BACKGROUND:Chemoradiotherapy (CRT) plays a key role in treating esophageal cancer (EC) but is associated with significant toxicity. Proton radiotherapy (PRT) may reduce this risk by limiting radiation dose to organs at risk. In the Netherlands, PRT is reimbursed only when eligibility criteria defined in a National Indication Protocol for Proton Therapy (NIPP) are met. This study describes the development and implementation of such a protocol for EC based on model-based selection. MATERIALS AND METHODS:A national multidisciplinary working group was formed aiming to develop the NIPP protocol through literature review, evaluation of prediction models, external model validation, and stakeholder engagement. RESULTS:Cardiac events and overall survival were identified as the most clinically relevant endpoints. As no existing models met the quality criteria for model-based selection, a validated 2-year mortality prediction model originally developed for lung cancer was externally validated in EC patients treated with definitive (dCRT) or neoadjuvant (nCRT) CRT. The model performed well after updates to the intercept (both cohorts) and slope (nCRT). Model-based selection was defined as a ≥ 5 % predicted absolute reduction in 2-year mortality with PRT versus photon radiotherapy, using Mean Heart Dose (MHD) and Gross Tumor Volume (GTV) as predictors. Additional selection criteria included WHO performance status 0-2 and exclusion of cT4, cN3, or cM1 status. The NIPP was approved by the Dutch Society of Radiation Oncology in June 2021 and by the Dutch Health Care Institute in October 2021. National implementation and prospective outcome evaluation are ongoing via the ProTRAIT registry. CONCLUSION:A national indication protocol enabling model-based selection for PRT in EC was successfully implemented in the Netherlands. The NIPP describes eligibility criteria for proton therapy reimbursement and enables reimbursement for individual patients who are expected to benefit from PRT.

Background and purposeRetrospective dose evaluation for organ-at-risk auto-contours has previously used small cohorts due to additional manual effort required for treatment planning on auto-contours. We aimed to do this at large scale, by a) proposing and assessing an automated plan optimization workflow that used existing clinical plan parameters and b) using it for head-and-neck auto-contour dose evaluation.Materials and methodsOur automated workflow emulated our clinic’s treatment planning protocol and reused existing clinical plan optimization parameters. This workflow recreated the original clinical plan (POG) with manual contours (PMC) and evaluated the dose effect (POG-PMC) on 70 photon and 30 proton plans of head-and-neck patients. As a use-case, the same workflow (and parameters) created a plan using auto-contours (PAC) of eight head-and-neck organs-at-risk from a commercial tool and evaluated their dose effect (PMC-PAC).ResultsFor plan recreation (POG-PMC), our workflow had a median impact of 1.0% and 1.5% across dose metrics of auto-contours, for photon and proton respectively. Computer time of automated planning was 25% (photon) and 42% (proton) of manual planning time. For auto-contour evaluation (PMC-PAC), we noticed an impact of 2.0% and 2.6% for photon and proton radiotherapy. All evaluations had a median ΔNTCP (Normal Tissue Complication Probability) less than 0.3%.ConclusionsThe plan replication capability of our automated program provides a blueprint for other clinics to perform auto-contour dose evaluation with large patient cohorts. Finally, despite geometric differences, auto-contours had a minimal median dose impact, hence inspiring confidence in their utility and facilitating their clinical adoption.

Real-world problems are often multi-objective, with decision-makers unable to specify a priori which trade-off between the conflicting objectives is preferable. Intuitively, building machine learning solutions in such cases would entail providing multiple predictions that span and uniformly cover the Pareto front of all optimal trade-off solutions. We propose a novel approach for multi-objective training of neural networks to approximate the Pareto front during inference. In our approach, we train the neural networks multi-objectively using a dynamic loss function, wherein each network’s losses (corresponding to multiple objectives) are weighted by their hypervolume maximizing gradients. Experiments on different multi-objective problems show that our approach returns well-spread outputs across different trade-offs on the approximated Pareto front without requiring the trade-off vectors to be specified a priori. Further, results of comparisons with the state-of-the-art approaches highlight the added value of our proposed approach, especially in cases where the Pareto front is asymmetric.

Manual or automatic delineation of the esophageal tumor in CT images is known to be very challenging. This is due to the low contrast between the tumor and adjacent tissues, the anatomical variation of the esophagus, as well as the occasional presence of foreign bodies (e.g. feeding tubes). Physicians therefore usually exploit additional knowledge such as endoscopic findings, clinical history, additional imaging modalities like PET scans. Achieving his additional information is time-consuming, while the results are error-prone and might lead to non-deterministic results. In this paper we aim to investigate if and to what extent a simplified clinical workflow based on CT alone, allows one to automatically segment the esophageal tumor with sufficient quality. For this purpose, we present a fully automatic end-to-end esophageal tumor segmentation method based on convolutional neural networks (CNNs). The proposed network, called Dilated Dense Attention Unet (DDAUnet), leverages spatial and channel attention gates in each dense block to selectively concentrate on determinant feature maps and regions. Dilated convolutional layers are used to manage GPU memory and increase the network receptive field. We collected a dataset of 792 scans from 288 distinct patients including varying anatomies with air pockets, feeding tubes and proximal tumors. Repeatability and reproducibility studies were conducted for three distinct splits of training and validation sets. The proposed network achieved a DSC value of 0.79 ± 0.20, a mean surface distance of 5.4 ± 20.2mm and 95% Hausdorff distance of 14.7 ± 25.0mm for 287 test scans, demonstrating promising results with a simplified clinical workflow based on CT alone. Our code is publicly available via https://github.com/yousefis/DenseUnet_Esophagus_Segmentation.

Real-world problems are often multi-objective with decision-makers unable to specify a priori which trade-off between the conflicting objectives is preferable. Intuitively, building machine learning solutions in such cases would entail providing multiple predictions that span and uniformly cover the Pareto front of all optimal trade-off solutions. We propose a novel approach for multi-objective training of neural networks to approximate the Pareto front during inference. In our approach, the neural networks are trained multi-objectively using a dynamic loss function, wherein each network's losses (corresponding to multiple objectives) are weighted by their hypervolume maximizing gradients. We discuss and illustrate why training processes to approximate Pareto fronts need to optimize on fronts of individual training samples instead of on only the front of average losses. Experiments on three multi-objective problems show that our approach returns outputs that are well-spread across different trade-offs on the approximated Pareto front without requiring the trade-off vectors to be specified a priori. Further, results of comparisons with the state-of-the-art approaches highlight the added value of our proposed approach, especially in asymmetric Pareto fronts.

Background and purposeAccess to healthcare data is indispensable for scientific progress and innovation. Sharing healthcare data is time-consuming and notoriously difficult due to privacy and regulatory concerns. The Personal Health Train (PHT) provides a privacy-by-design infrastructure connecting FAIR (Findable, Accessible, Interoperable, Reusable) data sources and allows distributed data analysis and machine learning. Patient data never leaves a healthcare institute.Materials and methodsLung cancer patient-specific databases (tumor staging and post-treatment survival information) of oncology departments were translated according to a FAIR data model and stored locally in a graph database. Software was installed locally to enable deployment of distributed machine learning algorithms via a central server. Algorithms (MATLAB, code and documentation publicly available) are patient privacy-preserving as only summary statistics and regression coefficients are exchanged with the central server. A logistic regression model to predict post-treatment two-year survival was trained and evaluated by receiver operating characteristic curves (ROC), root mean square prediction error (RMSE) and calibration plots.ResultsIn 4 months, we connected databases with 23 203 patient cases across 8 healthcare institutes in 5 countries (Amsterdam, Cardiff, Maastricht, Manchester, Nijmegen, Rome, Rotterdam, Shanghai) using the PHT. Summary statistics were computed across databases. A distributed logistic regression model predicting post-treatment two-year survival was trained on 14 810 patients treated between 1978 and 2011 and validated on 8 393 patients treated between 2012 and 2015.ConclusionThe PHT infrastructure demonstrably overcomes patient privacy barriers to healthcare data sharing and enables fast data analyses across multiple institutes from different countries with different regulatory regimens. This infrastructure promotes global evidence-based medicine while prioritizing patient privacy.

Manual or automatic delineation of the esophageal tumor in CT images is known to be very challenging. This is due to the low contrast between the tumor and adjacent tissues, the anatomical variation of the esophagus, as well as the occasional presence of foreign bodies (e.g. feeding tubes). Physicians therefore usually exploit additional knowledge such as endoscopic findings, clinical history, additional imaging modalities like PET scans. Achieving his additional information is time-consuming, while the results are error-prone and might lead to non-deterministic results. In this paper we aim to investigate if and to what extent a simplified clinical workflow based on CT alone, allows one to automatically segment the esophageal tumor with sufficient quality. For this purpose, we present a fully automatic end-to-end esophageal tumor segmentation method based on convolutional neural networks (CNNs). The proposed network, called Dilated Dense Attention Unet (DDAUnet), leverages spatial and channel attention gates in each dense block to selectively concentrate on determinant feature maps and regions. Dilated convolutional layers are used to manage GPU memory and increase the network receptive field. We collected a dataset of 792 scans from 288 distinct patients including varying anatomies with \mbox{air pockets}, feeding tubes and proximal tumors. Repeatability and reproducibility studies were conducted for three distinct splits of training and validation sets. The proposed network achieved a $\mathrm{DSC}$ value of $0.79 \pm 0.20$, a mean surface distance of $5.4 \pm 20.2mm$ and $95\%$ Hausdorff distance of $14.7 \pm 25.0mm$ for 287 test scans, demonstrating promising results with a simplified clinical workflow based on CT alone. Our code is publicly available via \url{this https URL}.

Background: Tumor hypoxia increases resistance to radiotherapy and systemic therapy. Our aim was to develop and validate a disease-agnostic and disease-specific CT (+FDG-PET) based radiomics hypoxia classification signature. Material and methods: A total of 808 patients with imaging data were included: N = 100 training/N = 183 external validation cases for a disease-agnostic CT hypoxia classification signature, N = 76 training/N = 39 validation cases for the H&N CT signature and N = 62 training/N = 36 validation cases for the Lung CT signature. The primary gross tumor volumes (GTV) were manually defined by experts on CT. In order to dichotomize between hypoxic/well-oxygenated tumors a threshold of 20% was used for the [F-18]-HX4-derived hypoxic fractions (HF). A random forest (RF)-based machine-learning classifier/regressor was trained to classify patients as hypoxia-positive/negative based on radiomic features. Results: A 11 feature "disease-agnostic CT model" reached AUC's of respectively 0.78 (95% confidence interval [CI], 0.62-0.94), 0.82 (95% CI, 0.67-0.96) and 0.78 (95% CI, 0.67-0.89) in three external validation datasets. A "disease-agnostic FDG-PET model" reached an AUC of 0.73 (0.95% CI, 0.49-0.97) in validation by combining 5 features. The highest "lung-specific CT model" reached an AUC of 0.80 (0.95% CI, 0.65-0.95) in validation with 4 CT features, while the "H&N-specific CT model" reached an AUC of 0.84 (0.95% CI, 0.64-1.00) in validation with 15 CT features. A tumor volume-alone model was unable to significantly classify patients as hypoxia-positive/negative. A significant survival split (P = 0.037) was found between CT-classified hypoxia strata in an external H&N cohort (n = 517), while 117 significant hypoxia gene-CT signature feature associations were found in an external lung cohort (n = 80). Conclusion: The disease-specific radiomics signatures perform better than the disease agnostic ones. By identifying hypoxic patients our signatures have the potential to enrich interventional hypoxia-targeting trials. (C) 2020 The Author(s). Published by Elsevier B.V.

Background and purpose: A higher radiation dose to the heart is known to be associated with increased mortality in non-small cell lung cancer (NSCLC) patients. It is however unknown what the contribution of the heart dose is when other risk factors for mortality are also accounted for. Materials and methods: We constructed and externally validated prediction models of mortality after definitive chemoradiotherapy for NSCLC. Models were developed in 145 stage I-IIIB NSCLC patients. Clinical (performance status, age, gross tumour volume (GTV) combining primary tumour and involved lymph nodes, current smoker) and dosimetric (mean lung (MLD) and heart (MHD) dose) variables were considered. Multivariable logistic regression models predicting 12 and 24 month mortality were built in 5-fold cross-validation. Discrimination and calibration was assessed in 3 external validation datasets containing 878 (via distributed learning), 127 and 96 NSCLC patients. Results: The best discriminating prediction models combined GTV, smoker and/or MHD: bootstrapping AUC (95% CI) of 0.74 (0.66-0.78) and 0.69 (0.55-0.74) at 12 and 24 months. At external validation, the 24 month mortality GTV-smoker-MHD model robustly showed moderate discrimination (AUC = 0.61- 0.64 before and 0.64-0.65 after model update) with limited 0.01-0.07 improvement over a GTV-only model, and calibration slope (0.64-0.65). This model can identify patients for whom a MHD reduction may be useful (e.g. PPV = 77%, NPV = 52% (60% cut-off)). Conclusions: Tumour volume is strongly related to mortality risk in the first 2 years after chemoradiotherapy for NSCLC. Modelling indicates that efforts to reduce cardiac dose may be relevant for small tumours and that smoking has an important negative association with survival. (C) 2019 Elsevier B.V. All rights reserved.

AbstractPre-requisites to better understand the chapter:knowledge of the major steps and procedures of developing a clinical prediction model.Logical position of the chapter with respect to the previous chapter:in the last chapters, you have learned how to develop and validate a clinical prediction model. You have been learning logistic regression as main algorithm to build the model. However, several different more complex algorithms can be used to build a clinical prediction model. In this chapter, the main machine learning based algorithms will be presented to you.Learning objectives:you will be presented with the definitions of: machine learning, supervised and unsupervised learning. The major algorithms for the last two categories will be introduced.

AbstractPre-requisites to better understand the chapter: basic knowledge of major sources of clinical data.Logical position of the chapter with respect to the previous chapter: in the previous chapter, you have learned what the major sources of clinical data are. In this chapter, we will dive into the main characteristics of presented data sources. In particular, we will learn how to distinguish and classify data according to its scale.Learning objectives: you will learn the major differences between data sources presented in previous chapters; how clinical data can be classified according to its scale. You will get familiar with the concept of ‘big’ clinical data; you will learn which are the major concerns limiting ‘big’ data exchange.