Outcome modeling techniques for prostate cancer radiotherapy: Data, models, and validation

Dependence of the AUC of NTCP models on the observational dose-range highlights cautions in comparison of discriminative performance

BACKGROUND

Normal tissue complication probability (NTCP) models are probabilistic models that describe the risk of radio-induced toxicity in tissues or organs. In the field of radiotherapy, the area under the ROC curve (AUC) is widely used to estimate the performance in risk prediction of NTCP models.

METHODS

In this work, we derived an analytical expression of the AUC for the logistic NTCP model in the case of both symmetrical and asymmetrical dose (to the normal tissue) windows around D50. Using numerical simulations, we studied the behavior of the AUC in general clinical settings, enforcing non-logistic NTCP models (Lyman-Kutcher-Burman and LogEUD) and including risk factors beyond the dose. We validated our findings using real-world radiotherapy data sets of prostate cancer patients.

RESULTS

Our analytical expression of the AUC made explicit the dependence on both the steepness of the logistic curve (β) and the dose window width (w), showing that an increase of w pushes AUC towards higher values. Increasing values of the AUC with increasing values of w were consistently observed across simulated data sets with diverse clinical settings from published studies and real clinical data sets.

CONCLUSION

Our results reveal that the AUC of NTCP models inherits intrinsic characteristics from the clinical setting of the data set on which the models are developed, and warn against the use of the AUC to compare the performance of models constructed upon data from trials in which substantially different dose ranges were administered or accounting for different risk factors beyond the dose.

Cone-beam computed tomography-based radiomics in prostate cancer: a mono-institutional study

PURPOSE

The purpose of the reported study was to investigate the value of cone-beam computed tomography (CBCT)-based radiomics for risk stratification and prediction of biochemical relapse in prostate cancer.

METHODS

The study population consisted of 31 prostate cancer patients. Radiomics features were extracted from weekly CBCT scans performed for verifying treatment position. From the data, logistic-regression models were learned for establishing tumor stage, Gleason score, level of prostate-specific antigen, and risk stratification, and for predicting biochemical recurrence. Performance of the learned models was assessed using the area under the receiver operating characteristic curve (AUC-ROC) or the area under the precision-recall curve (AUC-PRC).

RESULTS

Results suggest that the histogram-based Energy and Kurtosis features and the shape-based feature representing the standard deviation of the maximum diameter of the prostate gland during treatment are predictive of biochemical relapse and indicative of patients at high risk.

CONCLUSION

Our results suggest the usefulness of CBCT-based radiomics for treatment definition in prostate cancer.

Machine Learning-Based Models for Prediction of Toxicity Outcomes in Radiotherapy

In order to limit radiotherapy (RT)-related side effects, effective toxicity prediction and assessment schemes are essential. In recent years, the growing interest toward artificial intelligence and machine learning (ML) within the science community has led to the implementation of innovative tools in RT. Several researchers have demonstrated the high performance of ML-based models in predicting toxicity, but the application of these approaches in clinics is still lagging, partly due to their low interpretability. Therefore, an overview of contemporary research is needed in order to familiarize practitioners with common methods and strategies. Here, we present a review of ML-based models for predicting and classifying RT-induced complications from both a methodological and a clinical standpoint, focusing on the type of features considered, the ML methods used, and the main results achieved. Our work overviews published research in multiple cancer sites, including brain, breast, esophagus, gynecological, head and neck, liver, lung, and prostate cancers. The aim is to define the current state of the art and main achievements within the field for both researchers and clinicians.

Optimization of treatment strategy by using a machine learning model to predict survival time of patients with malignant glioma after radiotherapy

The purpose of this study was to predict the survival time of patients with malignant glioma after radiotherapy with high accuracy by considering additional clinical factors and optimize the prescription dose and treatment duration for individual patient by using a machine learning model. A total of 35 patients with malignant glioma were included in this study. The candidate features included 12 clinical features and 192 dose-volume histogram (DVH) features. The appropriate input features and parameters of the support vector machine (SVM) were selected using the genetic algorithm based on Akaike's information criterion, i.e. clinical, DVH, and both clinical and DVH features. The prediction accuracy of the SVM models was evaluated through a leave-one-out cross-validation test with residual error, which was defined as the absolute difference between the actual and predicted survival times after radiotherapy. Moreover, the influences of various values of prescription dose and treatment duration on the predicted survival time were evaluated. The prediction accuracy was significantly improved with the combined use of clinical and DVH features compared with the separate use of both features (P < 0.01, Wilcoxon signed rank test). Mean ± standard deviation of the leave-one-out cross-validation using the combined clinical and DVH features, only clinical features and only DVH features were 104.7 ± 96.5, 144.2 ± 126.1 and 204.5 ± 186.0 days, respectively. The prediction accuracy could be improved with the combination of clinical and DVH features, and our results show the potential to optimize the treatment strategy for individual patients based on a machine learning model.

CT imaging markers to improve radiation toxicity prediction in prostate cancer radiotherapy by stacking regression algorithm

PURPOSE

Radiomic features, clinical and dosimetric factors have the potential to predict radiation-induced toxicity. The aim of this study was to develop prediction models of radiotherapy-induced toxicities in prostate cancer patients based on computed tomography (CT) radiomics, clinical and dosimetric parameters.

METHODS

In this prospective study, prostate cancer patients were included, and radiotherapy-induced urinary and gastrointestinal (GI) toxicities were assessed by Common Terminology Criteria for adverse events. For each patient, clinical and dose volume parameters were obtained. Imaging features were extracted from pre-treatment rectal and bladder wall CT scan of patients. Stacking algorithm and elastic net penalized logistic regression were used in order to feature selection and prediction, simultaneously. The models were fitted by imaging (radiomics model) and clinical/dosimetric (clinical model) features alone and in combinations (clinical-radiomics model). Goodness of fit of the models and performance of classifications were assessed using Hosmer and Lemeshow test, - 2log (likelihood) and area under curve (AUC) of the receiver operator characteristic.

RESULTS

Sixty-four prostate cancer patients were studied, and 33 and 52 patients developed ≥ grade 1 GI and urinary toxicities, respectively. In GI modeling, the AUC for clinical, radiomics and clinical-radiomics models was 0.66, 0.71 and 0.65, respectively. To predict urinary toxicity, the AUC for radiomics, clinical and clinical-radiomics models was 0.71, 0.67 and 0.77, respectively.

CONCLUSIONS

We have shown that CT imaging features could predict radiation toxicities and combination of imaging and clinical/dosimetric features may enhance the predictive performance of radiotoxicity modeling.

Computed Tomography-based Radiomics for Risk Stratification in Prostate Cancer

PURPOSE

To explore the role of Computed tomography (CT)-based radiomics features in prostate cancer risk stratification.

METHODS AND MATERIALS

The study population consisted of 506 patients with prostate cancer collected from a clinically annotated database. After applying exclusion criteria, 342 patients were included in the final analysis. CT-based radiomics features were extracted from planning CT scans for prostate gland-only structure, and machine learning was used to train models for Gleason score (GS) and risk group (RG) classifications. Repeated cross-validation was used. The discriminatory performance of the developed models was assessed using receiver operating characteristic area under the curve (AUC) analysis.

RESULTS

Classifiers using CT-based radiomics features distinguished between GS ≤ 6 versus GS ≥ 7 with AUC = 0.90 and GS 7(3 + 4) versus GS 7(4 + 3) with AUC = 0.98. Developed classifiers also showed excellent performance in distinguishing low versus high RG (AUC = 0.96) and low versus intermediate RG (AUC = 1.00), but poorer performance was observed for GS 7 versus GS > 7 (AUC = 0.69). An overall modest performance was observed for validation on holdout data sets with the highest AUC of 0.75 for classifiers of low versus high RG and an AUC of 0.70 for GS 7 versus GS > 7.

CONCLUSIONS

Our results show that radiomics features from routinely acquired planning CT scans could provide insights into prostate cancer aggressiveness in a noninvasive manner. Assessing models on training data sets, the classifiers were especially accurate in discerning high-risk from low-risk patients and in classifying GS 7 versus GS > 7 and GS 7(3 + 4) versus G7(4 + 3); however, classifiers were less adept at distinguishing high RG versus intermediate RG. External validation and prospective studies are warranted to verify the presented findings. These findings could potentially guide targeted radiation therapy strategies in radical intent radiation therapy for prostate cancer.

Computational analysis of interfractional anisotropic shape variations of the rectum in prostate cancer radiation therapy

PURPOSE

To analyze the uncertainties of the rectum due to anisotropic shape variations by using a statistical point distribution model (PDM).

MATERIALS AND METHODS

The PDM was applied to the rectum contours that were delineated on planning computed tomography (CT) and cone-beam CT (CBCT) at 80 fractions of 11 patients. The standard deviations (SDs) of systematic and random errors of the shape variations of the whole rectum and the region in which the rectum overlapped with the PTV (ROP regions) were derived from the PDMs at all fractions of each patient. The systematic error was derived by using the PDMs of planning and average rectum surface determined from rectum surfaces at all fractions, while the random error was derived by using a PDM-based covariance matrix at all fractions of each patient.

RESULTS

Regarding whole rectum, the population SDs were larger than 1.0 mm along all directions for random error, and along the anterior, superior, and inferior directions for systematic error. The deviation is largest along the superior and inferior directions for systematic and random errors, respectively. For ROP regions, the population SDs of systematic error were larger than 1.0 mm along the superior and inferior directions. The population SDs of random error for the ROP regions were larger than 1.0 mm except along the right and posterior directions.

CONCLUSIONS

The anisotropic shape variations of the rectum, especially in the ROP regions, should be considered when determining a planning risk volume (PRV) margins for the rectum associated with the acute toxicities.

Knowledge-based treatment planning: An inter-technique and inter-system feasibility study for prostate cancer

PURPOSE

Helical Tomotherapy (HT) plans were used to create two RapidPlan knowledge-based (KB) models to generate plans with different techniques and to guide the optimization in a different treatment planning system for prostate plans. Feasibility and performance of these models were evaluated.

MATERIAL AND METHODS

two sets of 35 low risk (LR) and 30 intermediate risk (IR) prostate cancer cases who underwent HT treatments were selected to train RapidPlan models. The KB predicted constraints were used to perform new 20KB plans using RapidArc technique (KB-RA) (inter-technique validation), and to optimise 20 new HT (KB-HT) plans in the Tomoplan (inter-system validation). For each validation modality, KB plans were benchmarked with the manual plans created by an expert planner (EP).

RESULTS

RapidPlan was successfully configured using HT plans. The KB-RA plans fulfilled the clinical dose-volume requirements in 100% and 92% of cases for planning target volumes (PTVs) and organs at risk (OARs), respectively. For KB-HT plans these percentages were found to be a bit lower: 90% for PTVs and 86% for OARs. In comparison to EP plans, the KB-RA plans produced higher bladder doses for both LR and IR, and higher rectum doses for LR. KB-HT and EP plans produced similar results.

CONCLUSION

RapidPlan can be trained to create models by using plans of a different treatment modality. These models were suitable for generating clinically acceptable plans for inter-technique and inter-system applications. The use of KB models based on plans of consolidated technique could be useful with a new treatment modality.

Perspectives on making big data analytics work for oncology

Oncology, with its unique combination of clinical, physical, technological, and biological data provides an ideal case study for applying big data analytics to improve cancer treatment safety and outcomes. An oncology treatment course such as chemoradiotherapy can generate a large pool of information carrying the 5Vs hallmarks of big data. This data is comprised of a heterogeneous mixture of patient demographics, radiation/chemo dosimetry, multimodality imaging features, and biological markers generated over a treatment period that can span few days to several weeks. Efforts using commercial and in-house tools are underway to facilitate data aggregation, ontology creation, sharing, visualization and varying analytics in a secure environment. However, open questions related to proper data structure representation and effective analytics tools to support oncology decision-making need to be addressed. It is recognized that oncology data constitutes a mix of structured (tabulated) and unstructured (electronic documents) that need to be processed to facilitate searching and subsequent knowledge discovery from relational or NoSQL databases. In this context, methods based on advanced analytics and image feature extraction for oncology applications will be discussed. On the other hand, the classical p (variables)≫n (samples) inference problem of statistical learning is challenged in the Big data realm and this is particularly true for oncology applications where p-omics is witnessing exponential growth while the number of cancer incidences has generally plateaued over the past 5-years leading to a quasi-linear growth in samples per patient. Within the Big data paradigm, this kind of phenomenon may yield undesirable effects such as echo chamber anomalies, Yule-Simpson reversal paradox, or misleading ghost analytics. In this work, we will present these effects as they pertain to oncology and engage small thinking methodologies to counter these effects ranging from incorporating prior knowledge, using information-theoretic techniques to modern ensemble machine learning approaches or combination of these. We will particularly discuss the pros and cons of different approaches to improve mining of big data in oncology.

The role of medical physics in prostate cancer radiation therapy

Medical physics, both as a scientific discipline and clinical service, hugely contributed and still contributes to the advances in the radiotherapy of prostate cancer. The traditional translational role in developing and safely implementing new technology and methods for better optimizing, delivering and monitoring the treatment is rapidly expanding to include new fields such as quantitative morphological and functional imaging and the possibility of individually predicting outcome and toxicity. The pivotal position of medical physicists in treatment personalization probably represents the main challenge of current and next years and needs a gradual change of vision and training, without losing the traditional and fundamental role of physicists to guarantee a high quality of the treatment. The current focus issue is intended to cover traditional and new fields of investigation in prostate cancer radiation therapy with the aim to provide up-to-date reference material to medical physicists daily working to cure prostate cancer patients. The papers presented in this focus issue touch upon present and upcoming challenges that need to be met in order to further advance prostate cancer radiation therapy. We suggest that there is a smart future for medical physicists willing to perform research and innovate, while they continue to provide high-quality clinical service. However, physicists are increasingly expected to actively integrate their implicitly translational, flexible and high-level skills within multi-disciplinary teams including many clinical figures (first of all radiation oncologists) as well as scientists from other disciplines.