Bayesian network ensemble as a multivariate strategy to predict radiation pneumonitis risk

Performance Comparison of 10 State-of-the-Art Machine Learning Algorithms for Outcome Prediction Modeling of Radiation-Induced Toxicity

Purpose

To evaluate the efficacy of prominent machine learning algorithms in predicting normal tissue complication probability using clinical data obtained from 2 distinct disease sites and to create a software tool that facilitates the automatic determination of the optimal algorithm to model any given labeled data set.

Methods and Materials

We obtained 3 sets of radiation toxicity data (478 patients) from our clinic: gastrointestinal toxicity, radiation pneumonitis, and radiation esophagitis. These data comprised clinicopathological and dosimetric information for patients diagnosed with non-small cell lung cancer and anal squamous cell carcinoma. Each data set was modeled using 11 commonly employed machine learning algorithms (elastic net, least absolute shrinkage and selection operator [LASSO], random forest, random forest regression, support vector machine, extreme gradient boosting, light gradient boosting machine, k-nearest neighbors, neural network, Bayesian-LASSO, and Bayesian neural network) by randomly dividing the data set into a training and test set. The training set was used to create and tune the model, and the test set served to assess it by calculating performance metrics. This process was repeated 100 times by each algorithm for each data set. Figures were generated to visually compare the performance of the algorithms. A graphical user interface was developed to automate this whole process.

Results

LASSO achieved the highest area under the precision-recall curve (0.807 ± 0.067) for radiation esophagitis, random forest for gastrointestinal toxicity (0.726 ± 0.096), and the neural network for radiation pneumonitis (0.878 ± 0.060). The area under the curve was 0.754 ± 0.069, 0.889 ± 0.043, and 0.905 ± 0.045, respectively. The graphical user interface was used to compare all algorithms for each data set automatically. When averaging the area under the precision-recall curve across all toxicities, Bayesian-LASSO was the best model.

Conclusions

Our results show that there is no best algorithm for all data sets. Therefore, it is important to compare multiple algorithms when training an outcome prediction model on a new data set. The graphical user interface created for this study automatically compares the performance of these 11 algorithms for any data set.

Artificial intelligence uncertainty quantification in radiotherapy applications - A scoping review

BACKGROUND/PURPOSE

The use of artificial intelligence (AI) in radiotherapy (RT) is expanding rapidly. However, there exists a notable lack of clinician trust in AI models, underscoring the need for effective uncertainty quantification (UQ) methods. The purpose of this study was to scope existing literature related to UQ in RT, identify areas of improvement, and determine future directions.

METHODS

We followed the PRISMA-ScR scoping review reporting guidelines. We utilized the population (human cancer patients), concept (utilization of AI UQ), context (radiotherapy applications) framework to structure our search and screening process. We conducted a systematic search spanning seven databases, supplemented by manual curation, up to January 2024. Our search yielded a total of 8980 articles for initial review. Manuscript screening and data extraction was performed in Covidence. Data extraction categories included general study characteristics, RT characteristics, AI characteristics, and UQ characteristics.

RESULTS

We identified 56 articles published from 2015 to 2024. 10 domains of RT applications were represented; most studies evaluated auto-contouring (50 %), followed by image-synthesis (13 %), and multiple applications simultaneously (11 %). 12 disease sites were represented, with head and neck cancer being the most common disease site independent of application space (32 %). Imaging data was used in 91 % of studies, while only 13 % incorporated RT dose information. Most studies focused on failure detection as the main application of UQ (60 %), with Monte Carlo dropout being the most commonly implemented UQ method (32 %) followed by ensembling (16 %). 55 % of studies did not share code or datasets.

CONCLUSION

Our review revealed a lack of diversity in UQ for RT applications beyond auto-contouring. Moreover, we identified a clear need to study additional UQ methods, such as conformal prediction. Our results may incentivize the development of guidelines for reporting and implementation of UQ in RT.

Predicting radiation pneumonitis in lung cancer using machine learning and multimodal features: a systematic review and meta-analysis of diagnostic accuracy

OBJECTIVES

To evaluate the diagnostic accuracy of machine learning models incorporating multimodal features for predicting radiation pneumonitis in lung cancer through a systematic review and meta-analysis.

METHODS

Relevant studies were identified through a systematic search of PubMed, Web of Science, Embase, and the Cochrane Library from October 2003 to December 2023. Additional studies were located by reviewing bibliographies and relevant websites. Two independent researchers screened titles, abstracts, and full-text articles according to predefined inclusion and exclusion criteria. Data extraction was performed using standardized forms, and study quality was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 tool. The primary outcomes, including combined sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC), were calculated using STATA MP-64 software(Stata Corporation LLC, College Station, USA) with a random-effects model. Meta-analysis was conducted to synthesize diagnostic accuracy measures, and analyses of heterogeneity and publication bias were performed.

RESULTS

A total of 1,406 patients with primary lung cancer were included in this systematic review, drawing data from 9 studies. The pooled analysis revealed a sensitivity of 0.74 [0.58-0.85] and a specificity of 0.91 [0.87-0.95] for machine learning models in diagnosing radiation pneumonitis. The positive likelihood ratio (PLR) was 8.69 [5.21-14.50], the negative likelihood ratio (NLR) was 0.28 [0.16-0.49], and the diagnostic odds ratio (DOR) was 30.73 [11.96-78.97]. The area under the curve (AUC) was 0.93 [0.90-0.95], indicating excellent diagnostic performance. Meta-regression analysis identified that the number of machine learning models, year of publication, and study design contributed to heterogeneity among studies. No evidence of publication bias was found. Overall, machine learning models incorporating multimodal characteristics demonstrated 75% accuracy in predicting moderate to severe radiation pneumonitis.

CONCLUSION

In conclusion, by integrating the current machine learning (ML) algorithm's ability in big data mining, a predictive model can be constructed by combining multi-modal features such as genetics, imaging, and cell factors. By selecting multiple machine learning algorithm frameworks and competing for the best combination model based on research goals, the reliability and accuracy of the radiation pneumonitis prediction model can be greatly improved.

TRIAL REGISTRATION

PROSPERO (CRD42024497599).

Biomarkers associated with pulmonary exacerbations in a randomized trial of nintedanib for radiation pneumonitis

BACKGROUND AND PURPOSE

Radiation pneumonitis (RP) is a common side effect of thoracic radiotherapy and often has a long course characterized by acute exacerbations and progression to permanent lung fibrosis. There are no validated biomarkers of prognosis in patients diagnosed with RP.

MATERIALS AND METHODS

We analyzed a time course of serum chemokines, cytokines, and other proteins from patients with grade 2+ RP in a randomized clinical trial of a steroid taper plus nintedanib, a multiple tyrosine kinase inhibitor, versus placebo plus a steroid taper for the treatment of RP. Weighted gene correlation network analysis (WGCNA) and univariable zero inflated Poisson models were used to identify groups of correlated analytes and their associations with clinical outcomes.

RESULTS

Thirty enrolled patients had biomarker data available, and 17 patients had enough analytes tested for network analysis. WGNCA identified ten analytes, including transforming growth factor beta-1 (TGF-β1), monocyte chemoattractant protein-1 (MCP-1), and platelet-derived growth factor (PDGF), that in aggregate were correlated with the occurrence of pulmonary exacerbations (p = 0.008), the total number of acute pulmonary exacerbations (p = 0.002), and treatment arm (p = 0.036). By univariable analysis, an increase in rate of change of two components of the RP module were associated with an increased incidence rate of pulmonary exacerbations: interleukin 5 (IL-5, incidence rate ratio (IRR) 1.02, 95% CI 1.01-1.04, p = 0.002), and tumor necrosis factor superfamily 12 (TNFSF12, IRR 1.06, CI 1-1.11, p = 0.036). An increased slope of epidermal growth factor (EGF) was associated with a decreased incidence rate of exacerbations (IRR 0.94, CI 0.89-1, p = 0.036).

CONCLUSION

We identified a panel of serum biomarkers that showed association with nintedanib treatment and acute pulmonary exacerbations in patients with RP. A confirmatory study will be needed to validate this panel for use as a prognostic tool in patients with RP.

Artificial Intelligence Uncertainty Quantification in Radiotherapy Applications - A Scoping Review

Background/purpose

The use of artificial intelligence (AI) in radiotherapy (RT) is expanding rapidly. However, there exists a notable lack of clinician trust in AI models, underscoring the need for effective uncertainty quantification (UQ) methods. The purpose of this study was to scope existing literature related to UQ in RT, identify areas of improvement, and determine future directions.

Methods

We followed the PRISMA-ScR scoping review reporting guidelines. We utilized the population (human cancer patients), concept (utilization of AI UQ), context (radiotherapy applications) framework to structure our search and screening process. We conducted a systematic search spanning seven databases, supplemented by manual curation, up to January 2024. Our search yielded a total of 8980 articles for initial review. Manuscript screening and data extraction was performed in Covidence. Data extraction categories included general study characteristics, RT characteristics, AI characteristics, and UQ characteristics.

Results

We identified 56 articles published from 2015-2024. 10 domains of RT applications were represented; most studies evaluated auto-contouring (50%), followed by image-synthesis (13%), and multiple applications simultaneously (11%). 12 disease sites were represented, with head and neck cancer being the most common disease site independent of application space (32%). Imaging data was used in 91% of studies, while only 13% incorporated RT dose information. Most studies focused on failure detection as the main application of UQ (60%), with Monte Carlo dropout being the most commonly implemented UQ method (32%) followed by ensembling (16%). 55% of studies did not share code or datasets.

Conclusion

Our review revealed a lack of diversity in UQ for RT applications beyond auto-contouring. Moreover, there was a clear need to study additional UQ methods, such as conformal prediction. Our results may incentivize the development of guidelines for reporting and implementation of UQ in RT.

Embedding machine learning based toxicity models within radiotherapy treatment plan optimization

Objective.This study addresses radiation-induced toxicity (RIT) challenges in radiotherapy (RT) by developing a personalized treatment planning framework. It leverages patient-specific data and dosimetric information to create an optimization model that limits adverse side effects using constraints learned from historical data.Approach.The study uses the optimization with constraint learning (OCL) framework, incorporating patient-specific factors into the optimization process. It consists of three steps: optimizing the baseline treatment plan using population-wide dosimetric constraints; training a machine learning (ML) model to estimate the patient's RIT for the baseline plan; and adapting the treatment plan to minimize RIT using ML-learned patient-specific constraints. Various predictive models, including classification trees, ensembles of trees, and neural networks, are applied to predict the probability of grade 2+ radiation pneumonitis (RP2+) for non-small cell lung (NSCLC) cancer patients three months post-RT. The methodology is assessed with four high RP2+ risk NSCLC patients, with the goal of optimizing the dose distribution to constrain the RP2+ outcome below a pre-specified threshold. Conventional and OCL-enhanced plans are compared based on dosimetric parameters and predicted RP2+ risk. Sensitivity analysis on risk thresholds and data uncertainty is performed using a toy NSCLC case.Main results.Experiments show the methodology's capacity to directly incorporate all predictive models into RT treatment planning. In the four patients studied, mean lung dose and V20 were reduced by an average of 1.78 Gy and 3.66%, resulting in an average RP2+ risk reduction from 95% to 42%. Notably, this reduction maintains tumor coverage, although in two cases, sparing the lung slightly increased spinal cord max-dose (0.23 and 0.79 Gy).Significance.By integrating patient-specific information into learned constraints, the study significantly reduces adverse side effects like RP2+ without compromising target coverage. This unified framework bridges the gap between predicting toxicities and optimizing treatment plans in personalized RT decision-making.

Image masking using convolutional networks improves performance classification of radiation pneumonitis for non-small cell lung cancer

Radiation pneumonitis (RP) is a serious side effect of radiotherapy in patients with locally advanced non-small-cell lung cancer (NSCLC). Image cropping reduces training noise and may improve classification accuracy. This study proposes a prediction model for RP grade ≥ 2 using a convolutional neural network (CNN) model with image cropping. The 3D computed tomography (CT) images cropped in the whole-body, normal lung (nLung), and nLung regions overlapping the region over 20 Gy (nLung∩20 Gy) used in treatment planning were used as the input data. The output classifies patients as RP grade < 2 or RP grade ≥ 2. The sensitivity, specificity, accuracy, and area under the curve (AUC) were evaluated using the receiver operating characteristic curve (ROC). The accuracy, specificity, sensitivity, and AUC were 53.9%, 80.0%, 25.5%, and 0.58, respectively, for the whole-body method, and 60.0%, 81.7%, 36.4%, and 0.64, respectively, for the nLung method. For the nLung∩20 Gy method, the accuracy, specificity, sensitivity, and AUC improved to 75.7%, 80.0%, 70.9%, and 0.84, respectively. The CNN model, in which the input image is segmented in the normal lung considering the dose distribution, can help predict an RP grade ≥ 2 for NSCLC patients after definitive radiotherapy.

Personalized mid-course FDG-PET based adaptive treatment planning for non-small cell lung cancer using machine learning and optimization

Objective. Traditional radiotherapy (RT) treatment planning of non-small cell lung cancer (NSCLC) relies on population-wide estimates of organ tolerance to minimize excess toxicity. The goal of this study is to develop a personalized treatment planning based on patient-specific lung radiosensitivity, by combining machine learning and optimization.Approach. Sixty-nine non-small cell lung cancer patients with baseline and mid-treatment [18]F-fluorodeoxyglucose (FDG)-PET images were retrospectively analyzed. A probabilistic Bayesian networks (BN) model was developed to predict the risk of radiation pneumonitis (RP) at three months post-RT using pre- and mid-treatment FDG information. A patient-specific dose modifying factor (DMF), as a surrogate for lung radiosensitivity, was estimated to personalize the normal tissue toxicity probability (NTCP) model. This personalized NTCP was then integrated into a NTCP-based optimization model for RT adaptation, ensuring tumor coverage and respecting patient-specific lung radiosensitivity. The methodology was employed to adapt the treatment planning of fifteen NSCLC patients.Main results. The magnitude of the BN predicted risks corresponded with the RP severity. Average predicted risk for grade 1-4 RP were 0.18, 0.42, 0.63, and 0.76, respectively (p< 0.001). The proposed model yielded an average area under the receiver-operating characteristic curve (AUROC) of 0.84, outperforming the AUROCs of LKB-NTCP (0.77), and pre-treatment BN (0.79). Average DMF for the radio-tolerant (RP grade = 1) and radiosensitive (RP grade ≥ 2) groups were 0.8 and 1.63,p< 0.01. RT personalization resulted in five dose escalation strategies (average mean tumor dose increase = 6.47 Gy, range = [2.67-17.5]), and ten dose de-escalation (average mean lung dose reduction = 2.98 Gy [0.8-5.4]), corresponding to average NTCP reduction of 15% [4-27].Significance. Personalized FDG-PET-based mid-treatment adaptation of NSCLC RT could significantly lower the RP risk without compromising tumor control. The proposed methodology could help the design of personalized clinical trials for NSCLC patients.

Symptomatic Radiation Pneumonitis in NSCLC Patients Receiving EGFR-TKIs and Concurrent Once-daily Thoracic Radiotherapy: Predicting the Value of Clinical and Dose-volume Histogram Parameters

BACKGROUND

The incidence of symptomatic radiation pneumonitis (RP) and its relationship with dose-volume histogram (DVH) parameters in non-small cell lung cancer (NSCLC) patients receiving epidermal growth factor receptor-tyrosine kinase inhibitors (EGFR-TKIs) and concurrent once-daily thoracic radiotherapy (TRT) remain unclear. We aim to analyze the values of clinical factors and dose-volume histogram (DVH) parameters to predict the risk for symptomatic RP in these patients.

METHODS

Between 2011 and 2019, we retrospectively analyzed and identified 85 patients who had received EGFR-TKIs and once-daily TRT simultaneously (EGFR-TKIs group) and 129 patients who had received concurrent chemoradiotherapy (CCRT group). The symptomatic RP was recorded according to the Common Terminology Criteria for Adverse Event (CTCAE) criteria (grade 2 or above). Statistical analyses were performed using SPSS 26.0.

RESULTS

In total, the incidences of symptomatic (grade≥2) and severe RP (grade≥3) were 43.5% (37/85) and 16.5% (14/85) in EGFR-TKIs group vs 27.1% (35/129) and 10.1% (13/129) in CCRT group respectively. After 1:1 ratio between EGFR-TKIs group and CCRT group was matched by propensity score matching, chi-square test suggested that the incidence of symptomatic RP in the MATCHED EGFR-TKIs group was higher than that in the matched CCRT group (χ2=4.469, P=0.035). In EGFR-TKIs group, univariate and multivariate analyses indicated that the percentage of ipsilateral lung volume receiving ≥30 Gy (ilV30) [odds ratio (OR): 1.163, 95%CI: 1.036-1.306, P=0.011] and the percentage of total lung volume receiving ≥20 Gy (tlV20) (OR: 1.171, 95%CI: 1.031-1.330, P=0.015), with chronic obstructive pulmonary disease (COPD) or not (OR: 0.158, 95%CI: 0.041-0.600, P=0.007), were independent predictors of symptomatic RP. Compared to patients with lower ilV30/tlV20 values (ilV30 and tlV20<cut-off point values) and without COPD, patients with higher ilV30/tlV20 values (ilV30 and tlV20>cut-off point values) and COPD had a significantly higher risk for developing symptomatic RP, with a hazard ratio (HR) of 1.350 (95%CI: 1.190-1.531, P<0.001).

CONCLUSIONS

Patients receiving both EGFR-TKIs and once-daily TRT were more likely to develop symptomatic RP than patients receiving concurrent chemoradiotherapy. The ilV30, tlV20, and comorbidity of COPD may predict the risk of symptomatic RP among NSCLC patients receiving EGFR-TKIs and conventionally fractionated TRT concurrently.

Assessment of Tissue Toxicity Risk in Breast Radiotherapy using Bayesian Networks

PURPOSE

The purpose of this analysis is to predict worsening post-treatment normal tissue toxicity in patients undergoing accelerated partial breast irradiation (APBI) therapy and to quantitatively identify which diagnostic, anatomical, and dosimetric features are contributing to these outcomes.

METHODS

A retrospective study of APBI treatments was performed using 32 features pertaining to various stages of the patient's treatment journey. These features were used to inform and construct a Bayesian network (BN) based on both statistical analysis of feature distributions and relative clinical importance. The target feature for prediction was defined as a measurable worsening of telangiectasia, subcutaneous tissue induration, or fibrosis when compared against the observed baseline. Parameter learning for the network was performed using data from the 299 patients included in the ACCEL trial and predictive performance was measured. Feature importance for the BN was quantified using a novel information-theoretic approach.

RESULTS

Cross validated performance of the BN for predicting toxicity was consistently higher when compared against conventional machine learning (ML) techniques. The measured BN receiver operating characteristic area under the curve was 0.960±0.013 against the best ML result of 0.942±0.021 using 5-fold cross validation with separate test data across 100 trials. The volume of the clinical target volume, gross target volume, and baseline toxicity measurements were found to have the highest feature importance and mutual dependence with normal tissue toxicity in the network, representing the strongest contribution to patient outcomes.

CONCLUSIONS

The BN outperformed conventional ML techniques in predicting tissue toxicity outcomes and provided deeper insight into which features are contributing to these outcomes. This article is protected by copyright. All rights reserved.

Radiomic and radiogenomic modeling for radiotherapy: strategies, pitfalls, and challenges

The power of predictive modeling for radiotherapy outcomes has historically been limited by an inability to adequately capture patient-specific variabilities; however, next-generation platforms together with imaging technologies and powerful bioinformatic tools have facilitated strategies and provided optimism. Integrating clinical, biological, imaging, and treatment-specific data for more accurate prediction of tumor control probabilities or risk of radiation-induced side effects are high-dimensional problems whose solutions could have widespread benefits to a diverse patient population-we discuss technical approaches toward this objective. Increasing interest in the above is specifically reflected by the emergence of two nascent fields, which are distinct but complementary: radiogenomics, which broadly seeks to integrate biological risk factors together with treatment and diagnostic information to generate individualized patient risk profiles, and radiomics, which further leverages large-scale imaging correlates and extracted features for the same purpose. We review classical analytical and data-driven approaches for outcomes prediction that serve as antecedents to both radiomic and radiogenomic strategies. Discussion then focuses on uses of conventional and deep machine learning in radiomics. We further consider promising strategies for the harmonization of high-dimensional, heterogeneous multiomics datasets (panomics) and techniques for nonparametric validation of best-fit models. Strategies to overcome common pitfalls that are unique to data-intensive radiomics are also discussed.

Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study

BACKGROUND

Radiation pneumonitis (RP) is a dose-limiting toxicity in lung cancer radiotherapy (RT). As risk factors in the development of RP, patient and tumor characteristics, dosimetric parameters, and treatment features are intertwined, and it is not always possible to associate RP with a single parameter. This study aimed to determine the algorithm that most accurately predicted RP development with machine learning.

METHODS

Of the 197 cases diagnosed with stage III lung cancer and underwent RT and chemotherapy between 2014 and 2020, 193 were evaluated. The CTCAE 5.0 grading system was used for the RP evaluation. Synthetic minority oversampling technique was used to create a balanced data set. Logistic regression, artificial neural networks, eXtreme Gradient Boosting (XGB), Support Vector Machines, Random Forest, Gaussian Naive Bayes and Light Gradient Boosting Machine algorithms were used. After the correlation analysis, a permutation-based method was utilized for as a variable selection.

RESULTS

RP was seen in 51 of the 193 cases. Parameters affecting RP were determined as, total(t)V5, ipsilateral lung Dmax, contralateral lung Dmax, total lung Dmax, gross tumor volume, number of chemotherapy cycles before RT, tumor size, lymph node localization and asbestos exposure. LGBM was found to be the algorithm that best predicted RP at 85% accuracy (confidence interval: 0.73-0.96), 97% sensitivity, and 50% specificity.

CONCLUSION

When the clinical and dosimetric parameters were evaluated together, the LGBM algorithm had the highest accuracy in predicting RP. However, in order to use this algorithm in clinical practice, it is necessary to increase data diversity and the number of patients by sharing data between centers.

Bayesian Network to Predict the Risk of Post Influenza Vaccination Guillain-Barré Syndrome: A Development and Validation Study (Preprint)

Background

Identifying the key factors of Guillain-Barré syndrome (GBS) and predicting its occurrence are vital for improving the prognosis of patients with GBS. However, there are scarcely any publications on a forewarning model of GBS. A Bayesian network (BN) model, which is known to be an accurate, interpretable, and interaction-sensitive graph model in many similar domains, is worth trying in GBS risk prediction.

Objective

The aim of this study is to determine the most significant factors of GBS and further develop and validate a BN model for predicting GBS risk.

Methods

Large-scale influenza vaccine postmarketing surveillance data, including 79,165 US (obtained from the Vaccine Adverse Event Reporting System between 1990 and 2017) and 12,495 European (obtained from the EudraVigilance system between 2003 and 2016) adverse events (AEs) reports, were extracted for model development and validation. GBS, age, gender, and the top 50 prevalent AEs were included for initial BN construction using the R package bnlearn.

Results

Age, gender, and 10 AEs were identified as the most significant factors of GBS. The posttest probability of GBS suggested that male vaccinees aged 50-64 years and without erythema should be on the alert or be warned by clinicians about an increased risk of GBS, especially when they also experience symptoms of asthenia, hypesthesia, muscular weakness, or paresthesia. The established BN model achieved an area under the receiver operating characteristic curve of 0.866 (95% CI 0.865-0.867), sensitivity of 0.752 (95% CI 0.749-0.756), specificity of 0.882 (95% CI 0.879-0.885), and accuracy of 0.882 (95% CI 0.879-0.884) for predicting GBS risk during the internal validation and obtained values of 0.829, 0.673, 0.854, and 0.843 for area under the receiver operating characteristic curve, sensitivity, specificity, and accuracy, respectively, during the external validation.

Conclusions

The findings of this study illustrated that a BN model can effectively identify the most significant factors of GBS, improve understanding of the complex interactions among different postvaccination symptoms through its graphical representation, and accurately predict the risk of GBS. The established BN model could further assist clinical decision-making by providing an estimated risk of GBS for a specific vaccinee or be developed into an open-access platform for vaccinees’ self-monitoring.

Application of Radiomics for the Prediction of Radiation-Induced Toxicity in the IMRT Era: Current State-of-the-Art

Normal tissue complication probability (NTCP) models that were formulated in the Quantitative Analyses of Normal Tissue Effects in the Clinic (QUANTEC) are one of the pillars in support of everyday’s clinical radiation oncology. Because of steady therapeutic refinements and the availability of cutting-edge technical solutions, the ceiling of organs-at-risk-sparing has been reached for photon-based intensity modulated radiotherapy (IMRT). The possibility to capture heterogeneity of patients and tissues in the prediction of toxicity is still an unmet need in modern radiation therapy. Potentially, a major step towards a wider therapeutic index could be obtained from refined assessment of radiation-induced morbidity at an individual level. The rising integration of quantitative imaging and machine learning applications into radiation oncology workflow offers an unprecedented opportunity to further explore the biologic interplay underlying the normal tissue response to radiation. Based on these premises, in this review we focused on the current-state-of-the-art on the use of radiomics for the prediction of toxicity in the field of head and neck, lung, breast and prostate radiotherapy.

Predictive Modeling of Thoracic Radiotherapy Toxicity and the Potential Role of Serum Alpha-2-Macroglobulin

Background: To investigate the impact of alpha-2-macroglobulin (A2M), a suspected intrinsic radioprotectant, on radiation pneumonitis and esophagitis using multifactorial predictive models. Materials and Methods: Baseline A2M levels were obtained for 258 patients prior to thoracic radiotherapy (RT). Dose-volume characteristics were extracted from treatment plans. Spearman's correlation (Rs) test was used to correlate clinical and dosimetric variables with toxicities. Toxicity prediction models were built using least absolute shrinkage and selection operator (LASSO) logistic regression on 1,000 bootstrapped datasets. Results: Grade ≥2 esophagitis and pneumonitis developed in 61 (23.6%) and 36 (14.0%) patients, respectively. The median A2M level was 191 mg/dL (range: 94-511). Never/former/current smoker status was 47 (18.2%)/179 (69.4%)/32 (12.4%). We found a significant negative univariate correlation between baseline A2M levels and esophagitis (Rs = -0.18/p = 0.003) and between A2M and smoking status (Rs = 0.13/p = 0.04). Further significant parameters for grade ≥2 esophagitis included age (Rs = -0.32/p < 0.0001), chemotherapy use (Rs = 0.56/p < 0.0001), dose per fraction (Rs = -0.57/p < 0.0001), total dose (Rs = 0.35/p < 0.0001), and several other dosimetric variables with Rs > 0.5 (p < 0.0001). The only significant non-dosimetric parameter for grade ≥2 pneumonitis was sex (Rs = -0.32/p = 0.037) with higher risk for women. For pneumonitis D15 (lung) (Rs = 0.19/p = 0.006) and D45 (heart) (Rs = 0.16/p = 0.016) had the highest correlation. LASSO models applied on the validation data were statistically significant and resulted in areas under the receiver operating characteristic curve of 0.84 (esophagitis) and 0.78 (pneumonitis). Multivariate predictive models did not require A2M to reach maximum predictive power. Conclusion: This is the first study showing a likely association of higher baseline A2M values with lower risk of radiation esophagitis and with smoking status. However, the baseline A2M level was not a significant risk factor for radiation pneumonitis.

Computational advances of tumor marker selection and sample classification in cancer proteomics

Cancer proteomics has become a powerful technique for characterizing the protein markers driving transformation of malignancy, tracing proteome variation triggered by therapeutics, and discovering the novel targets and drugs for the treatment of oncologic diseases. To facilitate cancer diagnosis/prognosis and accelerate drug target discovery, a variety of methods for tumor marker identification and sample classification have been developed and successfully applied to cancer proteomic studies. This review article describes the most recent advances in those various approaches together with their current applications in cancer-related studies. Firstly, a number of popular feature selection methods are overviewed with objective evaluation on their advantages and disadvantages. Secondly, these methods are grouped into three major classes based on their underlying algorithms. Finally, a variety of sample separation algorithms are discussed. This review provides a comprehensive overview of the advances on tumor maker identification and patients/samples/tissues separations, which could be guidance to the researches in cancer proteomics.

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.

Genomics models in radiotherapy: From mechanistic to machine learning

Machine learning (ML) provides a broad framework for addressing high-dimensional prediction problems in classification and regression. While ML is often applied for imaging problems in medical physics, there are many efforts to apply these principles to biological data toward questions of radiation biology. Here, we provide a review of radiogenomics modeling frameworks and efforts toward genomically guided radiotherapy. We first discuss medical oncology efforts to develop precision biomarkers. We next discuss similar efforts to create clinical assays for normal tissue or tumor radiosensitivity. We then discuss modeling frameworks for radiosensitivity and the evolution of ML to create predictive models for radiogenomics.

Spatial signature of dose patterns associated with acute radiation-induced lung damage in lung cancer patients treated with stereotactic body radiation therapy

Thoracic radiation therapy (RT) is often associated with lung side effects, whose etiology is still controversial. Our aim was to explore correlations between local dose in the thoracic anatomy and the radiation-induced lung damage (RILD). To this end, we designed a robust scheme for voxel-based analysis (VBA) to explore dose patterns associated with RILD in non-small-cell lung cancer (NSCLC) patients receiving stereotactic body RT (SBRT). We analyzed 106 NSCLC SBRT patients (median prescription dose: 50 Gy; range: [40-54] Gy) in 4 fractions (range: [3-5]) with clinical and dosimetric records suitable for the analysis. The incidence of acute G1 RILD (RTOG grade  ⩾  1) was 68%. Each planning CT and dose map was spatially normalized to a common anatomical reference using a B-spline inter-patient registration algorithm after masking the gross tumor volume. The tumor-subtracted dose maps were converted into biologically effective dose maps (α/β  =  3 Gy). VBA was performed according to a non-parametric permutation test accounting for multiple comparison, based on a cluster analysis method. The underlying general linear model of RILD was designed to include dose maps and each non-dosimetric variable significantly correlated with RILD. The clusters of voxels with dose differences significantly correlated with RILD at a given p -level (S _p) were generated. The only non-dosimetric variable significantly correlated with RILD was the chronic obstructive pulmonary disease (p   =  0.034). Patients with G1 RILD received significantly (p   ⩽  0.05) higher doses in two voxel clusters S _0.05 in the lower-left lung (14 cm³) and in an area (64 cm³) largely included within the ventricles. The applied VBA represents a powerful tool to probe the dose susceptibility of inhomogeneous organs in clinical radiobiology studies. The identified subregions with dose differences associated with G1 RILD in both the heart and lower lungs endorse a trend of previously reported hypotheses on lung toxicity radiobiology.

Combining handcrafted features with latent variables in machine learning for prediction of radiation-induced lung damage

PURPOSE

There has been burgeoning interest in applying machine learning methods for predicting radiotherapy outcomes. However, the imbalanced ratio of a large number of variables to a limited sample size in radiation oncology constitutes a major challenge. Therefore, dimensionality reduction methods can be a key to success. The study investigates and contrasts the application of traditional machine learning methods and deep learning approaches for outcome modeling in radiotherapy. In particular, new joint architectures based on variational autoencoder (VAE) for dimensionality reduction are presented and their application is demonstrated for the prediction of lung radiation pneumonitis (RP) from a large-scale heterogeneous dataset.

METHODS

A large-scale heterogeneous dataset containing a pool of 230 variables including clinical factors (e.g., dose, KPS, stage) and biomarkers (e.g., single nucleotide polymorphisms (SNPs), cytokines, and micro-RNAs) in a population of 106 nonsmall cell lung cancer (NSCLC) patients who received radiotherapy was used for modeling RP. Twenty-two patients had grade 2 or higher RP. Four methods were investigated, including feature selection (case A) and feature extraction (case B) with traditional machine learning methods, a VAE-MLP joint architecture (case C) with deep learning and lastly, the combination of feature selection and joint architecture (case D). For feature selection, Random forest (RF), Support Vector Machine (SVM), and multilayer perceptron (MLP) were implemented to select relevant features. Specifically, each method was run for multiple times to rank features within several cross-validated (CV) resampled sets. A collection of ranking lists were then aggregated by top 5% and Kemeny graph methods to identify the final ranking for prediction. A synthetic minority oversampling technique was applied to correct for class imbalance during this process. For deep learning, a VAE-MLP joint architecture where a VAE aimed for dimensionality reduction and an MLP aimed for classification was developed. In this architecture, reconstruction loss and prediction loss were combined into a single loss function to realize simultaneous training and weights were assigned to different classes to mitigate class imbalance. To evaluate the prediction performance and conduct comparisons, the area under receiver operating characteristic curves (AUCs) were performed for nested CVs for both handcrafted feature selections and the deep learning approach. The significance of differences in AUCs was assessed using the DeLong test of U-statistics.

RESULTS

An MLP-based method using weight pruning (WP) feature selection yielded the best performance among the different hand-crafted feature selection methods (case A), reaching an AUC of 0.804 (95% CI: 0.761-0.823) with 29 top features. A VAE-MLP joint architecture (case C) achieved a comparable but slightly lower AUC of 0.781 (95% CI: 0.737-0.808) with the size of latent dimension being 2. The combination of handcrafted features (case A) and latent representation (case D) achieved a significant AUC improvement of 0.831 (95% CI: 0.805-0.863) with 22 features (P-value = 0.000642 compared with handcrafted features only (Case A) and P-value = 0.000453 compared to VAE alone (Case C)) with an MLP classifier.

CONCLUSION

The potential for combination of traditional machine learning methods and deep learning VAE techniques has been demonstrated for dealing with limited datasets in modeling radiotherapy toxicities. Specifically, latent variables from a VAE-MLP joint architecture are able to complement handcrafted features for the prediction of RP and improve prediction over either method alone.

The Role of Machine Learning in Knowledge-Based Response-Adapted Radiotherapy

With the continuous increase in radiotherapy patient-specific data from multimodality imaging and biotechnology molecular sources, knowledge-based response-adapted radiotherapy (KBR-ART) is emerging as a vital area for radiation oncology personalized treatment. In KBR-ART, planned dose distributions can be modified based on observed cues in patients' clinical, geometric, and physiological parameters. In this paper, we present current developments in the field of adaptive radiotherapy (ART), the progression toward KBR-ART, and examine several applications of static and dynamic machine learning approaches for realizing the KBR-ART framework potentials in maximizing tumor control and minimizing side effects with respect to individual radiotherapy patients. Specifically, three questions required for the realization of KBR-ART are addressed: (1) what knowledge is needed; (2) how to estimate RT outcomes accurately; and (3) how to adapt optimally. Different machine learning algorithms for KBR-ART application shall be discussed and contrasted. Representative examples of different KBR-ART stages are also visited.

Machine Learning and Radiogenomics: Lessons Learned and Future Directions

Due to the rapid increase in the availability of patient data, there is significant interest in precision medicine that could facilitate the development of a personalized treatment plan for each patient on an individual basis. Radiation oncology is particularly suited for predictive machine learning (ML) models due to the enormous amount of diagnostic data used as input and therapeutic data generated as output. An emerging field in precision radiation oncology that can take advantage of ML approaches is radiogenomics, which is the study of the impact of genomic variations on the sensitivity of normal and tumor tissue to radiation. Currently, patients undergoing radiotherapy are treated using uniform dose constraints specific to the tumor and surrounding normal tissues. This is suboptimal in many ways. First, the dose that can be delivered to the target volume may be insufficient for control but is constrained by the surrounding normal tissue, as dose escalation can lead to significant morbidity and rare. Second, two patients with nearly identical dose distributions can have substantially different acute and late toxicities, resulting in lengthy treatment breaks and suboptimal control, or chronic morbidities leading to poor quality of life. Despite significant advances in radiogenomics, the magnitude of the genetic contribution to radiation response far exceeds our current understanding of individual risk variants. In the field of genomics, ML methods are being used to extract harder-to-detect knowledge, but these methods have yet to fully penetrate radiogenomics. Hence, the goal of this publication is to provide an overview of ML as it applies to radiogenomics. We begin with a brief history of radiogenomics and its relationship to precision medicine. We then introduce ML and compare it to statistical hypothesis testing to reflect on shared lessons and to avoid common pitfalls. Current ML approaches to genome-wide association studies are examined. The application of ML specifically to radiogenomics is next presented. We end with important lessons for the proper integration of ML into radiogenomics.

A multiobjective Bayesian networks approach for joint prediction of tumor local control and radiation pneumonitis in nonsmall-cell lung cancer (NSCLC) for response-adapted radiotherapy

PURPOSE

Individualization of therapeutic outcomes in NSCLC radiotherapy is likely to be compromised by the lack of proper balance of biophysical factors affecting both tumor local control (LC) and side effects such as radiation pneumonitis (RP), which are likely to be intertwined. Here, we compare the performance of separate and joint outcomes predictions for response-adapted personalized treatment planning.

METHODS

A total of 118 NSCLC patients treated on prospective protocols with 32 cases of local progression and 20 cases of RP grade 2 or higher (RP2) were studied. Sixty-eight patients with 297 features before and during radiotherapy were used for discovery and 50 patients were reserved for independent testing. A multiobjective Bayesian network (MO-BN) approach was developed to identify important features for joint LC/RP2 prediction using extended Markov blankets as inputs to develop a BN predictive structure. Cross-validation (CV) was used to guide the MO-BN structure learning. Area under the free-response receiver operating characteristic (AU-FROC) curve was used to evaluate joint prediction performance.

RESULTS

Important features including single nucleotide polymorphisms (SNPs), micro RNAs, pretreatment cytokines, pretreatment PET radiomics together with lung and tumor gEUDs were selected and their biophysical inter-relationships with radiation outcomes (LC and RP2) were identified in a pretreatment MO-BN. The joint LC/RP2 prediction yielded an AU-FROC of 0.80 (95% CI: 0.70-0.86) upon internal CV. This improved to 0.85 (0.75-0.91) with additional two SNPs, changes in one cytokine and two radiomics PET image features through the course of radiotherapy in a during-treatment MO-BN. This MO-BN model outperformed combined single-objective Bayesian networks (SO-BNs) during-treatment [0.78 (0.67-0.84)]. AU-FROC values in the evaluation of the MO-BN and individual SO-BNs on the testing dataset were 0.77 and 0.68 for pretreatment, and 0.79 and 0.71 for during-treatment, respectively.

CONCLUSIONS

MO-BNs can reveal possible biophysical cross-talks between competing radiotherapy clinical endpoints. The prediction is improved by providing additional during-treatment information. The developed MO-BNs can be an important component of decision support systems for personalized response-adapted radiotherapy.

A method to combine target volume data from 3D and 4D planned thoracic radiotherapy patient cohorts for machine learning applications

BACKGROUND AND PURPOSE

The gross tumour volume (GTV) is predictive of clinical outcome and consequently features in many machine-learned models. 4D-planning, however, has prompted substitution of the GTV with the internal gross target volume (iGTV). We present and validate a method to synthesise GTV data from the iGTV, allowing the combination of 3D and 4D planned patient cohorts for modelling.

MATERIAL AND METHODS

Expert delineations in 40 non-small cell lung cancer patients were used to develop linear fit and erosion methods to synthesise the GTV volume and shape. Quality was assessed using Dice Similarity Coefficients (DSC) and closest point measurements; by calculating dosimetric features; and by assessing the quality of random forest models built on patient populations with and without synthetic GTVs.

RESULTS

Volume estimates were within the magnitudes of inter-observer delineation variability. Shape comparisons produced mean DSCs of 0.8817 and 0.8584 for upper and lower lobe cases, respectively. A model trained on combined true and synthetic data performed significantly better than models trained on GTV alone, or combined GTV and iGTV data.

CONCLUSIONS

Accurate synthesis of GTV size from the iGTV permits the combination of lung cancer patient cohorts, facilitating machine learning applications in thoracic radiotherapy.

Radiation-induced lung toxicity in non-small-cell lung cancer: Understanding the interactions of clinical factors and cytokines with the dose-toxicity relationship

BACKGROUND AND PURPOSE

Current methods to estimate risk of radiation-induced lung toxicity (RILT) rely on dosimetric parameters. We aimed to improve prognostication by incorporating clinical and cytokine data, and to investigate how these factors may interact with the effect of mean lung dose (MLD) on RILT.

MATERIALS AND METHODS

Data from 125 patients treated from 2004 to 2013 with definitive radiotherapy for stages I-III NSCLC on four prospective clinical trials were analyzed. Plasma levels of 30 cytokines were measured pretreatment, and at 2 and 4weeks midtreatment. Penalized logistic regression models based on combinations of MLD, clinical factors, and cytokine levels were developed. Cross-validated estimates of log-likelihood and area under the receiver operating characteristic curve (AUC) were used to assess accuracy.

RESULTS

In prognosticating grade 3 or greater RILT by MLD alone, cross-validated log-likelihood and AUC were -28.2 and 0.637, respectively. Incorporating clinical features and baseline cytokine levels increased log-likelihood to -27.6 and AUC to 0.669. Midtreatment cytokine data did not further increase log-likelihood or AUC. Of the 30 cytokines measured, higher levels of 13 decreased the effect of MLD on RILT, corresponding to a lower odds ratio for RILT per Gy MLD, while higher levels of 4 increased the association.

CONCLUSIONS

Although the added prognostic benefit from cytokine data in our model was modest, understanding how clinical and biologic factors interact with the MLD-RILT relationship represents a novel framework for understanding and investigating the multiple factors contributing to radiation-induced toxicity.

Radiogenomics and radiotherapy response modeling

Advances in patient-specific information and biotechnology have contributed to a new era of computational medicine. Radiogenomics has emerged as a new field that investigates the role of genetics in treatment response to radiation therapy. Radiation oncology is currently attempting to embrace these recent advances and add to its rich history by maintaining its prominent role as a quantitative leader in oncologic response modeling. Here, we provide an overview of radiogenomics starting with genotyping, data aggregation, and application of different modeling approaches based on modifying traditional radiobiological methods or application of advanced machine learning techniques. We highlight the current status and potential for this new field to reshape the landscape of outcome modeling in radiotherapy and drive future advances in computational oncology.

Prediction of Postoperative Clinical Recovery of Drop Foot Attributable to Lumbar Degenerative Diseases, via a Bayesian Network

BACKGROUND

Drop foot resulting from degenerative lumbar diseases can impair activities of daily living. Therefore, predictors of recovery of this symptom have been investigated using univariate or/and multivariate analyses. However, the conclusions have been somewhat controversial. Bayesian network models, which are graphic and intuitive to the clinician, may facilitate understanding of the prognosis of drop foot resulting from degenerative lumbar diseases.

QUESTIONS/PURPOSES

(1) To show a layered correlation among predictors of recovery from drop foot resulting from degenerative lumbar diseases; and (2) to develop support tools for clinical decisions to treat drop foot resulting from lumbar degenerative diseases.

METHODS

Between 1993 and 2013, we treated 141 patients with decompressive lumbar spine surgery who presented with drop foot attributable to degenerative diseases. Of those, 102 (72%) were included in this retrospective study because they had drop foot of recent development and had no diseases develop that affect evaluation of drop foot after surgery. Specifically, 28 (20%) patients could not be analyzed because their records were not available at a minimum of 2 years followup after surgery and 11 (8%) were lost owing to postoperative conditions that affect the muscle strength evaluation. Eight candidate variables were sex, age, herniated soft disc, duration of the neurologic injury (duration), preoperative tibialis anterior muscle strength (pretibialis anterior), leg pain, cauda equina syndrome, and number of involved levels. Manual muscle testing was used to assess the tibialis anterior muscle strength. Drop foot was defined as a tibialis anterior muscle strength score of less than 3 of 5 (5 = movement against gravity and full resistance, 4 = movement against gravity and moderate resistance, 3 = movement against gravity through full ROM, 3- = movement against gravity through partial ROM, 2 = movement with gravity eliminated through full ROM, 1 = slight contraction but no movement, and 0 = no contraction). The two outcomes of interest were postoperative tibialis anterior muscle strength (posttibialis anterior) of 3 or greater and posttibialis anterior strength of 4 or greater at 2 years after surgery. We developed two separate Bayesian network models with outcomes of interest for posttibialis anterior strength of 3 or greater and posttibialis anterior strength of 4 or greater. The two outcomes correspond to "good" and "excellent" results based on previous reports, respectively. Direct predictors are defined as variables that have the tail of the arrow connecting the outcome of interest, whereas indirect predictors are defined as variables that have the tail of the arrow connecting either direct predictors or other indirect predictors that have the tail of the arrow connecting direct predictors. Sevenfold cross validation and receiver-operating characteristic (ROC) curve analyses were performed to evaluate the accuracy and robustness of the Bayesian network models.

RESULTS

Both of our Bayesian network models showed that weaker muscle power before surgery (pretibialis anterior ≤ 1) and longer duration of neurologic injury before treatment (> 30 days) were associated with a decreased likelihood of return of function by 2 years. The models for posttibialis anterior muscle strength of 3 or greater and posttibialis anterior muscle strength of 4 or greater were the same in terms of the graphs, showing that the two direct predictors were pretibialis anterior muscle strength (score ≤ 1 or ≥ 2) and duration (≤ 30 days or > 30 days). Age, herniated soft disc, and leg pain were identified as indirect predictors. We developed a decision-support tool in which the clinician can enter pretibialis anterior muscle strength and duration, and from this obtain the probability estimates of posttibialis anterior muscle strength. The probability estimates of posttibialis anterior muscle strength of 3 or greater and posttibialis anterior muscle strength of 4 or greater were 94% and 85%, respectively, in the most-favorable conditions (pretibialis anterior ≥ 2; duration ≤ 30 days) and 18% and 14%, respectively, in the least-favorable conditions (pretibialis anterior ≤ 1; duration > 30 days). On the sevenfold cross validation, the area under the ROC curve yielded means of 0.78 (95% CI, 0.68-0.87) and 0.74 (95% CI, 0.64-0.84) for posttibialis anterior muscle strength of 3 or greater and posttibialis anterior muscle strength of 4 or greater, respectively.

CONCLUSIONS

The results of this study suggest that the clinician can understand intuitively the layered correlation among predictors by Bayesian network models. Based on the models, the decision-support tool successfully provided the probability estimates of posttibialis anterior muscle strength to treat drop foot attributable to lumbar degenerative diseases. These models were shown to be robust on the internal validation but should be externally validated in other populations.

LEVEL OF EVIDENCE

Level III, therapeutic study.

Unraveling biophysical interactions of radiation pneumonitis in non-small-cell lung cancer via Bayesian network analysis

BACKGROUND

In non-small-cell lung cancer radiotherapy, radiation pneumonitis≥grade 2 (RP2) depends on patients' dosimetric, clinical, biological and genomic characteristics.

METHODS

We developed a Bayesian network (BN) approach to explore its potential for interpreting biophysical signaling pathways influencing RP2 from a heterogeneous dataset including single nucleotide polymorphisms, micro RNAs, cytokines, clinical data, and radiation treatment plans before and during the course of radiotherapy. Model building utilized 79 patients (21 with RP2) with complete data, and model testing used 50 additional patients with incomplete data. A developed large-scale Markov blanket approach selected relevant predictors. Resampling by k-fold cross-validation determined the optimal BN structure. Area under the receiver-operating characteristics curve (AUC) measured performance.

RESULTS

Pre- and during-treatment BNs identified biophysical signaling pathways from the patients' relevant variables to RP2 risk. Internal cross-validation for the pre-BN yielded an AUC=0.82 which improved to 0.87 by incorporating during treatment changes. In the testing dataset, the pre- and during AUCs were 0.78 and 0.82, respectively.

CONCLUSIONS

Our developed BN approach successfully handled a high number of heterogeneous variables in a small dataset, demonstrating potential for unraveling relevant biophysical features that could enhance prediction of RP2, although the current observations would require further independent validation.

The big data effort in radiation oncology: Data mining or data farming?

Although large volumes of information are entered into our electronic health care records, radiation oncology information systems and treatment planning systems on a daily basis, the goal of extracting and using this big data has been slow to emerge. Development of strategies to meet this goal is aided by examining issues with a data farming instead of a data mining conceptualization. Using this model, a vision of key data elements, clinical process changes, technology issues and solutions, and role for professional societies is presented. With a better view of technology, process and standardization factors, definition and prioritization of efforts can be more effectively directed.

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.

Using machine learning to predict radiation pneumonitis in patients with stage I non-small cell lung cancer treated with stereotactic body radiation therapy

To develop a patient-specific 'big data' clinical decision tool to predict pneumonitis in stage I non-small cell lung cancer (NSCLC) patients after stereotactic body radiation therapy (SBRT). 61 features were recorded for 201 consecutive patients with stage I NSCLC treated with SBRT, in whom 8 (4.0%) developed radiation pneumonitis. Pneumonitis thresholds were found for each feature individually using decision stumps. The performance of three different algorithms (Decision Trees, Random Forests, RUSBoost) was evaluated. Learning curves were developed and the training error analyzed and compared to the testing error in order to evaluate the factors needed to obtain a cross-validated error smaller than 0.1. These included the addition of new features, increasing the complexity of the algorithm and enlarging the sample size and number of events. In the univariate analysis, the most important feature selected was the diffusion capacity of the lung for carbon monoxide (DLCO adj%). On multivariate analysis, the three most important features selected were the dose to 15 cc of the heart, dose to 4 cc of the trachea or bronchus, and race. Higher accuracy could be achieved if the RUSBoost algorithm was used with regularization. To predict radiation pneumonitis within an error smaller than 10%, we estimate that a sample size of 800 patients is required. Clinically relevant thresholds that put patients at risk of developing radiation pneumonitis were determined in a cohort of 201 stage I NSCLC patients treated with SBRT. The consistency of these thresholds can provide radiation oncologists with an estimate of their reliability and may inform treatment planning and patient counseling. The accuracy of the classification is limited by the number of patients in the study and not by the features gathered or the complexity of the algorithm.

Big Data Analytics for Prostate Radiotherapy

Radiation therapy is a first-line treatment option for localized prostate cancer and radiation-induced normal tissue damage are often the main limiting factor for modern radiotherapy regimens. Conversely, under-dosing of target volumes in an attempt to spare adjacent healthy tissues limits the likelihood of achieving local, long-term control. Thus, the ability to generate personalized data-driven risk profiles for radiotherapy outcomes would provide valuable prognostic information to help guide both clinicians and patients alike. Big data applied to radiation oncology promises to deliver better understanding of outcomes by harvesting and integrating heterogeneous data types, including patient-specific clinical parameters, treatment-related dose-volume metrics, and biological risk factors. When taken together, such variables make up the basis for a multi-dimensional space (the "RadoncSpace") in which the presented modeling techniques search in order to identify significant predictors. Herein, we review outcome modeling and big data-mining techniques for both tumor control and radiotherapy-induced normal tissue effects. We apply many of the presented modeling approaches onto a cohort of hypofractionated prostate cancer patients taking into account different data types and a large heterogeneous mix of physical and biological parameters. Cross-validation techniques are also reviewed for the refinement of the proposed framework architecture and checking individual model performance. We conclude by considering advanced modeling techniques that borrow concepts from big data analytics, such as machine learning and artificial intelligence, before discussing the potential future impact of systems radiobiology approaches.

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

Prostate cancer is a frequently diagnosed malignancy worldwide and radiation therapy is a first-line approach in treating localized as well as locally advanced cases. The limiting factor in modern radiotherapy regimens is dose to normal structures, an excess of which can lead to aberrant radiation-induced toxicities. Conversely, dose reduction to spare adjacent normal structures risks underdosing target volumes and compromising local control. As a result, efforts aimed at predicting the effects of radiotherapy could invaluably optimize patient treatments by mitigating such toxicities and simultaneously maximizing biochemical control. In this work, we review the types of data, frameworks and techniques used for prostate radiotherapy outcome modeling. Consideration is given to clinical and dose-volume metrics, such as those amassed by the QUANTEC initiative, and also to newer methods for the integration of biological and genetic factors to improve prediction performance. We furthermore highlight trends in machine learning that may help to elucidate the complex pathophysiological mechanisms of tumor control and radiation-induced normal tissue side effects.

Bayesian network ensemble as a multivariate strategy to predict radiation pneumonitis risk

PURPOSE

Prediction of radiation pneumonitis (RP) has been shown to be challenging due to the involvement of a variety of factors including dose-volume metrics and radiosensitivity biomarkers. Some of these factors are highly correlated and might affect prediction results when combined. Bayesian network (BN) provides a probabilistic framework to represent variable dependencies in a directed acyclic graph. The aim of this study is to integrate the BN framework and a systems' biology approach to detect possible interactions among RP risk factors and exploit these relationships to enhance both the understanding and prediction of RP.

METHODS

The authors studied 54 nonsmall-cell lung cancer patients who received curative 3D-conformal radiotherapy. Nineteen RP events were observed (common toxicity criteria for adverse events grade 2 or higher). Serum concentration of the following four candidate biomarkers were measured at baseline and midtreatment: alpha-2-macroglobulin, angiotensin converting enzyme (ACE), transforming growth factor, interleukin-6. Dose-volumetric and clinical parameters were also included as covariates. Feature selection was performed using a Markov blanket approach based on the Koller-Sahami filter. The Markov chain Monte Carlo technique estimated the posterior distribution of BN graphs built from the observed data of the selected variables and causality constraints. RP probability was estimated using a limited number of high posterior graphs (ensemble) and was averaged for the final RP estimate using Bayes' rule. A resampling method based on bootstrapping was applied to model training and validation in order to control under- and overfit pitfalls.

RESULTS

RP prediction power of the BN ensemble approach reached its optimum at a size of 200. The optimized performance of the BN model recorded an area under the receiver operating characteristic curve (AUC) of 0.83, which was significantly higher than multivariate logistic regression (0.77), mean heart dose (0.69), and a pre-to-midtreatment change in ACE (0.66). When RP prediction was made only with pretreatment information, the AUC ranged from 0.76 to 0.81 depending on the ensemble size. Bootstrap validation of graph features in the ensemble quantified confidence of association between variables in the graphs where ten interactions were statistically significant.

CONCLUSIONS

The presented BN methodology provides the flexibility to model hierarchical interactions between RP covariates, which is applied to probabilistic inference on RP. The authors' preliminary results demonstrate that such framework combined with an ensemble method can possibly improve prediction of RP under real-life clinical circumstances such as missing data or treatment plan adaptation.

Application and Exploration of Big Data Mining in Clinical Medicine

OBJECTIVE

To review theories and technologies of big data mining and their application in clinical medicine.

DATA SOURCES

Literatures published in English or Chinese regarding theories and technologies of big data mining and the concrete applications of data mining technology in clinical medicine were obtained from PubMed and Chinese Hospital Knowledge Database from 1975 to 2015.

STUDY SELECTION

Original articles regarding big data mining theory/technology and big data mining's application in the medical field were selected.

RESULTS

This review characterized the basic theories and technologies of big data mining including fuzzy theory, rough set theory, cloud theory, Dempster-Shafer theory, artificial neural network, genetic algorithm, inductive learning theory, Bayesian network, decision tree, pattern recognition, high-performance computing, and statistical analysis. The application of big data mining in clinical medicine was analyzed in the fields of disease risk assessment, clinical decision support, prediction of disease development, guidance of rational use of drugs, medical management, and evidence-based medicine.

CONCLUSION

Big data mining has the potential to play an important role in clinical medicine.

Modeling the risk of radiation-induced lung fibrosis: Irradiated heart tissue is as important as irradiated lung

PURPOSE

We used normal tissue complication probability (NTCP) modeling to explore the impact of heart irradiation on radiation-induced lung fibrosis (RILF).

MATERIALS AND METHODS

We retrospectively reviewed for RILF 148 consecutive Hodgkin lymphoma (HL) patients treated with sequential chemo-radiotherapy (CHT-RT). Left, right, total lung and heart dose-volume and dose-mass parameters along with clinical, disease and treatment-related characteristics were analyzed. NTCP modeling by multivariate logistic regression analysis using bootstrapping was performed. Models were evaluated by Spearman Rs coefficient and ROC area.

RESULTS

At a median time of 13months, 18 out of 115 analyzable patients (15.6%) developed RILF after treatment. A three-variable predictive model resulted to be optimal for RILF. The two models most frequently selected by bootstrap included increasing age and mass of heart receiving >30Gy as common predictors, in combination with left lung V5 (Rs=0.35, AUC=0.78), or alternatively, the lungs near maximum dose D2% (Rs=0.38, AUC=0.80).

CONCLUSION

CHT-RT may cause lung injury in a small, but significant fraction of HL patients. Our results suggest that aging along with both heart and lung irradiation plays a fundamental role in the risk of developing RILF.