposed also reveals the relationship between different inverse planning schemes and gives important insight into the problem of therapeutic plan optimization. In particular, we show that the EUD-based optimization is a special case of the general inverse planning formalism described in this paper.

Fitting late rectal bleeding data using different NTCP models: results from an Italian multi-centric study (AIROPROS0101)

BACKGROUND AND PURPOSE

Recent investigations demonstrated a significant correlation between rectal dose-volume patterns and late rectal toxicity. The reduction of the DVH to a value expressing the probability of complication would be suitable. To fit different normal tissue complication probability (NTCP) models to clinical outcome on late rectal bleeding after external beam radiotherapy (RT) for prostate cancer.

PATIENTS AND METHODS

Rectal dose-volume histograms of the rectum (DVH) and clinical records of 547 prostate cancer patients (pts) pooled from five institutions previously collected and analyzed were considered. All patients were treated in supine position with 3 or 4-field techniques: 123 patients received an ICRU dose between 64 and 70 Gy, 255 patients between 70 and 74 Gy and 169 patients between 74 and 79.2 Gy; 457/547 patients were treated with conformal RT and 203/547 underwent radical prostatectomy before RT. Minimum follow-up was 18 months. Patients were considered as bleeders if showing grade 2/3 late bleeding (slightly modified RTOG/EORTC scoring system) within 18 months after the end of RT. Four NTCP models were considered: (a) the Lyman model with DVH reduced to the equivalent uniform dose (LEUD, coincident with the classical Lyman-Kutcher-Burman, LKB, model), (b) logistic with DVH reduced to EUD (LOGEUD), (c) Poisson coupled to EUD reduction scheme and (d) relative seriality (RS). The parameters for the different models were fit to the patient data using a maximum likelihood analysis. The 68% confidence intervals (CI) of each parameter were also derived.

RESULTS

Forty six out of five hundred and forty seven patients experienced grade 2/3 late bleeding: 38/46 developed rectal bleeding within 18 months and were then considered as bleeders The risk of rectal bleeding can be well calculated with a 'smooth' function of EUD (with a seriality parameter n equal to 0.23 (CI 0.05), best fit result). Using LEUD the relationship between EUD and NTCP can be described with a TD50 of 81.9 Gy (CI 1.8 Gy) and a steepness parameter m of 0.19 (CI 0.01); when using LOGEUD, TD50 is 82.2 Gy and k is 7.85. Best fit parameters for RS are s=0.49, gamma=1.69, TD50=83.1 Gy. Qualitative as well as quantitative comparisons (chi-squared statistics, P=0.005) show that the models fit the observed complication rates very well. The results found in the overall population were substantially confirmed in the subgroup of radically treated patients (LEUD: n=0.24 m=0.14 TD50=75.8 Gy). If considering just the grade 3 bleeders (n=9) the best fit is found in correspondence of a n-value around 0.06, suggesting that for severe bleeding the rectum is more serial.

CONCLUSIONS

Different NTCP models fit quite accurately the considered clinical data. The results are consistent with a rectum 'less serial' than previously reported investigations when considering grade 2 bleeding while a more serial behaviour was found for severe bleeding. EUD may be considered as a robust and simple parameter correlated with the risk of late rectal bleeding.

Target volume uncertainty and a method to visualize its effect on the target dose prescription

PURPOSE

To consider the uncertainty in the construction of target boundaries for optimization, and to demonstrate how the principles of mathematical programming can be applied to determine and display the effect on the tumor dose of making small changes to the target boundary.

METHODS

The effect on the achievable target dose of making successive small shifts to the target boundary within its range of uncertainty was found by constructing a mixed-integer linear program that automated the placement of the beam angles using the initial target volume.

RESULTS

The method was demonstrated using contours taken from a nasopharynx case, with dose limits placed on surrounding structures. In the illustrated case, enlarging the target anteriorly to provide greater assurance of disease coverage did not force a sacrifice in the minimum or mean tumor doses. However, enlarging the margin posteriorly, near a critical structure, dramatically changed the minimum, mean, and maximum tumor doses.

CONCLUSION

Tradeoffs between the position of the target boundary and the minimum target dose can be developed using mixed-integer programming, and the results projected as a guide to contouring and plan selection.

Spinal cord tolerance to high-dose fractionated 3D conformal proton-photon irradiation as evaluated by equivalent uniform dose and dose volume histogram analysis

PURPOSE

To evaluate cervical spinal cord tolerance using equivalent uniform dose (EUD) and dose volume histogram (DVH) analysis after proton-photon radiotherapy.

METHODS AND MATERIAL

The 3D dose distributions were analyzed in 85 patients with cervical vertebral tumors. Mean follow-up was 41.3 months. The mean prescribed dose was 76.3 Cobalt Gray Equivalent (CGE = proton dose x RBE 1.1). Dose constraints to the center and the surface of the cervical cord were 55-58 CGE and 67-70 CGE, respectively. Dose parameters, DVH and EUD, were calculated for each patient. The spinal cord toxicity was graded using the European Organization for Research and Treatment of Cancer (EORTC) and Radiation Therapy Oncology Group (RTOG) late effects scoring system.

RESULTS

Thirteen patients experienced Grade 1-2 toxicity. Four patients had Grade 3 toxicity. For the dose range used in this study, none of the dosimetric parameters was found to be associated with the observed distribution of cord toxicities. The only factor significantly associated with cord toxicity was the number of surgeries before irradiation.

CONCLUSION

The data and our analysis suggest that the integrity of the normal musculoskeletal supportive tissues and vascular supply may be important confounding factors of toxicity at these dose levels. The results also indicate that the cervical spinal cord dose constraints used in treating these patients are appropriate for conformal proton-photon radiotherapy.

Characterization of rectal normal tissue complication probability after high-dose external beam radiotherapy for prostate cancer

PURPOSE

Conformal radiotherapy (RT) has allowed radiation dose escalation to improve the outcome of prostate cancer. With higher doses, concern exists that rectal injury may increase. This study analyzed the utility and limitations of the widely used Lyman-Kutcher- Burman (LKB) normal tissue complication probability model in projecting the hazards of rectal complication with high-dose RT.

METHODS AND MATERIALS

A total of 128 patients were included in this study. These patients were treated with three-dimensional conformal RT alone at the University of Texas M.D. Anderson Cancer Center between 1992 and 1999. Patients were treated to 46 Gy with a four-field box technique followed by a six-field arrangement to boost the total dose to 78 Gy. All doses were delivered at 2 Gy/fraction to the isocenter. The minimal follow-up was 2 years. The end point for analysis was Grade 2 or worse rectal bleeding by 2 years. The LKB model was fitted to the data using the maximal likelihood method.

RESULTS

Of the 128 patients, 29 experienced Grade 2 or worse rectal bleeding by 2 years. For the entire cohort, the parameters obtained from the fit of the LKB model were as follows: the volume factor was n = 3.91 (95% confidence interval [CI] 0.031 to infinity ), dose associated with 50% chance of complication for uniform whole rectal irradiation [TD50(1)] was 53.6 Gy (95% CI 50.0-75.1), and a determinant of the steepness of the dose-response curve, (m), was 0.156 (95% CI 0.036-0.271). A statistically significant difference was found in the rate of postradiation rectal bleeding in patients with hemorrhoids vs. those without hemorrhoids. The parameters obtained for the patients without hemorrhoids were as follows: n = 0.746 (95% CI 0.026 to infinity ), TD50(1) 56.7 Gy (95% CI 49.9-75.2), and m 0.092 (95% CI 0.019-0.189).

CONCLUSION

Our analysis suggests a dose response for rectal bleeding probability along with a volume effect. We found that the LKB model might have limited utility in determining a large volume effect. We further suggest that LKB model should be used with caution in clinical practice.

The dose–volume constraint satisfaction problem for inverse treatment planning with field segments

The prescribed goals of radiation treatment planning are often expressed in terms of dose-volume constraints. We present a novel formulation of a dose-volume constraint satisfaction search for the discretized radiation therapy model. This approach does not rely on any explicit cost function. Inverse treatment planning uses the aperture-based approach with predefined, according to geometric rules, segmental fields. The solver utilizes the simultaneous version of the cyclic subgradient projection algorithm. This is a deterministic iterative method designed for solving the convex feasibility problems. A prescription is expressed with the set of inequalities imposed on the dose at the voxel resolution. Additional constraint functions control the compliance with selected points of the expected cumulative dose-volume histograms. The performance of this method is tested on prostate and head-and-neck cases. The relationships with other models and algorithms of similar conceptual origin are discussed. The demonstrated advantages of the method are: the equivalence of the algorithmic and prescription parameters, the intuitive setup of free parameters, and the improved speed of the method as compared to similar iterative as well as other techniques. The technique reported here will deliver approximate solutions for inconsistent prescriptions.

A TCP-NTCP estimation module using DVHs and known radiobiological models and parameter sets

Radiotherapy treatment plan evaluation relies on an implicit estimation of the tumor control probability (TCP) and normal tissue complication probability (NTCP) arising from a given dose distribution. A potential application of radiobiological modeling to radiotherapy is the ranking of treatment plans via a more explicit determination of TCP and NTCP values. Although the limited predictive capabilities of current radiobiological models prevent their use as a primary evaluative tool, radiobiological modeling predictions may still be a valuable complement to clinical experience. A convenient computational module has been developed for estimating the TCP and the NTCP arising from a dose distribution calculated by a treatment planning system, and characterized by differential (frequency) dose‐volume histograms (DDVHs). The radiobiological models included in the module are sigmoidal dose response and Critical Volume NTCP models, a Poisson TCP model, and a TCP model incorporating radiobiological parameters describing linear‐quadratic cell kill and repopulation. A number of sets of parameter values for the different models have been gathered in databases. The estimated parameters characterize the radiation response of several different normal tissues and tumor types. The system also allows input and storage of parameters by the user, which is particularly useful because of the rapidly increasing number of parameter estimates available in the literature. Potential applications of the system include the following: comparing radiobiological predictions of outcome for different treatment plans or types of treatment; comparing the number of observed outcomes for a cohort of patient DVHs to the predicted number of outcomes based on different models/parameter sets; and testing of the sensitivity of model predictions to uncertainties in the parameter values. The module thus helps to amalgamate and make more accessible current radiobiological modeling knowledge, and may serve as a useful aid in the prospective and retrospective analysis of radiotherapy treatment plans.

PACS number: 87.53.Tf

Are we doing any good by doing really well? (Where's the Bacon?)

Francis Bacon, who with Rene Decartes laid the intellectual foundations for Western science in the seventeenth century, asserted that the purpose of all knowledge is "action in the production of works for ... the relief of man's estate." We assess briefly several aspects of a few of the current efforts directed to the production of such "works" with respect to such "relief" as they may provide: cancer mortality, the medical literature, evidence-based medicine, clinical trials, observational databases and criteria for the promotion and tenure of the medical faculty. We suggest why each of these efforts appears to have failed to some degree and then propose some measures that may possibly serve as correctives.