Critical volume model analysis of lung complication data from different strains of mice

Mechanistic model of radiotherapy-induced lung fibrosis using coupled 3D agent-based and Monte Carlo simulations

BACKGROUND

Mechanistic modelling of normal tissue toxicities is unfolding as an alternative to the phenomenological normal tissue complication probability models. The latter, currently used in the clinics, rely exclusively on limited patient data and neglect spatial dose distribution information. Among the various approaches, agent-based models are appealing as they provide the means to include patient-specific parameters and simulate long-term effects in complex systems. However, Monte Carlo tools remain the state-of-the-art for modelling radiation transport and provide measurements of the delivered dose with unmatched precision.

METHODS

In this work, we develop and characterize a coupled 3D agent-based - Monte Carlo model that mechanistically simulates the onset of the radiation-induced lung fibrosis in an alveolar segment. To the best of our knowledge, this is the first such model.

RESULTS

Our model replicates extracellular matrix patterns, radiation-induced lung fibrosis severity indexes and functional subunits survivals that show qualitative agreement with experimental studies and are consistent with our past results. Moreover, in accordance with experimental results, higher functional subunits survival and lower radiation-induced lung fibrosis severity indexes are achieved when a 5-fractions treatment is simulated. Finally, the model shows increased sensitivity to more uniform protons dose distributions with respect to more heterogeneous ones from photon irradiation.

CONCLUSIONS

This study lays thus the groundwork for further investigating the effects of different radiotherapeutic treatments on the onset of radiation-induced lung fibrosis via mechanistic modelling.

An Agent-Based Model of Radiation-Induced Lung Fibrosis

Early- and late-phase radiation-induced lung injuries, namely pneumonitis and lung fibrosis (RILF), severely constrain the maximum dose and irradiated volume in thoracic radiotherapy. As the most radiosensitive targets, epithelial cells respond to radiation either by undergoing apoptosis or switching to a senescent phenotype that triggers the immune system and damages surrounding healthy cells. Unresolved inflammation stimulates mesenchymal cells' proliferation and extracellular matrix (ECM) secretion, which irreversibly stiffens the alveolar walls and leads to respiratory failure. Although a thorough understanding is lacking, RILF and idiopathic pulmonary fibrosis share multiple pathways and would mutually benefit from further insights into disease progression. Furthermore, current normal tissue complication probability (NTCP) models rely on clinical experience to set tolerance doses for organs at risk and leave aside mechanistic interpretations of the undergoing processes. To these aims, we implemented a 3D agent-based model (ABM) of an alveolar duct that simulates cell dynamics and substance diffusion following radiation injury. Emphasis was placed on cell repopulation, senescent clearance, and intra/inter-alveolar bystander senescence while tracking ECM deposition. Our ABM successfully replicates early and late fibrotic response patterns reported in the literature along with the ECM sigmoidal dose-response curve. Moreover, surrogate measures of RILF severity via a custom indicator show qualitative agreement with published fibrosis indices. Finally, our ABM provides a fully mechanistic alveolar survival curve highlighting the need to include bystander damage in lung NTCP models.

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.

Preclinical Evaluation of Dose-Volume Effects and Lung Toxicity Occurring In and Out-of-Field

PURPOSE

The aim of this study was to define the dose and dose-volume relationship of radiation-induced pulmonary toxicities occurring in and out-of-field in mouse models of early inflammatory and late fibrotic response.

MATERIALS AND METHODS

Early radiation-induced inflammation and fibrosis were investigated in C3H/NeJ and C57BL/6J mice, respectively. Animals were irradiated with 20 Gy delivered to the upper region of the right lung as a single fraction or as 3 consecutive fractions using the Small Animal Radiation Research Platform (Xstrahl Inc, Camberley, UK). Cone beam computed tomography was performed for image guidance before irradiation and to monitor late toxicity. Histologic sections were examined for neutrophil and macrophage infiltration as markers of early inflammatory response and type I collagen staining as a marker of late-occurring fibrosis. Correlation was evaluated with the dose-volume histogram parameters calculated for individual mice and changes in the observed cone beam computed tomography values.

RESULTS

Mean lung dose and the volume receiving over 10 Gy (V10) showed significant correlation with late responses for single and fractionated exposures in directly targeted volumes. Responses observed outside the target volume were attributed to nontargeted effects and showed no dependence on either mean lung dose or V10.

CONCLUSIONS

Quantitative assessment of normal tissue response closely correlates early and late pulmonary response with clinical parameters, demonstrating this approach as a potential tool to facilitate clinical translation of preclinical studies. Out-of-field effects were observed but did not correlate with dosimetric parameters, suggesting that nontargeted effects may have a role in driving toxicities outside the treatment field.

A survey of changing trends in modelling radiation lung injury in mice: bringing out the good, the bad, and the uncertain

Within this millennium there has been resurgence in funding and research dealing with animal models of radiation-induced lung injury to identify and establish predictive biomarkers and effective mitigating agents that are applicable to humans. Most have been performed on mice but there needs to be assurance that the emphasis on such models is not misplaced. We therefore considered it timely to perform a comprehensive appraisal of the literature dealing with radiation lung injury of mice and to critically evaluate the validity and clinical relevance of the research. A total of 357 research papers covering the period of 1970-2015 were extensively reviewed. Whole thorax irradiation (WTI) has become the most common treatment for studying lung injury in mice and distinct trends were seen with regard to the murine strain, radiation dose, intended pathology investigated, length of study, and assays. Recently, the C57BL/6 strain has been increasingly used in the majority of these studies with the notion that they are susceptible to pulmonary fibrosis. Nonetheless, many of these investigations depend on animal survival as the primary end point and neglect the importance of radiation pneumonitis and the anomaly of lethal pleural effusions. A relatively large variation in survival times of C5BL/6 mice is also seen among different institutions pointing to the need for standardization of radiation treatments and environmental conditions. An analysis of mitigating drug treatments is complicated by the fact that the majority of studies are limited to the C57BL/6 strain with a premature termination of the experiments and do not establish whether the treatment actually prevents or simply delays the progression of radiation injury. This survey of the literature has pointed to several improvements that need to be considered in establishing a reliable preclinical murine model of radiation lung injury. The lethality end point should also be used cautiously and with greater emphasis on other assays such as non-invasive lung functional and imaging monitoring in order to quantify specific pulmonary injury that can be better extrapolated to radiation toxicity encountered in our own species.

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.

Mechanistic modelling of radiotherapy-induced lung toxicity

OBJECTIVE

This work explores the biological basis of a mechanistic model of radiation-induced lung damage; uniquely, the model makes a connection between the cellular radiobiology involved in lung irradiation and the full three-dimensional distribution of radiation dose.

METHODS

Local tissue damage and loss of global organ function, in terms of radiation pneumonitis (RP), were modelled as different levels of radiation injury. Parameters relating to the former could be derived from the local dose-response function, and the latter from the volume effect of the organ. The literature was consulted to derive information on a threshold dose and volume-effect mechanisms.

RESULTS

Simulations of local tissue damage supported the alveolus as a functional subunit (FSU) which can be regenerated from a single surviving stem cell. A moderate interpatient variation in stem cell radiosensitivity (15%) resulted in a great variation in tissue response between 8 and 20 Gy. The threshold of FSU inactivation within a critical functioning volume leading to RP was found to be approximately 47% and the degree of health status variation (influencing the volume effect) in a population was estimated at 25%.

CONCLUSION

This work has shown that it is possible to make sense of the way the lung responds to radiation by modelling RP mechanistically, from cell death to tissue damage to loss of organ function.

ADVANCES IN KNOWLEDGE

Simulations were able to provide parameter values, currently not available in the literature, related to the response of the lung to irradiation.

Mechanistic simulation of normal-tissue damage in radiotherapy--implications for dose-volume analyses

A radiobiologically based 3D model of normal tissue has been developed in which complications are generated when 'irradiated'. The aim is to provide insight into the connection between dose-distribution characteristics, different organ architectures and complication rates beyond that obtainable with simple DVH-based analytical NTCP models. In this model the organ consists of a large number of functional subunits (FSUs), populated by stem cells which are killed according to the LQ model. A complication is triggered if the density of FSUs in any 'critical functioning volume' (CFV) falls below some threshold. The (fractional) CFV determines the organ architecture and can be varied continuously from small (series-like behaviour) to large (parallel-like). A key feature of the model is its ability to account for the spatial dependence of dose distributions. Simulations were carried out to investigate correlations between dose-volume parameters and the incidence of 'complications' using different pseudo-clinical dose distributions. Correlations between dose-volume parameters and outcome depended on characteristics of the dose distributions and on organ architecture. As anticipated, the mean dose and V(20) correlated most strongly with outcome for a parallel organ, and the maximum dose for a serial organ. Interestingly better correlation was obtained between the 3D computer model and the LKB model with dose distributions typical for serial organs than with those typical for parallel organs. This work links the results of dose-volume analyses to dataset characteristics typical for serial and parallel organs and it may help investigators interpret the results from clinical studies.

Analysis of normal tissue complication probability of the lung using a reliability model

The volume effect of normal tissues and organs is an important factor for predicting normal tissue complication probability (NTCP) following partial, heterogeneous irradiation of organs at risk, and reducing the late sequela by conformal radiation therapy. We have previously developed a reliability model for calculation of NTCP, assuming a parallel architecture of functional subunits (FSU), where a critical number (k) out of the total number of FSUs (N) must be intact for the organ to maintain its function. Published data on radiation-induced lethal pneumonitis and altered breathing rate following partial volume irradiation of the mouse lung were analysed, and critical fraction and corresponding spatial density distribution of FSUs were estimated using this model. The critical fraction (k/N) seemed to be similar for the two endpoints, and a value of 0.7 was found to provide good fit to the experimental data. The critical fraction did not vary throughout the lung, and variation in volume effect cannot therefore be attributed to heterogeneous tissue architecture. On the other hand, our analysis revealed that the observed variation in volume effect of mouse lung may be attributed to heterogeneous spatial distribution in FSU density or also the spatial variation in inactivation probability of the FSUs.