Search

Radiotherapy (RT) datasets can suffer from variations in annotation of organ at risk (OAR) and target structures. Annotation standards exist, but their description for prostate targets is limited. This restricts the use of such data for supervised machine learning purposes as it requires properly annotated data. The aim of this work was to develop a modality independent deep learning (DL) model for automatic classification and annotation of prostate RT DICOM structures. Delineated prostate organs at risk (OAR), support‐ and target structures (gross tumor volume [GTV]/clinical target volume [CTV]/planning target volume [PTV]), along with or without separate vesicles and/or lymph nodes, were extracted as binary masks from 1854 patients. An image modality independent 2D InceptionResNetV2 classification network was trained with varying amounts of training data using four image input channels. Channel 1–3 consisted of orthogonal 2D projections from each individual binary structure. The fourth channel contained a summation of the other available binary structure masks. Structure classification performance was assessed in independent CT (n = 200 pat) and magnetic resonance imaging (MRI) (n = 40 pat) test datasets and an external CT (n = 99 pat) dataset from another clinic. A weighted classification accuracy of 99.4% was achieved during training. The unweighted classification accuracy and the weighted average F1 score among different structures in the CT test dataset were 98.8% and 98.4% and 98.6% and 98.5% for the MRI test dataset, respectively. The external CT dataset yielded the corresponding results 98.4% and 98.7% when analyzed for trained structures only, and results from the full dataset yielded 79.6% and 75.2%. Most misclassifications in the external CT dataset occurred due to multiple CTVs and PTVs being fused together, which was not included in the training data. Our proposed DL‐based method for automated renaming and standardization of prostate radiotherapy annotations shows great potential. Clinic specific contouring standards however need to be represented in the training data for successful use. Source code is available at https://github.com/jamtheim/DicomRTStructRenamerPublic

In this paper, we examine the problem of target detection for the multistatic passive radar. Passive radar systems leverage the existing wireless sources, such as radio/TV stations and cellular signals that are referred to as illuminators of opportunity (IOs), to illuminate the environment and provide surveillance functions. Usually, these IO source signals are sparse or locally sparse in the frequency domain. We develop a passive multistatic detector by exploiting the sparsity or local sparsity of the IO signals. To improve the computational efficiency, two fast implementations of the proposed detector are also introduced. Simulation results show that the proposed approaches outperform the conventional passive detection methods that model the IO signals as unknown without any specific structures.

Identification of prostate gold fiducial markers in magnetic resonance imaging (MRI) images is challenging when CT images are not available, due to misclassifications from intra-prostatic calcifications. It is also a time consuming task and automated identification methods have been suggested as an improvement for both objectives. Multi-echo gradient echo (MEGRE) images have been utilized for manual fiducial identification with 100% detection accuracy. The aim is therefore to develop an automatic deep learning based method for fiducial identification in MRI images intended for MRI-only prostate radiotherapy. MEGRE images from 326 prostate cancer patients with fiducials were acquired on a 3T MRI, post-processed with N4 bias correction, and the fiducial center of mass (CoM) was identified. A 9 mm radius sphere was created around the CoM as ground truth. A deep learning HighRes3DNet model for semantic segmentation was trained using image augmentation. The model was applied to 39 MRI-only patients and 3D probability maps for fiducial location and segmentation were produced and spatially smoothed. In each of the three largest probability peaks, a 9 mm radius sphere was defined. Detection sensitivity and geometric accuracy was assessed. To raise awareness of potential false findings a ‘BeAware’ score was developed, calculated from the total number and quality of the probability peaks. All datasets, annotations and source code used were made publicly available. The detection sensitivity for all fiducials were 97.4%. Thirty-six out of thirty-nine patients had all fiducial markers correctly identified. All three failed patients generated a user notification using the BeAware score. The mean absolute difference between the detected fiducial and ground truth CoM was 0.7 ± 0.9 [0 3.1] mm. A deep learning method for automatic fiducial identification in MRI images was developed and evaluated with state-of-the-art results. The BeAware score has the potential to notify the user regarding patients where the proposed method is uncertain.

In this paper, we propose a gridless method for estimating an unknown number of fundamental frequencies. Starting with a conventional dictionary matrix, containing sets of candidate fundamental frequencies and their corresponding harmonics, a nonconvex log-sum cost function is formed such that it imposes the harmonic structure and treats every fundamental frequency in the dictionary as a parameter. The cost function is iteratively decreased by minimizing a surrogate function, and, in each iteration, the fundamental frequencies are refined, whereas redundant parameters are omitted from the dictionary. The proposed method is tested on both real and simulated data, showing its preferred performance as compared to other state-of-the-art multipitch estimators.

In this paper, we propose a blind channel deconvolution method based on the Least Absolute Shrinkage and Selection Operator (LASSO). Using a Toeplitz structured formulation of the received signal, we form an iterative blind deconvolution scheme, alternatively estimating the underwater impulse response, and alternatively the transmitted waveform. The resulting optimizations are convex, and we formulate a computationally efficient solver using the Alternating Direction Method of Multipliers (ADMM). We illustrate the performance of the resulting estimator using both simulated and measured underwater signals.

Nonuniform sampling (NUS) of multidimensional NMR data offers significant time savings while improving spectral resolution or increasing sensitivity per unit time. However, NUS has not been widely used for quantitative analysis because of the nonlinearity of most methods used to model NUS data, which leads to problems in estimating signal intensities, relaxation rate constants, and their error bounds. Here, we present an approach that avoids these limitations by combining accordion spectroscopy and NUS in the indirect dimensions of multidimensional spectra and then applying sparse exponential mode analysis, which is well suited for analyzing accordion-type relaxation data in a NUS context. By evaluating the Cramer-Rao lower bound of the variances of the estimated relaxation rate constants, we achieve a robust benchmark for the underlying reconstruction model. Furthermore, we design NUS schemes optimized with respect to the information theoretical lower bound of the error in the parameters of interest, given a specified number of sampling points. The accordion-NUS method compares favorably with conventional relaxation experiments in that it produces identical results, within error, while shortening the length of the experiment by an order of magnitude. Thus, our approach enables rapid acquisition of NMR relaxation data for optimized use of spectrometer time or accurate measurements on samples of limited lifetime.

Passive radar employs existing illuminators of opportunity (IOs) to probe the environment and perform surveillance functions. Such IO signals are often sparse or locally sparse in the frequency domain. In this paper, we propose a new target detector for multistatic passive radars with multiple distributed receivers by exploiting the sparsity structure of the IO waveform. Two fast implementations of the proposed detector are derived to improve the computational efficiency. Simulation results show that the proposed approaches outperform conventional passive detection methods that models the IO signals as unknown without any specific structures, especially when the observation data size is limited.

The decomposition of nonlinear chirp modes is a challenging task, typically requiring prior knowledge of the number of modes a signal contains. In this work, we present a greedy nonlinear chirp mode estimation (NCME) technique that forms the used decomposition basis from the signal itself, using an arctangent demodulation technique. The resulting decomposition is formed by considering the residual energy and smoothness of the instantaneous amplitude. We also derive a computationally efficient implementation of the resulting estimator, using the alternating direction method of multipliers (ADMM). Numerical simulations and experimental data analysis illustrate the effectiveness and advantages of the proposed method.

In this paper, we propose a time-recursive multipitch estimation algorithm using a sparse reconstruction framework, assuming that only a few pitches from a large set of candidates are active at each time instant. The proposed algorithm does not require any training data, and instead utilizes a sparse recursive least-squares formulation augmented by an adaptive penalty term specifically designed to enforce a pitch structure on the solution. The amplitudes of the active pitches are also recursively updated, allowing for a smooth and more accurate representation. When evaluated on a set of ten music pieces, the proposed method is shown to outperform other general purpose multipitch estimators in either accuracy or computational speed, although not being able to yield performance as good as the state-of-the art methods, which are being optimally tuned and specifically trained on the present instruments. However, the method is able to outperform such a technique when used without optimal tuning, or when applied to instruments not included in the training data.

In this paper, we propose a blind channel deconvolution method based on a sparse reconstruction framework exploiting a wideband dictionary under the (relatively weak) assumption that the transmitted signal may be assumed to be well modelled as a sum of sinusoids. Using a Toeplitz structured formulation of the received signal, we form an iterative blind deconvolution scheme, alternatively estimating the underwater impulse response and the transmitted waveform. The resulting optimization problems are convex, and we formulate a computationally efficient solver using the Alternating Direction Method of Multipliers (ADMM). We illustrate the performance of the resulting estimator using both simulated and measured underwater signals.

In this work, we consider the problem of estimating the parameters of polynomially damped sinusoidal signals, commonly encountered in, for instance, spectroscopy. Generally, finding the parameter values of such signals constitutes a high-dimensional problem, often further complicated by not knowing the number of signal components or their specific signal structures. In order to alleviate the computational burden, we herein propose a mismatched estimation procedure using simplified, approximate signal models. Despite the approximation, we show that such a procedure is expected to yield predictable results, allowing for statistically and computationally efficient estimates of the signal parameters.

In this work, we investigate the possibility of incorporating a zooming procedure for the well known iterative adaptive approach (IAA), and thereby allow for higher precision on relevant areas of the spectrum. These kinds of zooming schemes have been used successfully together with several other methods, and have in many cases shown dramatical decrease in the computational cost. It has earlier been noted that the IAA method does not easily allow for these kind of zooming approaches, as the covariance formulation dictates that the resolution must be the same over all regions of the spectrum. In this paper, we present an iterative zooming procedure which allows for an efficient local estimation of IAA spectrum. The proposed algorithm is shown to allow for significant speedups without sacrificing estimation performance.

In this work, we present a computationally efficient algorithm for estimating fault modes in ball bearing systems. The presented method generalizes and improves upon earlier developed sparse reconstruction techniques, allowing for detecting multiple fault modes. The measured signal is corrupted with additive and multiplicative noise, yielding a signal that is highly erratic. Fortunately, the damaged ball bearings give rise to strong periodical structures which may be exploited when forming the proposed detector. Numerical simulations illustrate the preferred performance of the proposed method.

Estimating the parameters of non-uniformly sampled multi-dimensional damped modes is computationally cumbersome, especially if the model order of the signal is not assumed to be known a priori. In this work, we examine the possibility of using the recently introduced wideband dictionary framework to formulate a computationally efficient estimator that iteratively refines the estimates of the candidate frequency and damping coefficients for each component. The proposed wideband dictionary allows for the use of a coarse initial grid without increasing the risk of not identifying closely spaced components, resulting in a substantial reduction in computational complexity. The performance of the proposed method is illustrated using both simulated and real spectroscopy data, clearly showing the improved performance as compared to previous techniques.

In this work, we introduce an antenna placement algorithm for indoor radio networks. The algorithm aims to minimize the number of antennas required to provide sufficient coverage in an area of interest, minimizing the cost of equipment and installation work. The optimization algorithm exploits a semi-deterministic model for the most dominant radio paths. Each path is in turn determined with the A⋆ path finding algorithm. Both the proposed antenna placement algorithm and the used indoor radio propagation model are evaluated using real measurements, confirming the efficiency of the method.

We consider the over-fitting problem for multinomial probabilistic Latent Semantic Analysis (pLSA) in collaborative filtering, using a regularization approach. For big data applications, the computational complexity is at a premium and we, therefore, consider a maximum a posteriori approach based on conjugate priors that ensure that complexity of each step remains the same as compared to the un-regularized method. In the numerical section, we show that the proposed regularization method and training scheme yields an improvement on commonly used data sets, as compared to previously proposed heuristics.

In this work, we propose a method for finding an optimal, non-uniform, sampling scheme for a general class of signals in which the signal measurements may be non-linear functions of the parameters to be estimated. Formulated as a convex optimization problem reminiscent of the sensor selection problem, the method determines an optimal sampling scheme given a suitable estimation bound on the parameters of interest. The formulation also allows for putting emphasis on a particular set of parameters of interest by scaling the optimization problem in such a way that the bound to be minimized becomes more sensitive to these parameters. For the case of imprecise a priori knowledge of these parameters, we present a framework for customizing the sampling scheme to take such uncertainty into account. Numerical examples illustrate the efficiency of the proposed scheme.

This work proposes a frequency and amplitude estimator tailored for noise corrupted signals that have been clipped. Formulated as a sparse reconstruction problem, the proposed algorithm estimates the signal parameters by solving an atomic norm minimization problem. The estimator also exploits the waveform information provided by the clipped samples, incorporated in the form of linear constraints that have been augmented by slack variables as to provide robustness to noise. Numerical examples indicate that the algorithm offers preferable performance as compared to methods not exploiting the saturated samples.