Image Based Data Mining Using Per-voxel Cox Regression

CT- and MR-based image-based data mining are consistent in the brain

PURPOSE

Image-based data mining (IBDM) is a voxel-based analysis technique to investigate dose-response. Most often, IBDM uses radiotherapy planning CTs because of their broad accessibility, however, it was unknown whether CT provided sufficient soft tissue contrast for brain IBDM. This study evaluates whether MR-based IBDM improves upon CT-based IBDM for studies of children with brain tumours.

METHODS

We compared IBDM pipelines using either CT- or MRI-based spatial normalisation in 128 children (ages 3.3-19.7 years) who received photon radiotherapy for primary brain tumours at a single institution. We quantified spatial-normalisation accuracy using contour comparison measures (centre-of-mass separation, average contour distance-to-agreement (DTavg), and Hausdorff distance) at multiple anatomic loci. We performed an end-to-end test of CT- and MRI-IBDM using modified clinical dose distributions and simulated effect labels to detect associations in pre-defined anatomic loci. Accuracy was assessed via sensitivity and specificity.

RESULTS

Spatial normalisation accuracy was comparable for both modalities, with a significant but small improvement for MRI compared to CT in all structures except the brainstem. The median (range) difference between the DTavg for the two pipelines was 0.37 (0.00-2.91) mm. The end-to-end test revealed no significant difference in sensitivity of the IBDM-identified regions for the two pipelines. Specificity slightly improved for MR-IBDM at the 99% significance level.

CONCLUSION

CT-based IBDM was comparable to MR-based IBDM, despite a small advantage in spatial normalisation accuracy with MRI. The use of CT-IBDM over MR-IBDM is useful for multi-institutional retrospective IBDM studies, where the availability of standardised MRI data can be limited.

Review and recommendations on deformable image registration uncertainties for radiotherapy applications

Deformable image registration (DIR) is a versatile tool used in many applications in radiotherapy (RT). DIR algorithms have been implemented in many commercial treatment planning systems providing accessible and easy-to-use solutions. However, the geometric uncertainty of DIR can be large and difficult to quantify, resulting in barriers to clinical practice. Currently, there is no agreement in the RT community on how to quantify these uncertainties and determine thresholds that distinguish a good DIR result from a poor one. This review summarises the current literature on sources of DIR uncertainties and their impact on RT applications. Recommendations are provided on how to handle these uncertainties for patient-specific use, commissioning, and research. Recommendations are also provided for developers and vendors to help users to understand DIR uncertainties and make the application of DIR in RT safer and more reliable.

Voxel-based analysis: Roadmap for clinical translation

Voxel-based analysis (VBA) allows the full, 3-dimensional, dose distribution to be considered in radiotherapy outcome analysis. This provides new insights into anatomical variability of pathophysiology and radiosensitivity by removing the need for a priori definition of organs assumed to drive the dose response associated with patient outcomes. This approach may offer powerful biological insights demonstrating the heterogeneity of the radiobiology across tissues and potential associations of the radiotherapy dose with further factors. As this methodological approach becomes established, consideration needs to be given to translating VBA results to clinical implementation for patient benefit. Here, we present a comprehensive roadmap for VBA clinical translation. Technical validation needs to demonstrate robustness to methodology, where clinical validation must show generalisability to external datasets and link to a plausible pathophysiological hypothesis. Finally, clinical utility requires demonstration of potential benefit for patients in order for successful translation to be feasible. For each step on the roadmap, key considerations are discussed and recommendations provided for best practice.

Dose outside of the prostate is associated with improved outcomes for high-risk prostate cancer patients treated with brachytherapy boost

Background

One in three high-risk prostate cancer patients treated with radiotherapy recur. Detection of lymph node metastasis and microscopic disease spread using conventional imaging is poor, and many patients are under-treated due to suboptimal seminal vesicle or lymph node irradiation. We use Image Based Data Mining (IBDM) to investigate association between dose distributions, and prognostic variables and biochemical recurrence (BCR) in prostate cancer patients treated with radiotherapy. We further test whether including dose information in risk-stratification models improves performance.

Method

Planning CTs, dose distributions and clinical information were collected for 612 high-risk prostate cancer patients treated with conformal hypo-fractionated radiotherapy, intensity modulated radiotherapy (IMRT), or IMRT plus a single fraction high dose rate (HDR) brachytherapy boost. Dose distributions (including HDR boost) of all studied patients were mapped to a reference anatomy using the prostate delineations. Regions where dose distributions significantly differed between patients that did and did-not experience BCR were assessed voxel-wise using 1) a binary endpoint of BCR at four-years (dose only) and 2) Cox-IBDM (dose and prognostic variables). Regions where dose was associated with outcome were identified. Cox proportional-hazard models with and without region dose information were produced and the Akaike Information Criterion (AIC) was used to assess model performance.

Results

No significant regions were observed for patients treated with hypo-fractionated radiotherapy or IMRT. Regions outside the target where higher dose was associated with lower BCR were observed for patients treated with brachytherapy boost. Cox-IBDM revealed that dose response was influenced by age and T-stage. A region at the seminal vesicle tips was identified in binary- and Cox-IBDM. Including the mean dose in this region in a risk-stratification model (hazard ratio=0.84, p=0.005) significantly reduced AIC values (p=0.019), indicating superior performance, compared with prognostic variables only. The region dose was lower in the brachytherapy boost patients compared with the external beam cohorts supporting the occurrence of marginal misses.

Conclusion

Association was identified between BCR and dose outside of the target region in high-risk prostate cancer patients treated with IMRT plus brachytherapy boost. We show, for the first-time, that the importance of irradiating this region is linked to prognostic variables.

Demystifying the Results of RTOG 0617: Identification of Dose Sensitive Cardiac Subregions Associated With Overall Survival

INTRODUCTION

The RTOG 0617 trial presented a worse survival for patients with lung cancer treated in the high-dose (74 Gy) arm. In multivariable models, radiation level and whole-heart volumetric dose parameters were associated with survival. In this work, we consider heart subregions to explain the observed survival difference between radiation levels.

METHODS

Voxel-based analysis identified anatomical regions where the dose was associated with survival. Bootstrapping clinical and dosimetric variables into an elastic net model selected variables associated with survival. Multivariable Cox regression survival models assessed the significance of dose to the heart subregion, compared with whole heart v5 and v30. Finally, the trial outcome was assessed after propensity score matching of patients on lung dose, heart subregion dose, and tumor volume.

RESULTS

A total of 458 patients were eligible for voxel-based analysis. A region of significance (p < 0.001) was identified in the base of the heart. Bootstrapping selected mean lung dose, radiation level, log tumor volume, and heart region dose. The multivariable Cox model exhibited dose to the heart region (p = 0.02), and tumor volume (p = 0.03) were significantly associated with survival, and radiation level was not significant (p = 0.07). The models exhibited that whole heart v5 and v30 were not associated with survival, with radiation level being significant (p < 0.05). In the matched cohort, no significant survival difference was seen between radiation levels.

CONCLUSIONS

Dose to the base of the heart is associated with overall survival, partly removing the radiation level effect, and explaining that worse survival in the high-dose arm is owing, in part, to the heart subregion dose. By defining a heart avoidance region, future dose escalation trials may be feasible.

Technical note: MAMBA-Multi-pAradigM voxel-Based Analysis: A computational cookbot

BACKGROUND

Voxel-Based (VB) analysis embraces a multifaceted ensemble of sophisticated techniques, lying at the boundary between image processing and statistical modeling, that allow for a frequentist inference of pathophysiological properties anchored to an anatomical reference. VB methods has been widely adopted in neuroimaging studies and, more recently, they are gaining momentum in radiation oncology research. However, the price for the power of VB analysis is the complexity of the underlying mathematics and algorithms.

PURPOSE

In this paper, we present the Multi-pAradigM voxel-Based Analysis (MAMBA) toolbox, which is intended for a flexible application of VB analysis in a wide variety of scenarios in medical imaging and radiation oncology.

METHODS

The MAMBA toolbox is implemented in Matlab. It provides open-source functions to compute VB statistical models of the input data, according to a great variety of regression schemes, and to derive VB maps of the observed significance level, performing a non-parametric permutation inference. The toolbox allows for including VB and global outcomes, as well as an arbitrary amount of VB and global Explanatory Variables (EVs). In addition, the Matlab Parallel Computing Toolbox is exploited to take advantage of the perfect parallelizability of most workloads.

RESULTS

The use of MAMBA was demonstrated by means of several realistic examples on a synthetic dataset mimicking a radiation oncology scenario.

CONCLUSION

MAMBA is an open-source toolbox, freely available for academic and non-commercial purposes. It is designed to make state-of-the-art VB analysis accessible to research scientists without the programming resources needed to build from scratch their own software solutions. At the same time, the source code is handed out for more experienced users to complement their own tools, also customizing user-defined models. MAMBA guarantees high generality and flexibility in the design of the statistical models, significantly expanding on the features of available free tools for VB analysis. The presented toolbox aims at increasing the reach of VB studies as well as the sharing of research results.

Cardiac Function Modifies the Impact of Heart Base Dose on Survival: A Voxel-Wise Analysis of Patients With Lung Cancer From the PET-Plan Trial

INTRODUCTION

Heart dose has emerged as an independent predictor of overall survival in patients with NSCLC treated with radiotherapy. Several studies have identified the base of the heart as a region of enhanced dose sensitivity and a potential target for cardiac sparing. We present a dosimetric analysis of overall survival in the multicenter, randomized PET-Plan trial (NCT00697333) and for the first time include left ventricular ejection fraction (EF) at baseline as a metric of cardiac function.

METHODS

A total of 205 patients with inoperable stage II or III NSCLC treated with 60 to 72 Gy in 2 Gy fractions were included in this study. A voxel-wise image-based data mining methodology was used to identify anatomical regions where higher dose was significantly associated with worse overall survival. Univariable and multivariable Cox proportional hazards models tested the association of survival with dose to the identified region, established prognostic factors, and baseline cardiac function.

RESULTS

A total of 172 patients remained after processing and censoring for follow-up. At 2-years posttreatment, a highly significant region was identified within the base of the heart (p < 0.005), centered on the origin of the left coronary artery and the region of the atrioventricular node. In multivariable analysis, the number of positron emission tomography-positive nodes (p = 0.02, hazard ratio = 1.13, 95% confidence interval: 1.02-1.25) and mean dose to the cardiac subregion (p = 0.02, hazard ratio = 1.11 Gy-1, 95% confidence interval: 1.02-1.21) were significantly associated with overall survival. There was a significant interaction between EF and region dose (p = 0.04) for survival, with contrast plots revealing a larger effect of region dose on survival in patients with lower EF values.

CONCLUSIONS

This work validates previous image-based data mining studies by revealing a strong association between dose to the base of the heart and overall survival. For the first time, an interaction between baseline cardiac health and heart base dose was identified, potentially suggesting preexisting cardiac dysfunction exacerbates the impact of heart dose on survival.

Image-based data mining applies to data collected from children

PURPOSE

Image-based data mining (IBDM) is a novel voxel-based method for analyzing radiation dose responses that has been successfully applied in adult data. Because anatomic variability and side effects of interest differ for children compared to adults, we investigated the feasibility of IBDM for pediatric analyses.

METHODS

We tested IBDM with CT images and dose distributions collected from 167 children (aged 10 months to 20 years) who received proton radiotherapy for primary brain tumors. We used data from four reference patients to assess IBDM sensitivity to reference selection. We quantified spatial-normalization accuracy via contour distances and deviations of the centers-of-mass of brain substructures. We performed dose comparisons with simplified and modified clinical dose distributions with a simulated effect, assessing their accuracy via sensitivity, positive predictive value (PPV) and Dice similarity coefficient (DSC).

RESULTS

Spatial normalizations and dose comparisons were insensitive to reference selection. Normalization discrepancies were small (average contour distance < 2.5 mm, average center-of-mass deviation < 6 mm). Dose comparisons identified differences (p < 0.01) in 81% of simplified and all modified clinical dose distributions. The DSCs for simplified doses were high (peak frequency magnitudes of 0.9-1.0). However, the PPVs and DSCs were low (maximum 0.3 and 0.4, respectively) in the modified clinical tests.

CONCLUSIONS

IBDM is feasible for childhood late-effects research. Our findings may inform cohort selection in future studies of pediatric radiotherapy dose responses and facilitate treatment planning to reduce treatment-related toxicities and improve quality of life among childhood cancer survivors.

Radial Data Mining to Identify Density-Dose Interactions That Predict Distant Failure Following SABR

Purpose

Lower dose outside the planned treatment area in lung stereotactic radiotherapy has been linked to increased risk of distant metastasis (DM) possibly due to underdosage of microscopic disease (MDE). Independently, tumour density on pretreatment computed tomography (CT) has been linked to risk of MDE. No studies have investigated the interaction between imaging biomarkers and incidental dose. The interaction would showcase whether the impact of dose on outcome is dependent on imaging and, hence, if imaging could inform which patients require dose escalation outside the gross tumour volume (GTV). We propose an image-based data mining methodology to investigate density-dose interactions radially from the GTV to predict DM with no a priori assumption on location.

Methods

Dose and density were quantified in 1-mm annuli around the GTV for 199 patients with early-stage lung cancer treated with 60 Gy in 5 fractions. Each annulus was summarised by three density and three dose parameters. For parameter combinations, Cox regressions were performed including a dose-density interaction in independent annuli. Heatmaps were created that described improvement in DM prediction due to the interaction. Regions of significant improvement were identified and studied in overall outcome models.

Results

Dose-density interactions were identified that significantly improved prediction for over 50% of bootstrap resamples. Dose and density parameters were not significant when the interaction was omitted. Tumour density variance and high peritumour density were associated with DM for patients with more cold spots (less than 30-Gy EQD2) and non-uniform dose about 3 cm outside of the GTV. Associations identified were independent of the mean GTV dose.

Conclusions

Patients with high tumour variance and peritumour density have increased risk of DM if there is a low and non-uniform dose outside the GTV. The dose regions are independent of tumour dose, suggesting that incidental dose may play an important role in controlling occult disease. Understanding such interactions is key to identifying patients who will benefit from dose-escalation. The methodology presented allowed spatial dose-density interactions to be studied at the exploratory stage for the first time. This could accelerate the clinical implementation of imaging biomarkers by demonstrating the impact of incidental dose for tumours of varying characteristics in routine data.

Deep Learning for Radiotherapy Outcome Prediction Using Dose Data - A Review

Artificial intelligence, and in particular deep learning using convolutional neural networks, has been used extensively for image classification and segmentation, including on medical images for diagnosis and prognosis prediction. Use in radiotherapy prognostic modelling is still limited, however, especially as applied to toxicity and tumour response prediction from radiation dose distributions. We review and summarise studies that applied deep learning to radiotherapy dose data, in particular studies that utilised full three-dimensional dose distributions. Ten papers have reported on deep learning models for outcome prediction utilising spatial dose information, whereas four studies used reduced dimensionality (dose volume histogram) information for prediction. Many of these studies suffer from the same issues that plagued early normal tissue complication probability modelling, including small, single-institutional patient cohorts, lack of external validation, poor data and model reporting, use of late toxicity data without taking time-to-event into account, and nearly exclusive focus on clinician-reported complications. They demonstrate, however, how radiation dose, imaging and clinical data may be technically integrated in convolutional neural networks-based models; and some studies explore how deep learning may help better understand spatial variation in radiosensitivity. In general, there are a number of issues specific to the intersection of radiotherapy outcome modelling and deep learning, for example translation of model developments into treatment plan optimisation, which will require further combined effort from the radiation oncology and artificial intelligence communities.

Current status and application of proton therapy for esophageal cancer

Esophageal cancer remains one of the leading causes of death from cancer across the world despite advances in multimodality therapy. Although early-stage disease can often be treated surgically, the current state of the art for locally advanced disease is concurrent chemoradiation, followed by surgery whenever possible. The uniform midline tumor location puts a strong importance on the need for precise delivery of radiation that would minimize dose to the heart and lungs, and the biophysical properties of proton beam makes this modality potential ideal for esophageal cancer treatment. This review covers the current state of knowledge of proton therapy for esophageal cancer, focusing on published retrospective single- and multi-institutional clinical studies, and emerging data from prospective clinical trials, that support the benefit of protons vs photon-based radiation in reducing postoperative complications, cardiac toxicity, and severe radiation induced immune suppression, which may improve survival outcomes for patients. In addition, we discuss the incorporation of immunotherapy to the curative management of esophageal cancers in the not-too-distant future. However, there is still a lack of high-level evidence to support proton therapy in the treatment of esophageal cancer, and proton therapy has its limitations in clinical application. It is expected to see the results of future large-scale randomized clinical trials and the continuous improvement of proton radiotherapy technology.

Probing thoracic dose patterns associated to pericardial effusion and mortality in patients treated with photons and protons for locally advanced non-small-cell lung cancer

PURPOSE

To investigate thoracic dose-response patterns for pericardial effusion (PCE) and mortality in patients treated for locally advanced Non-Small-Cell Lung Cancer (NSCLC) by Intensity Modulated RT (IMRT) or Passive-Scattering Proton Therapy (PSPT).

METHODS

Among 178 patients, 43.5% developed grade ≥ 2 PCE. Clinical and dosimetric factors associated with PCE or overall survival (OS) were identified via multi-variable Cox proportional hazards modeling. The Voxel-Based Analyses (VBAs) of local dose differences between patients with and without PCE and mortality was performed. The robustness of VBA results was assessed by a novel characterization of spatial properties of dose distributions based on probabilistic independent component analysis (PICA) and connectograms.

RESULTS

Several non-dosimetric variables were selected by the multivariable analysis for the considered outcomes, while the time-dependent PCE onset was uncorrelated with the OS (p = 0.34) at a multi-variable Cox analysis. Despite the significant PSPT dosimetric advantage, the RT technique did not affect the occurrence of PCE or OS. VBAs highlighted largely overlapping clusters significantly associated with PCE endpoints in heart and lungs. No significant dosimetric patterns related to mortality endpoints were found. PICA identified 43 components homogeneously scattered within thorax, while connectograms showed modest correlations between doses in main cardio-pulmonary substructures.

CONCLUSIONS

Spatially resolved analysis highlighted dose patterns related to radiation-induced cardiac toxiciy and the observed organ-based dose-response mismatch in PSPT and IMRT. Indeed, the thoracic regions spared by PSPT poorly overlapped with the areas involved in PCE development, as highlited by VBA. PICA and connectograms proved valuable tools for assessing the robusteness of obtained VBA inferences.

Spatial descriptions of radiotherapy dose: normal tissue complication models and statistical associations

For decades, dose-volume information for segmented anatomy has provided the essential data for correlating radiotherapy dosimetry with treatment-induced complications. Dose-volume information has formed the basis for modelling those associations via normal tissue complication probability (NTCP) models and for driving treatment planning. Limitations to this approach have been identified. Many studies have emerged demonstrating that the incorporation of information describing the spatial nature of the dose distribution, and potentially its correlation with anatomy, can provide more robust associations with toxicity and seed more general NTCP models. Such approaches are culminating in the application of computationally intensive processes such as machine learning and the application of neural networks. The opportunities these approaches have for individualising treatment, predicting toxicity and expanding the solution space for radiation therapy are substantial and have clearly widespread and disruptive potential. Impediments to reaching that potential include issues associated with data collection, model generalisation and validation. This review examines the role of spatial models of complication and summarises relevant published studies. Sources of data for these studies, appropriate statistical methodology frameworks for processing spatial dose information and extracting relevant features are described. Spatial complication modelling is consolidated as a pathway to guiding future developments towards effective, complication-free radiotherapy treatment.

Novel methodology to assess the effect of contouring variation on treatment outcome

PURPOSE

Contouring variation is one of the largest systematic uncertainties in radiotherapy, yet its effect on clinical outcome has never been analyzed quantitatively. We propose a novel, robust methodology to locally quantify target contour variation in a large patient cohort and find where this variation correlates with treatment outcome. We demonstrate its use on biochemical recurrence for prostate cancer patients.

METHOD

We propose to compare each patient's target contours to a consistent and unbiased reference. This reference was created by auto-contouring each patient's target using an externally trained deep learning algorithm. Local contour deviation measured from the reference to the manual contour was projected to a common frame of reference, creating contour deviation maps for each patient. By stacking the contour deviation maps, time to event was modeled pixel-wise using a multivariate Cox proportional hazards model (CPHM). Hazard ratio (HR) maps for each covariate were created, and regions of significance found using cluster-based permutation testing on the z-statistics. This methodology was applied to clinical target volume (CTV) contours, containing only the prostate gland, from 232 intermediate- and high-risk prostate cancer patients. The reference contours were created using ADMIRE® v3.4 (Elekta AB, Sweden). Local contour deviations were computed in a spherical coordinate frame, where differences between reference and clinical contours were projected in a 2D map corresponding to sampling across the coronal and transverse angles every 3°. Time to biochemical recurrence was modeled using the pixel-wise CPHM analysis accounting for contour deviation, patient age, Gleason score, and treated CTV volume.

RESULTS

We successfully applied the proposed methodology to a large patient cohort containing data from 232 patients. In this patient cohort, our analysis highlighted regions where the contour variation was related to biochemical recurrence, producing expected and unexpected results: (a) the interface between prostate-bladder and prostate-seminal vesicle interfaces where increase in the manual contour relative to the reference was related to a reduction of risk of biochemical recurrence by 4-8% per mm and (b) the prostate's right, anterior and posterior regions where an increase in the manual contour relative to the reference contours was related to an increase in risk of biochemical recurrence by 8-24% per mm.

CONCLUSION

We proposed and successfully applied a novel methodology to explore the correlation between contour variation and treatment outcome. We analyzed the effect of contour deviation of the prostate CTV on biochemical recurrence for a cohort of more than 200 prostate cancer patients while taking basic clinical variables into account. Applying this methodology to a larger dataset including additional clinically important covariates and externally validating it can more robustly identify regions where contour variation directly relates to treatment outcome. For example, in the prostate case we use to demonstrate our novel methodology, external validation will help confirm or reject the counter-intuitive results (larger contours resulting in higher risk). Ultimately, the results of this methodology could inform contouring protocols based on actual patient outcomes.