Search

We introduce CQnet, a neural network with origins in the CQ algorithm for solving convex split-feasibility problems and forward-backward splitting. CQnet's trajectories are interpretable as particles that are tracking a changing constraint set via its point-to-set distance function while being elements of another constraint set at every layer. More than just a convex-geometric interpretation, CQnet accommodates learned and deterministic constraints that may be sample or data-specific and are satisfied by every layer and the output. Furthermore, the states in CQnet progress toward another constraint set at every layer. We provide proof of stability/nonexpansiveness with minimal assumptions. The combination of constraint handling and stability put forward CQnet as a candidate for various tasks where prior knowledge exists on the network states or output., 12 pages, 7 figures

Convolutional Neural Networks (CNN) have recently seen tremendous success in various computer vision tasks. However, their application to problems with high dimensional input and output, such as high-resolution image and video segmentation or 3D medical imaging, has been limited by various factors. Primarily, in the training stage, it is necessary to store network activations for back propagation. In these settings, the memory requirements associated with storing activations can exceed what is feasible with current hardware, especially for problems in 3D. Motivated by the propagation of signals over physical networks, that are governed by the hyperbolic Telegraph equation, in this work we introduce a fully conservative hyperbolic network for problems with high dimensional input and output. We introduce a coarsening operation that allows completely reversible CNNs by using a learnable Discrete Wavelet Transform and its inverse to both coarsen and interpolate the network state and change the number of channels. We show that fully reversible networks are able to achieve results comparable to the state of the art in 4D time-lapse hyper spectral image segmentation and full 3D video segmentation, with a much lower memory footprint that is a constant independent of the network depth. We also extend the use of such networks to Variational Auto Encoders with high resolution input and output., 21 pages, 9 figures, Updated work to include additional numerical experiments, a section about VAEs and learnable wavelets

Purpose The aim of this study is to assess how clinical outcomes in real-world (effectiveness) correspond to the outcomes in clinical trials (efficacy) of systemic treatments for extensive disease small cell lung cancer (ED SCLC).Methods All patients diagnosed with ED SCLC between 2008 and 2014 in six Dutch large teaching hospitals (Santeon network) were identified and followed-up from date of diagnosis until death or end of data collection. For every patient, an efficacy-effectiveness factor (EE factor) was calculated by dividing individual patients' overall survival (OS) by the pooled median OS assessed from clinical trials with the respective treatment.Results From 792 diagnosed patients, 568 (72%) started with first-line treatment. Overall, the median EE factor was 0.79 (P = 2) and a higher age at diagnosis (age >= 65 years) were independent predictors for a lower EE factor. The EE gap was 43% in patients with both age >= 65 years and ECOG >= 2 (EE factor 0.57). The mean age and the proportion of patients with ECOG >= 2 in real-world were different from those in clinical trials (mean age of 66 versus 62 years, and ECOG >= 2 25% versus 17%; both P Conclusion OS of patients with ED SCLC treated with systemic therapy in real-world practice is 21% shorter than for patients included in trials. Age at diagnosis and performance status partly explain this gap.

BACKGROUND: The Value-Based Health Care concept defines patient value as patient-relevant outcomes divided by costs. The aim of the present study was to assess the development of systemic treatment costs over the years compared with changes in overall survival (OS) at the level of a diagnosis of stage IV non-small cell lung cancer (NSCLC). METHODS: All patients diagnosed (in 2008-2014) with stage IV NSCLC and treated with systemic treatment in six Dutch large teaching hospitals (Santeon network) were included. We collected data on OS and amounts of drug units (milligrams) for every drug in the applied systemic cancer treatments, until death. These amounts were multiplied by Dutch unit costs (Euros/mg) expressed in 2018 Euros to construct total drug costs per line of treatment per patient. Costs for day care visits were added for drugs requiring parenteral administration. RESULTS: Data were collected from 1214 patients. Median OS and mean total drug costs showed no significant variation over the years (p = 0.437 and p = 0.693, respectively). Mean total drug costs per 1 year of survival ranged from €20,665 to €26,438 during the period under study. Costs for first-line systemic treatment were significantly higher in 2011-2014 compared with 2008-2010. CONCLUSION: This study shows that overall drug costs were stable over the years, despite a relative increase in first-line treatment costs. Median OS remained at around 8 months from year to year. These trend data are very relevant as background for the assessment of costs and achieved outcomes in the more recent years.

This retrospective study was performed to evaluate real-world oncological outcomes of patients treated with chemo-based therapy for muscle-invasive or metastatic bladder cancer (MIBC/mBC) and compare results to data from RCTs and other cohorts. Among 1578 patients diagnosed, 470 (30%) had MIBC/mBC. Median overall survival (mOS) for RC alone (47 months), first-line (13 months) and second-line (7 months) chemotherapy, and chemotherapy for recurrent disease (8 months) were similar to literature. Treatment with neoadjuvant and induction chemotherapy (NAIC) was only utilized in 9% of patients, and often in patients with poor disease status, resulting in a lower mOS compared to literature (35 and 20 months, respectively). Patients treated with chemotherapy had many adversities to treatment, with only 50%, 13%, 18% and 7% of patients in NAIC, first-line, salvage after RC, and second-line setting completing the full pre-planned chemotherapy treatment. Real-world data shows NAIC before RC is underutilized. Adversities during chemotherapy treatment are frequent, with many patients requiring dose reduction or early treatment termination, resulting in poor treatment response. Although treatment efficacy between RCTs and real-world patients is quite similar, there are large differences in baseline characteristics and treatment patterns. Possibly, results from retrospective studies on real-world data can deliver missing evidence on efficacy of chemotherapy treatment on older and ‘unfit’ patients.

When pixel-level masks or partial annotations are not available for training neural networks for semantic segmentation, it is possible to use higher-level information in the form of bounding boxes, or image tags. In the imaging sciences, many applications do not have an object-background structure and bounding boxes are not available. Any available annotation typically comes from ground truth or domain experts. A direct way to train without masks is using prior knowledge on the size of objects/classes in the segmentation. We present a new algorithm to include such information via constraints on the network output, implemented via projection-based point-to-set distance functions. This type of distance functions always has the same functional form of the derivative, and avoids the need to adapt penalty functions to different constraints, as well as issues related to constraining properties typically associated with non-differentiable functions. Whereas object size information is known to enable object segmentation from bounding boxes from datasets with many general and medical images, we show that the applications extend to the imaging sciences where data represents indirect measurements, even in the case of single examples. We illustrate the capabilities in case of a) one or more classes do not have any annotation; b) there is no annotation at all; c) there are bounding boxes. We use data for hyperspectral time-lapse imaging, object segmentation in corrupted images, and sub-surface aquifer mapping from airborne-geophysical remote-sensing data. The examples verify that the developed methodology alleviates difficulties with annotating non-visual imagery for a range of experimental settings.

Recently established equivalences between differential equations and the structure of neural networks enabled some interpretation of training of a neural network as partial-differential-equation (PDE) constrained optimization. We add to the previously established connections, explicit regularization that is particularly beneficial in the case of single large-scale examples with partial annotation. We show that explicit regularization of model parameters in PDE constrained optimization translates to regularization of the network output. Examination of the structure of the corresponding Lagrangian and backpropagation algorithm do not reveal additional computational challenges. A hyperspectral imaging example shows that minimum prior information together with cross-validation for optimal regularization parameters boosts the segmentation accuracy.

The large spatial/frequency scale of hyperspectral and airborne magnetic and gravitational data causes memory issues when using convolutional neural networks for (sub-) surface characterization. Recently developed fully reversible networks can mostly avoid memory limitations by virtue of having a low and fixed memory requirement for storing network states, as opposed to the typical linear memory growth with depth. Fully reversible networks enable the training of deep neural networks that take in entire data volumes, and create semantic segmentations in one go. This approach avoids the need to work in small patches or map a data patch to the class of just the central pixel. The cross-entropy loss function requires small modifications to work in conjunction with a fully reversible network and learn from sparsely sampled labels without ever seeing fully labeled ground truth. We show examples from land-use change detection from hyperspectral time-lapse data, and regional aquifer mapping from airborne geophysical and geological data.

Recently established equivalences between differential equations and the structure of neural networks enabled some interpretation of training of a neural network as partial-differential-equation (PDE) constrained optimization. We add to the previously established connections, explicit regularization that is particularly beneficial in the case of single large-scale examples with partial annotation. We show that explicit regularization of model parameters in PDE constrained optimization translates to regularization of the network output. Examination of the structure of the corresponding Lagrangian and backpropagation algorithm do not reveal additional computational challenges. A hyperspectral imaging example shows that minimum prior information together with cross-validation for optimal regularization parameters boosts the segmentation accuracy.

Many works on inverse problems in the imaging sciences consider regularization via one or more penalty functions or constraint sets. When the models/images are not easily described using one or a few penalty functions/constraints, additive model descriptions for regularization lead to better imaging results. These include cartoon-texture decomposition, morphological component analysis, and robust principal component analysis; methods that typically rely on penalty functions. We propose a regularization framework, based on the Minkowski set, that merges the strengths of additive models and constrained formulations. We generalize the Minkowski set, such that the model parameters are the sum of two components, each of which is constrained to an intersection of sets. Furthermore, the sum of the components is also an element of another intersection of sets. These generalizations allow us to include multiple pieces of prior knowledge on each of the components, as well as on the sum of components, which is necessary to ensure physical feasibility of partial-differential-equation based parameters estimation problems. We derive the projection operation onto the generalized Minkowski sets and construct an algorithm based on the alternating direction method of multipliers. We illustrate how we benefit from using more prior knowledge in the form of the generalized Minkowski set using seismic waveform inversion and video background-anomaly separation., 18 pages, 3 figures

Factors that limit the size of the input and output of a neural network include memory requirements for the network states/activations to compute gradients, as well as memory for the convolutional kernels or other weights. The memory restriction is especially limiting for applications where we want to learn how to map volumetric data to the desired output, such as video-to-video. Recently developed fully reversible neural networks enable gradient computations using storage of the network states for a couple of layers only. While this saves a tremendous amount of memory, it is the convolutional kernels that take up most memory if fully reversible networks contain multiple invertible pooling/coarsening layers. Invertible coarsening operators such as the orthogonal wavelet transform cause the number of channels to grow explosively. We address this issue by combining fully reversible networks with layers that contain the convolutional kernels in a compressed form directly. Specifically, we introduce a layer that has a symmetric block-low-rank structure. In spirit, this layer is similar to bottleneck and squeeze-and-expand structures. We contribute symmetry by construction, and a combination of notation and flattening of tensors allows us to interpret these network structures in linear algebraic fashion as a block-low-rank matrix in factorized form and observe various properties. A video segmentation example shows that we can train a network to segment the entire video in one go, which would not be possible, in terms of memory requirements, using non-reversible networks and previously proposed reversible networks.

Inverse problems in the imaging sciences encompass a variety of applications. The primary problem of interest is the identification of physical parameters from observed data that come from experiments governed by partial-differential-equations. The secondary type of imaging problems attempts to reconstruct images and video that are corrupted by, for example, noise, subsampling, blur, or saturation. The quality of the solution of an inverse problem is sensitive to issues such as noise and missing entries in the data. The non-convex seismic full-waveform inversion problem suffers from parasitic local minima that lead to wrong solutions that may look realistic even for noiseless data. To meet some of these challenges, I propose solution strategies that constrain the model parameters at every iteration to help guide the inversion. To arrive at this goal, I present new practical workflows, algorithms, and software, that avoid manual tuning-parameters and that allow us to incorporate multiple pieces of prior knowledge. Opposed to penalty methods, I avoid balancing the influence of multiple pieces of prior knowledge by working with intersections of constraint sets. I explore and present advantages of constraints for imaging. Because the resulting problems are often non-trivial to solve, especially on large 3D grids, I introduce faster algorithms, dedicated to computing projections onto intersections of multiple sets. To connect prior knowledge more directly to problem formulations, I also combine ideas from additive models, such as cartoon-texture decomposition and robust principal component analysis, with intersections of multiple constraint sets for the regularization of inverse problems. The result is an extension of the concept of a Minkowski set. Examples from non-unique physical parameter estimation problems show that constraints in combination with projection methods provide control over the model properties at every iteration. This can lead to improved results when the constraints are carefully relaxed.

Detecting a specific horizon in seismic images is a valuable tool for geological interpretation. Because hand-picking the locations of the horizon is a time-consuming process, automated computational methods were developed starting three decades ago. Older techniques for such picking include interpolation of control points however, in recent years neural networks have been used for this task. Until now, most networks trained on small patches from larger images. This limits the networks ability to learn from large-scale geologic structures. Moreover, currently available networks and training strategies require label patches that have full and continuous annotations, which are also time-consuming to generate. We propose a projected loss-function for training convolutional networks with a multi-resolution structure, including variants of the U-net. Our networks learn from a small number of large seismic images without creating patches. The projected loss-function enables training on labels with just a few annotated pixels and has no issue with the other unknown label pixels. Training uses all data without reserving some for validation. Only the labels are split into training/testing. Contrary to other work on horizon tracking, we train the network to perform non-linear regression, and not classification. As such, we propose labels as the convolution of a Gaussian kernel and the known horizon locations that indicate uncertainty in the labels. The network output is the probability of the horizon location. We demonstrate the proposed computational ingredients on two different datasets, for horizon extrapolation and interpolation. We show that the predictions of our methodology are accurate even in areas far from known horizon locations because our learning strategy exploits all data in large seismic images., 24 pages, 13 figures

The divergence between clinical trial results and real-world outcomes is largely unknown for many cancer types. The present study aims overall to assess the efficacy–effectiveness gap (difference between outcomes in clinical trials and the real world) in systemic treatment for metastatic nonsmall cell lung cancer (NSCLC). All patients diagnosed with stage IV NSCLC between 2008 and 2014 within a network of seven Dutch large teaching hospitals (Santeon) were studied. For every patient, an efficacy–effectiveness (EE) factor was calculated by dividing individual patients' overall survival (OS) by the pooled median OS assessed from clinical trials with the respective treatment. From 2989 diagnosed patients, 1214 (41%) started with first-line treatment. For all studied regimens, real-world OS was shorter than OS reported in clinical trials. Overall, the EE factor was 0.77 (95% CI 0.70–0.85; p, Survival of patients with metastatic NSCLC treated with chemotherapy or targeted therapy in real-world practice is nearly one quarter shorter than for patients included in clinical trials. These real-world data provide useful information for clinicians. http://ow.ly/Khd230minFF

BACKGROUND: The present study aims to give a detailed overview of day-to-day practice in the systemic treatment of NSCLC stage IIIB/IV and its clinical outcomes in six large teaching hospitals in the Netherlands in the period 2008-2012. METHODS: A retrospective observational cohort study was conducted in the Care for Outcome registry. Patients diagnosed with stage IIIB/IV NSCLC were included and drug data were collected. Outcomes included percentage of patients treated with systemic treatment, percentage of different first line treatment options, survival, and number and percentage of switches, dose reductions (

Introduction The growing interest in comparative effectiveness research (CER) based on data from routine clinical practice also extends towards lung oncology. Although CER studies using real world data (RWD) have the potential to assist clinical decision-making, concerns about the quality and validity of studies with observational data subsist. The primary objective of the present study is to assess the current status of observational CER in the field of lung oncology, both quantitatively as qualitatively. Methods We performed a systematic electronic literature database search in MEDLINE and EMBASE (up to 1 July 2015). The quality of all selected studies was assessed according to the Good ReseArch for Comparative Effectiveness (GRACE) checklist. Results The first selection included 657 publications. After screening the corresponding abstracts and full-text papers, 38 studies remained. A total of 36 studies included patients with advanced NSCLC. The comparison of the effectiveness of gefitinib versus erlotinib was the main objective in 22% of the studies. The median number of patients per study was 202 (range 21–10064). The number of publications increased over the years whereas the quality score remained stable over the years with several common shortcomings (checklist items M5, D1, D4, D6). Discussion The growing interest in clinical oncology CER studies using RWD is reflected in an increasing number of publications in the recent years. The studies have several common methodological shortcomings possibly limiting their applicability in clinical decision-making. To fulfil the promise of RWD CER in lung oncology effort should be continued to overcome these shortcomings.

BACKGROUND: The present study aims to give a detailed overview of day-to-day practice in the systemic treatment of NSCLC stage IIIB/IV and its clinical outcomes in six large teaching hospitals in the Netherlands in the period 2008-2012. METHODS: A retrospective observational cohort study was conducted in the Care for Outcome registry. Patients diagnosed with stage IIIB/IV NSCLC were included and drug data were collected. Outcomes included percentage of patients treated with systemic treatment, percentage of different first line treatment options, survival, and number and percentage of switches, dose reductions (

Height is a classic complex trait with common variants in a growing list of genes known to contribute to the phenotype. Using a genecentric genotyping array targeted toward cardiovascular-related loci, comprising 49,320 SNPs across approximately 2000 loci, we evaluated the association of common and uncommon SNPs with adult height in 114,223 individuals from 47 studies and six ethnicities. A total of 64 loci contained a SNP associated with height at array-wide significance (p < 2.4 × 10−6), with 42 loci surpassing the conventional genome-wide significance threshold (p < 5 × 10−8). Common variants with minor allele frequencies greater than 5% were observed to be associated with height in 37 previously reported loci. In individuals of European ancestry, uncommon SNPs in IL11 and SMAD3, which would not be genotyped with the use of standard genome-wide genotyping arrays, were strongly associated with height (p < 3 × 10−11). Conditional analysis within associated regions revealed five additional variants associated with height independent of lead SNPs within the locus, suggesting allelic heterogeneity. Although underpowered to replicate findings from individuals of European ancestry, the direction of effect of associated variants was largely consistent in African American, South Asian, and Hispanic populations. Overall, we show that dense coverage of genes for uncommon SNPs, coupled with large-scale meta-analysis, can successfully identify additional variants associated with a common complex trait.

Height is a classic complex trait with common variants in a growing list of genes known to contribute to the phenotype. Using a genecentric genotyping array targeted toward cardiovascular-related loci, comprising 49,320 SNPs across approximately 2000 loci, we evaluated the association of common and uncommon SNPs with adult height in 114,223 individuals from 47 studies and six ethnicities. Atotal of 64 loci contained a SNP associated with height at array-wide significance (p < 2.4× 10(-6)), with 42 loci surpassing the conventional genome-wide significance threshold (p < 5× 10(-8)). Common variants with minor allele frequencies greater than 5% were observed to be associated with height in 37 previously reported loci. In individuals of European ancestry, uncommon SNPs in IL11 and SMAD3, which would not be genotyped with the use of standard genome-wide genotyping arrays, were strongly associated with height (p < 3× 10(-11)). Conditional analysis within associated regions revealed five additional variants associated with height independent of lead SNPs within the locus, suggesting allelic heterogeneity. Although underpowered to replicate findings from individuals of European ancestry, the direction of effect of associated variants was largely consistent in African American, South Asian, and Hispanic populations. Overall, we show that dense coverage of genes for uncommon SNPs, coupled with large-scale meta-analysis, can successfully identify additional variants associated with a common complex trait.

In developed countries cardiovascular disease is one of the leading causes of death. Cardiovascular drugs such as platelet aggregation inhibitors, oral anticoagulants, antihypertensives and cholesterol lowering drugs are abundantly prescribed to reduce risk of cardiovascular disease. Notable interindividual variation exists in the response to these pharmacotherapeutic interventions, which can be partially explained by factors such as gender, age, diet, concomitant drug use and environmental factors. Notwithstanding, a great part of this variability remains unknown. To a smaller or larger extent, genetic variability may contribute to the variability in response to these cardiovascular drugs. This review gives an overview of pharmacogenetic studies of genes that were reported to be associated with four commonly prescribed drugs/drug classes (platelet aggregation inhibitors, coumarins, antihypertensives and statins) and were studied at least 2 times with a similar outcome measure. In the field of cardiovascular drug therapy, polymorphisms in candidate genes such as the cycloxygenase-1, vitamin K reductase complex subunit 1, CYP2C9, alpha adducin and 3-hydroxy-3-methylglutaryl-CoA reductase have received a great amount of interest in the pharmacogenetics of aspirin, coumarins, antihypertensives and statins respectively. However, only variations in VKORC1 and CYP2C9 have consistently been associated with drug response (coumarins) and have clinical implications. Clinical trials should provide evidence for the effectiveness of genotyping before this procedure will be a part of every day anticoagulant therapy. In spite of the tremendous amount of publications in this field, there is no reason to advocate for genetic testing for any other drugs cardiovascular drug therapy yet. Current approaches in pharmacogenetic research do not seem to lead to results that meet our expectations of individualized medicine. Therefore, new approaches are needed addressing issues and challenges such as the number of SNPs studied, study power, study design and application of new statistical methods in (pharmaco-)genetic analysis.