Search

The RADxSM Tech initiative required a massive mobilization of the biomedical community. It was chartered with the extremely ambitious goal of rapidly developing and deploying innovative tests to detect people infected with the SARS-CoV-2 virus. It needed to do so at a scale and with urgency to get the country back to daily activities such as school and work as soon as possible. It required forming and supporting a diversity of teams with members from around the country and beyond. These teams collaborated in complex workflows that needed to be carefully monitored and tracked. This paper describes the key elements of the secure, web-based infrastructure that was configured to enable the efficient and effective operation of RADx Tech's key processes and address its unique and urgent challenges. One such challenge was to manage the flow of applications through a multi-stage, interactive selection process (using the CoLab platform) and another was to support and facilitate the progress of projects selected for support and funding through an accelerated commercialization program (using the GAITS platform).

Stimulus-driven brain-computer interfaces (BCIs), such as the P300 speller, rely on using sensory stimuli to elicit specific neural signal components called event-related potentials (ERPs) to control external devices. However, psychophysical factors, such as refractory effects and adjacency distractions, may negatively impact ERP elicitation and BCI performance. Although conventional BCI stimulus presentation paradigms usually design stimulus presentation schedules in a pseudo-random manner, recent studies have shown that controlling the stimulus selection process can enhance ERP elicitation. In prior work, we developed an algorithm to adaptively select BCI stimuli using an objective criterion that maximizes the amount of information about the user's intent that can be elicited with the presented stimuli given current data conditions. Here, we enhance this adaptive BCI stimulus selection algorithm to mitigate adjacency distractions and refractory effects by modeling temporal dependencies of ERP elicitation in the objective function and imposing spatial restrictions in the stimulus search space. Results from simulations using synthetic data and human data from a BCI study show that the enhanced adaptive stimulus selection algorithm can improve spelling speeds relative to conventional BCI stimulus presentation paradigms.Clinical relevance-Increased communication rates with our enhanced adaptive stimulus selection algorithm can potentially facilitate the translation of BCIs as viable communication alternatives for individuals with severe neuromuscular limitations.

MRI scanners with higher B0 fields typically provide a stronger Signal to Noise Ratio (SNR) which results in MRI images with their resolution or shorter acquisition time [1] . Stronger B0 fields require higher operating frequencies of the B1 field, and the shorter wavelengths cause in homogeneities in the B1 field distribution due to scattering and interference in the tissues. At higher B0 RF fields are limited to not cause excessive absorption of power in the tissues. Specifically, electric fields induce currents in conductive tissues causing heat dissipation which raise local temperature potentially leading to patient discomfort and even tissue damage. Hence, local Specific Absorprtion Rate (SAR) and temperature are limited.

High-permittivity pads have been used to enhance MRI performed in high-field static magnets (1-4). Here we perform an evaluation of fields and displacement currents associated with resonances in a thin pad or plate of high-permittivity material (HPM) in order to formulate basic design principles.

Thermal infrared (IR) imaging has increasingly been used for remote sensing applications, which has required the adaptation of processing techniques for “natural images” (e.g., RGB images) to this unique domain in order to accommodate such differences as texture and resolution. While Convolutional Neural Networks (CNNs) have recently been shown to perform well for classification of IR images, we consider the common scenario in which the target object is partially occluded. Recent work demonstrates that deep CNNs struggle to generalize under occlusion, and Compositional CNNs (CompNets) have been proposed as deep models to mitigate this shortcoming. In this work, we apply CompNets to IR imagery, discuss the considerations in moving from natural images to IR, and analyze performance on a dataset that has been artificially occluded. Our results indicate that CompNets do, indeed, bolster robustness to occlusion in the IR domain.

Speech intelligibility in cochlear implant (CI) users degrades considerably in listening environments with reverberation and noise. Previous research in automatic speech recognition (ASR) has shown that phoneme-based speech enhancement algorithms improve ASR system performance in reverberant environments as compared to a global model. However, phoneme-specific speech processing has not yet been implemented in CIs. In this paper, we propose a causal deep learning framework for classifying phonemes using features extracted at the time-frequency resolution of a CI processor. We trained and tested long short-term memory networks to classify phonemes and manner of articulation in anechoic and reverberant conditions. The results showed that CI-inspired features provide slightly higher levels of performance than traditional ASR features. To the best of our knowledge, this study is the first to provide a classification framework with the potential to categorize phonetic units in real-time in a CI.

Recently, Convolutional Neural Networks (CNNs) have demonstrated impressive performance on several visual recognition benchmark datasets utilizing overhead imagery. However, most of these analyses performed on benchmark datasets involve testing pre-trained CNNs on imagery that was collected over roughly the same locations as the training imagery. In this work we propose a benchmark problem - termed cross-geographical domain (xGD) adaptation - designed to evaluate the performance of CNNs in which they are tested on imagery collected over previously unseen geo-locations - a more challenging and practical scenario that we term cross-domain testing. We focus this work on building segmentation due to the availability of appropriate datasets. The results indicate that CNNs generalize poorly to data processed from geographic locations that were not present in training. Surprisingly, we found that larger models (pre-trained on ImageNet) generalize as well as small models in cross-domain testing, and sometimes better. This work provides the first comprehensive results for cross-domain recognition, raising awareness of this important problem. We hope that xGD can serve as a benchmark for future work; xGD uses publicly-available data, and we release our design details with this publication.

Access to electricity positively correlates with many beneficial socioeconomic outcomes in the developing world including improvements in education, health, and poverty. Efficient planning for electricity access requires information on the location of existing electric transmission and distribution infrastructure; however, the data on existing infrastructure is often unavailable or expensive. We propose a deep learning based method to automatically detect electric transmission infrastructure from aerial imagery and quantify those results with traditional object detection performance metrics. In addition, we explore two challenges to applying these techniques at scale: (1) how models trained on particular geographies generalize to other locations and (2) how the spatial resolution of imagery impacts infrastructure detection accuracy. Our approach results in object detection performance with an F1 score of 0.53 (0.47 precision and 0.60 recall). Using training data that includes more diverse geographies improves performance across the 4 geographies that we examined. Image resolution significantly impacts object detection performance and decreases precipitously as the image resolution decreases.

Segmentation convolutional neural networks (SCNNs) are now popular for the semantic segmentation (i.e., dense pixel-wise labeling) of remote sensing imagery, such as color or hyperspectral satellite imagery. One desirable property of SCNNs when applied to remote sensing problems is rotational equivariance. This property implies that the class label assigned to a particular pixel (building, road, etc.) does not change if the input imagery is rotated by an arbitrary angle. We argue that recently proposed methods to make rotational equivariant SCNNs fall into two broad categories: easily employed methods that are somewhat ineffective, and highly effective methods that are complicated and potentially incompatible with state-of-the-art SCNN techniques. We propose a simple addition to the standard SCNN loss function that encourages the SCNN to be rotationally equivariant, and is easily added to modern SCNNs. We test the method on the Inria building labeling dataset and compare it to the popular simple approach of adding random rotational augmentations of the input imagery during training. We show that the proposed approach (i) achieves improved equivariance and (ii) yields performance improvements on average.

Segmentation convolutional neural networks (CNNs) are now popular for the semantic segmentation (i.e., dense pixel-wise labeling) of remote sensing imagery, such as color or hyperspectral satellite imagery. In recent years a large number of hand-labeled datasets of overhead imagery have emerged, leading to breakthrough performance for CNNs. However, these datasets are typically used in isolation of one another because they are either (i) annotated with heterogeneous object type labels, or (ii) they are collected over different geographic areas. This imposes a major bottleneck on the value of these datasets. In this work we present what we call a class-asymmetric loss function that makes it possible to train a single multi-class network using multiple datasets that are heterogeneously-labeled. We show, for example, that it is possible to train a segmentation algorithm for Buildings, roads, and background using two datasets: one annotated with buildings and one annotated with buildings. We propose a class asymmetric loss that under certain common conditions, allows for one to train models on datasets in which the target class is unlabeled.

Deep neural networks have emerged as popular machine learning tools due to their ability to automatically learn feature representations from raw input data. An auto-encoder neural network is a special network that can be trained in an unsupervised manner for automated feature learning. Unsupervised analysis of EEG signals is highly desirable since supervised analysis requires manual labeling of EEG signals which can be labor intensive and time consuming given the large amount of EEG data collected. We present a deep convolutional auto-encoder neural network to automatically learn feature representations from raw EEG signals in an unsupervised manner. We use the features extracted from the auto-encoder neural network for clustering EEG signals into sleep stages. For clustering, we test two algorithms: K-means – which is a single-membership model, and the latent Dirichlet allocation (LDA) topic model – which is a mixed membership model. Results are presented demonstrating an improvement in clustering performance using auto-encoder features compared to standard manually extracted features.

Individuals with cochlear implants (CIs) experience more difficulty understanding speech in reverberant environ-ments than normal hearing listeners. As a result, recent research has targeted mitigating the effects of late reverberant signal reflections in CIs by using a machine learning approach to detect and delete affected segments in the CI stimulus pattern. Previous work has trained electrode-specific classification models to mitigate late reverberant signal reflections based on features extracted from only the acoustic activity within the electrode of interest. Since adjacent CI electrodes tend to be activated concurrently during speech, we hypothesized that incorporating additional information from the other electrode channels, termed cross-channel information, as features could improve classification performance. Cross-channel information extracted in real-world conditions will likely contain errors that will impact classification performance. To simulate extracting cross-channel information in realistic conditions, we developed a graphical model based on the Ising model to systematically introduce errors to specific types of cross-channel information. The Ising-like model allows us to add errors while maintaining the important geometric information contained in cross-channel information, which is due to the spectro-temporal structure of speech. Results suggest the potential utility of leveraging cross-channel information to improve the performance of the reverberation mitigation algorithm from the baseline channel-based features, even when the cross-channel information contains errors.

In this paper we evaluate through simulations the SNR improvement in the head when a high-permittivity helmet former is used with a 28 channel receive array coil. The impact on the SNR of parallel imaging applications is also considered. An overall improvement in the brain is observed when the high-permittivity helmet is used with an average SNR increase of about 51%, and a maximum improvement of 220% in a location on the side of the head. During parallel imaging applications a significant improvement can be still observed in presence of the high-permittivity helmet, but some SNR reduction is observed in the head when high acceleration factors are performed.

In this paper we evaluate an 8-channel electric dipole array placed around a high permittivity material helmet former in simulation. Transmit efficiency and coil coupling are evaluated, with the intent to build an 8-channel transmit array around a tight fitting dense receive array and a dielectric helmet for head imaging at 7 Tesla.

Recently, convolutional neural networks (CNNs) have received substantial attention in the literature for object recognition (e.g., buildings and roads) in several remote sensing data modalities (e.g., aerial color imagery). Although CNNs have exhibited excellent recognition performance, recent research suggests that trained CNNs can often perform very poorly when applied to data collected over new geographic regions, and for which little labeled training data is available. In this work, we consider the adversarial discriminative domain adaptation (ADDA) approach to address this limitation, due its recent success on related problems. A limitation of ADDA is that it is unsupervised, so in this work we extend ADDA to a semi-supervised algorithm, in which we assume that both labeled and unlabeled data are available in the new domain (e.g., in new geographic region to be evaluated). We compare semi-supervised ADDA to ADDA and a standard fine-tuning approach wherein available labeled data is used for standard CNN training. We perform experiments on two remote sensing datasets and the results indicate that semi-supervised ADDA consistently improves over the other approaches when small amounts of labeled training data are available in the new domain.

In this work we consider the application of convolutional neural networks (CNNs) for the semantic segmentation of remote sensing imagery (e.g., aerial color or hyperspectral imagery). In segmentation the goal is to provide a dense pixel-wise labeling of the input imagery. However, remote sensing imagery is usually stored in the form of very large images, called “tiles”, which are too large to be segmented directly using most CNNs and their associated hardware. During label inference (i.e., obtaining labels for a new large tile) smaller sub-images, called “patches”, are extracted uniformly over a tile and the resulting label maps are “stitched” (or concatenated) to create a tile-sized label map. This approach suffers from computational inefficiency and risks of discontinuities at the boundaries between the output of individual patches. In this work we propose a simple alternative approach in which the input size of the CNN is dramatically increased only during label inference. We evaluate the performance of the proposed approach against a standard stitching approach using two popular segmentation CNN models on the INRIA building labeling dataset. The results suggest that the proposed approach substantially reduces label inference time, while also yielding modest overall label accuracy increases. This approach also contributed to our winning entry (overall performance) in the INRIA building labeling competition.

Latent Dirichlet allocation (LDA) is a statistical model that is often used to discover topics or themes in a large collection of documents. In the LDA model, topics are modeled as discrete distributions over a finite vocabulary of words. The LDA is also a popular choice to model other datasets spanning a discrete domain, such as population genetics and social networks. However, in order to model data spanning a continuous domain with the LDA, discrete approximations of the data need to be made. These discrete approximations to continuous data can lead to loss of information and may not represent the true structure of the underlying data. We present an augmented version of the LDA topic model, where topics are represented using Gaussian mixture models (GMMs), which are multi-modal distributions spanning a continuous domain. This augmentation of the LDA topic model with Gaussian mixture topics is denoted by the GMM-LDA model. We use Gibbs sampling to infer model parameters. We demonstrate the utility of the GMM-LDA model by applying it to the problem of clustering sleep states in electroencephalography (EEG) data. Results are presented demonstrating superior clustering performance with our GMM-LDA algorithm compared to the standard LDA and other clustering algorithms.

In January 2017, Psyche and a second mission concept were selected by NASA for flight as part of the 14th Discovery mission competition. Assigned for an initial launch date in 2023, the Psyche team was given direction shortly after selection to research the possibility for earlier opportunities. Ultimately, the team was able to identify a launch opportunity in 2022 with a reduced flight time to its destination. This was accomplished in large part to crosscutting trades centered on the electrical power subsystem. These trades were facilitated through the Psyche mission's planned use of Solar Electric Propulsion (SEP), which enables substantial flexibility with respect to trajectory design. In combination with low-thrust trajectory analysis tools, the team was able to robustly converge to solutions with a higher fidelity and accuracy of results. These trades also took advantage of the 1300 series product line produced by Space Systems Loral (SSL), which enabled power growth while maintaining strong system-level heritage through its modular design that has been utilized on a large number of geostationary (GEO) communications satellites. This paper presents an overview of the Psyche mission concept, and the unique architecture that enables the use of commercially developed electric propulsion and space power systems from Space Systems Loral to provide flexibility in mission design. This paper then discusses the trades that allowed the Psyche team to meet a 2022 launch date.

Immersive virtual reality (VR) systems offer flexible control of an interactive environment, along with precise position and orientation tracking of realistic movements. Immersive VR can also be used in conjunction with neurophysiological monitoring techniques, such as electroencephalography (EEG), to record neural activity as users perform complex tasks. As such, the fusion of VR, kinematic tracking, and EEG offers a powerful testbed for naturalistic neuroscience research. In this study, we combine these elements to investigate the cognitive and neural mechanisms that underlie motor skill learning during a multi-day simulated marksmanship training regimen conducted with 20 participants. On each of 3 days, participants performed 8 blocks of 60 trials in which a simulated clay pigeon was launched from behind a trap house. Participants attempted to shoot the moving target with a firearm game controller, receiving immediate positional feedback and running scores after each shot. Over the course of the 3 days that individuals practiced this protocol, shot accuracy and precision improved significantly while reaction times got significantly faster. Furthermore, results demonstrate that more negative EEG amplitudes produced over the visual cortices correlate with better shooting performance measured by accuracy, reaction times, and response times, indicating that early visual system plasticity underlies behavioral learning in this task. These findings point towards a naturalistic neuroscience approach that can be used to identify neural markers of marksmanship performance.

Non-intrusive load monitoring (NILM) systems estimate the amount of energy each appliance consumes using as input the aggregate building energy consumption. Typically, NILM results are presented for a single sampling rate. To evaluate tradeoffs between end-uses and sensor costs, it is important to study the performance of NILM systems across sampling rates. In this work, we examine the performance of two NILM systems over a range of low frequency sampling rates on two datasets. Our results empirically demonstrate how NILM classification performance degrades nonlinearly as the sampling rate decreases and how varied this degradation is across appliance types. The results also suggest that reporting algorithm accuracy for a single sampling rate may not be sufficient for thorough algorithm performance evaluation. These findings can assist policy and decision makers in identifying the right smart meter hardware to meet appliance-level energy efficiency and building automation goals.

Convolutional neural networks (CNNs) have recently achieved unprecedented performance for the automatic recognition of objects (e.g., buildings, roads, or vehicles) in color aerial imagery. Although these results are promising, questions remain about their practical applicability. This is because there is a wide variability in the visual characteristics of remote sensing imagery across different geographic locations, and CNNs are often trained and tested on imagery from nearby (or the same) geographic locations. It is therefore unclear whether trained CNNs will perform well on new, previously unseen, geographic locations, which is an important practical consideration. In this work we investigate this problem when applying CNNs for solar array detection on a large aerial imagery dataset comprised of two nearby US cities. We compare the performance of CNNs under two conditions: training and testing on the same city vs training on one city and testing on another city. We discuss several subtle difficulties with these experiments and make recommendations. We show that there can be substantial performance loss in second case, when compared to the first. We also investigate how much training data is required from the unseen city in order to fine-tune the CNN so that it performs well. We investigate several different fine-tuning strategies, yielding a clear winner.

We consider the problem of detecting objects (such as trees, rooftops, roads, or cars) in remote sensing data including, for example, color or hyperspectral imagery. Many detection algorithms applied to this problem operate by assigning a decision statistic to all, or a subset, of spatial locations in the imagery for classification purposes. In this work we investigate a recently proposed method, called Local Averaging for Improved Predictions (LAIP), which can be used for trading off the classification accuracy of detector decision statistics with their spatial precision. We explore the behaviors of LAIP on controlled synthetic data, as we vary several experimental conditions: (a) the difficulty of the detection problem, (b) the spatial area over which LAIP is applied, and (c) how it behaves when the estimated ROC curve of the detector becomes increasingly inaccurate. These results provide basic insights about the conditions under which LAIP is effective.

Brain-computer interfaces (BCIs) can provide an alternative means of communication for individuals with severe neuromuscular limitations. The P300-based BCI speller relies on eliciting and detecting transient event-related potentials (ERPs) in electroencephalography (EEG) data, in response to a user attending to rarely occurring target stimuli amongst a series of non-target stimuli. However, in most P300 speller implementations, the stimuli to be presented are randomly selected from a limited set of options and stimulus selection and presentation are not optimized based on previous user data. In this work, we propose a data-driven method for stimulus selection based on the expected discrimination gain metric. The data-driven approach selects stimuli based on previously observed stimulus responses, with the aim of choosing a set of stimuli that will provide the most information about the user's intended target character. Our approach incorporates knowledge of physiological and system constraints imposed due to real-time BCI implementation. Simulations were performed to compare our stimulus selection approach to the row-column paradigm, the conventional stimulus selection method for P300 spellers. Results from the simulations demonstrated that our adaptive stimulus selection approach has the potential to significantly improve performance from the conventional method: up to 34% improvement in accuracy and 43% reduction in the mean number of stimulus presentations required to spell a character in a 72-character grid. In addition, our greedy approach to stimulus selection provides the flexibility to accommodate design constraints., This paper has been accepted for the 2017 IEEE International Conference on Systems, Man and Cybernetics (SMC)

As we rely increasingly on numerical simulations to evaluate safety of RF fields in MRI, there lies a danger in treating the simulations like a ”black box” and trusting results that may not be correct. An intuitive understanding of what to expect in general can help avoid such a situation. Here we examine what to expect in terms of Specific energy Absorption Rate (SAR) distributions for different types of surface coils and for volume coils both in the quasi-static regime and when wavelength effects become significant.

In this paper we evaluate the safety (maximum local SAR and temperature) of three different algorithms to drive the currents of a transmit array: 1) the birdcage coil algorithm; 2) the phase-only optimization algorithm; 3) the phase and amplitude optimization algorithm. The last two algorithms are evaluated when optimizing the transmit efficiency of RF fields in two different locations, one close to the center, and one close to the extremities of a human body subject. All the fields were normalized so that the whole-body average SAR is equal to 2W/kg, and temperature simulations were computed for a continuous exposure 1 hour long.

In this work we present a systematic evaluation of the potential of combining dielectric materials with an array of electric dipoles for MRI. Design parameters include the permittivity and length of a dielectric sleeve, as well as the dipole length and position of the tuning inductors. Results show that the combined approach improves transmit efficiency and SNR by ∼10 to 15 % compared to an optimized dipole array without dielectric sleeve. The resulting induced current densities reflect an improved correspondence with the ideal current pattern, which explains the gains in performance.

A 29-channel receive-only dense dipole head array for 7T MRI was built, methods for isolating the coil elements from each other and detuning each coil element during transmit was explored, and its performance was evaluated and compared with a commercial 24ch head array (Nova Medical. Wilmington, MA).

The probability of an intervention being effective is likely increased if it is designed following a behavioural analysis and with the aid of evidence-based intervention frameworks. For example, the Behaviour Change Wheel (BCW) framework and its associated COM-B model of behaviour have been used successfully as a starting point for designing behaviour change interventions. However, the BCW framework can also be applied at a later stage in the design process, such as when an intervention has been designed but before it is deployed. Here we describe the application of the BCW framework and COM-B model to evaluate and refine already designed interventions. We use a multidisciplinary multi-site project (‘City4Age’) as a case study. The project aims to promote independent living of older adults through the deployment of interventions using wearable and environment-based technology. We conducted face-to-face interviews with site representatives to identify the target behaviours, perceived barriers and facilitators, intervention functions, and modes of delivery for each planned intervention. Additionally, literature reviews were conducted to identify evidence-based facilitators and barriers for each targeted behaviour. Subsequently, we 1) compared the intervention functions proposed by the project-sites with those most likely to be effective according to the BCW; and 2) assessed the congruency of the barriers and facilitators identified by the pilot-sites with those identified in the literature. For five planned interventions across two project-sites (Birmingham and Singapore), two had intervention functions unlikely to be effective according to the BCW. The two planned interventions to promote social engagement did not address barriers or facilitators evident in the literature, indicating they required refinement. Applying the BCW framework allowed to identify which interventions needed refining. It also helped in providing specific guidance in our recommendations for improvements prior to deployment.

In this work, the problem of developing algorithms that automatically infer information about small-scale solar photovoltaic (PV) arrays in high resolution aerial imagery is considered. Such algorithms potentially offer a faster and cheaper solution to collecting small-scale PV information, such as their location and capacity. Existing work on this topic has focused on the automatic identification and annotation of panels in the aerial imagery. We extend this work by showing that we can reliably infer the capacity of PV arrays given only (i) color aerial imagery and (ii) a precise annotation of the array location. First we demonstrate that accurate capacity estimates can be obtained simply by estimating the visible surface area of a solar array, regardless of tilt. We then build a more sophisticated model where we use additional image information related to the properties of the solar array to further improve the capacity predictions. We use a dataset of 362 manually annotated Google Earth images of solar arrays with known electricity generation capacity in North Carolina to measure the predictive performance of our models.

In this work we consider the problem of developing algorithms that automatically identify small-scale solar photovoltaic arrays in high resolution aerial imagery. Such algorithms potentially offer a faster and cheaper solution to collecting small-scale photovoltaic (PV) information, such as their location, capacity, and the energy they produce. Here we build on previous algorithmic work by employing convolutional neural networks (CNNs), which have recently yielded major improvements in other image object recognition problems. We propose a CNN architecture for our recognition problem and then measure its detection performance on the same (publicly available) dataset that was used in previous publications. The results indicate that the CNN yields substantial performance improvements over previous results. We also investigate the recently popular approach of pre-training for CNNs.

In this work, we consider the problem of detecting target objects in remote sensing imagery; such as detecting rooftops, trees, or cars in color/hyperspectral imagery. Many detection algorithms for this problem work by assigning a decision statistic (or “confidence”) to all, or a subset, of spatial locations in the data. A threshold is then applied to the statistics to identify detections. The detection theory underpinning this general approach assumes that a yes/no decision must be made, individually, for each location. In some applications, however, the precise location of the detected objects may be less important than knowing how many total objects there are. In this work we propose two methods that can permit a generic detection algorithm to gradually lower the spatial certainty, or resolution, of its detections in order to improve the accuracy of the overall number of detected objects. We validate the proposed methods on a controlled synthetic dataset as well as a real dataset from previously published work on solar photovoltaic array detection in color aerial imagery.

A large body of recent research has focused on the development of supervised buried threat detection algorithms for ground penetrating radar (GPR) data. Such algorithms learn to automatically identify landmines in GPR data based on threat data and non-threat data examples. Training data typically consists of small 2-dimensional images that are extracted from a larger image, or volume, of GPR data. Currently, the most popular criterion for choosing training (or testing) patches is high GPR signal energy. In this work, we investigate translational variance in the patches, which occurs when relevant GPR signals (e.g., hyperbolic landmine signals) are not consistently centered, or aligned, in the extracted patches. In this work, we (i) provide evidence suggesting that translational variance is introduced into the data when popular energy based patch extraction methods are employed, and (ii) estimate the classification performance loss in supervised algorithms due to this effect. We present a simple method to help alleviate the translational variance problem. We hypothesize that reducing translational variance prior to supervised learning may facilitate the use, and success, of image features.

Forward-looking ground-penetrating radar (FLGPR) has been investigated as a remote sensing modality for buried threat detection (e.g., landmines). Generally in this context, raw FLGPR data is beamformed into two-dimensional images and then automated algorithms are applied to detect buried threats. The beamformed images represent a crude estimate of the energy reflected from the subsurface (i) across a particular spatial extent, and (ii) at a particular depth. Typically, a single focus depth (e.g., −0.25 meters) is used to beamform all of the imagery. In this work we hypothesize that this approach disregards important factors such as the likely variable burial depth of threats, and varying contours of the ground. We examine this hypothesis by beamforming multiple images, each at a separate focus depth, and stack them to form 3-dimensional (3D) volumes. Several popular FLGPR detection algorithms are applied to both the 3D volumes and the conventional 2D images on a large collection of FLGPR data. The results show that the 3D detectors consistently outperform their 2D counterparts. This suggests that the use of a single beamforming focus depth for imaging is a potential performance bottleneck for threat detection in FLGPR data.

Ground Penetrating Radar (GPR) is a remote sensing modality that has been researched extensively for buried threat detection. For this purpose, algorithms can be developed to automatically determine the presence of such threats. To train such algorithms, small 2-dimensional images can be extracted from the larger image, or volume, of GPR data. One thread of research in the buried threat detection literature is to use visual descriptors from the computer vision literature. One recent, very successful approach in that field is the use of deep convolutional neural networks (CNNs). Applying CNNs requires a large number of design choices which complicate their use. In this work, we investigate their application to GPR data and adapt several recent advances from the CNN literature to improve detection performance on GPR data. In particular, we investigate the initialization step of pretraining and propose a dataset augmentation protocol. The efficacy of these approaches are evaluated on several architectures with a relatively similar number of network parameters to learn. The results indicate that both pretraining and dataset augmentation help achieve higher detection performance.

In September 2015, NASA selected five mission concepts from a field of 27 to proceed to the next stage (step 2) of the latest Discovery mission competition. Each team submitted a Mission Concept Study to NASA in August 2016, and in January of 2017 NASA selected Psyche and a second mission for flight. This paper describes Psyche, a unique investigation of a metal world, which is the only one of the original five mission concepts studied in detail to propose the use of Electric Propulsion (EP) to accomplish its mission objectives. Psyche will harness commercially developed EP and space power systems with strong system-level heritage to accomplish a deep space NASA science mission at comparatively low technical-risk and cost-risk. This paper describes the Psyche mission concept and the unique Solar Electric Propulsion (SEP) architecture that allows the use of SSL's commercial SPT-140 Hall thruster propulsion system at solar distances of up to 3.3 AU with only minimal modifications. Building on previous work analyzing SEP systems for Discovery-class missions, this paper describes the heritage, design, and testing which have been conducted on the power and propulsion systems to develop the Psyche mission, addresses the differences between GEO and deep-space environments, and describes actions taken to ensure that GEO heritage systems can be operated reliably in deep-space.

The P300-based brain-computer interface (BCI) speller relies on eliciting and detecting specific brain responses to target stimulus events, termed event-related potentials (ERPs). In a visual speller, ERPs are elicited when the user's desired character, i.e. the “target,” is flashed on a computer screen. The P300 speller is currently limited by its relatively slow typing speed due to the need for repetitive data measurements that are necessary to achieve reasonable signal-to-noise ratios. In addition, refractory effects limit the ability to elicit ERPs with every target stimulus event presentation. In this paper, we present a new method to design the stimulus presentation paradigm for the P300 speller by exploiting an information-theoretic approach to maximize the information content that is presented to the user while also mitigating refractory effects. We present results with real-time BCI use which demonstrate significant performance improvements with our performance-based paradigm compared to the conventional stimulus presentation paradigm.

Power generation from distributed solar photovoltaic (PV) arrays has grown rapidly in recent years. As a result, there is interest in collecting information about the quantity, power capacity, and energy generated by such arrays; and to do so over small geo-spatial regions (e.g., counties, cities, or even smaller regions). Unfortunately, existing sources of such information are dispersed, limited in geospatial resolution, and otherwise incomplete or publically unavailable. As result, we recently proposed a new approach for collecting such distributed PV information that relies on computer algorithms to automatically detect PV arrays in high resolution aerial imagery [1], Here, we build on this work by investigating a detection algorithm based on a Random Forest (RF) classifier, and we consider its detection performance using several different sets of image features. The proposed method is developed and tested using a very large collection of publicly available [2] aerial imagery, covering 112.5 km2 of surface area, with 2,328 manually annotated PV array locations. The results indicate that a combination of local color and texture (using the popular texton feature) features yield the best detection performance.

Most state-of-the-art dictionary learning algorithms (DLAs) are iterative, and must begin with an initial estimate of the dictionary, referred to as the seed. A seed can be generated randomly, but it has been shown that choosing a more intelligent seed often yields a better solution. For example, a seed inferred using data from a related problem, or one handcrafted based on a priori knowledge of the problem at hand can yield better solutions. Seed dictionaries appear to encode valuable a priori information however, most DLAs discard the seed after initialization. This work investigates the questions of whether the information encoded in a good seed can be leveraged further, by potentially using the seed to influence learning after initialization. This is achieved by modifying the popular DLA K-SVD to use the seed as a prior during learning, by penalizing differences between the learned dictionary and the seed. The resulting algorithm, referred to as Seed Shrinkage Dictionary Learning (SSDL), is examined against K-SVD on image denoising experiments using several benchmark images. The results indicate that utilizing the seed as a prior in this way consistently yields improved denoising performance in our experiments. This simple approach motivates the development of more sophisticated approaches that leverage a priori information in useful seeds.

Academic investigators are generating a plethora of insights and technologies that have the potential to significantly improve patient care. However, to address the imperative to improve the quality, cost and access to care with ever more constrained funding, the efficiency and the consistency with which they are translated into cost effective products and/or services need to improve. Healthcare commercialization programs (HCPs) are described and proposed as an option that institutions can add to their portfolio to improve translational research. In helping teams translate specific healthcare innovations into practice, HCPs expand the skillset of investigators and enhance an institution’s innovation capacity. Lessons learned are shared from configuring and delivering HCPs, which build on the fundamentals of the National Science Foundation’s Innovation Corps program, to address the unique challenges in supporting healthcare innovations and innovators., Healthcare Commercialization Programs (HCPs): Filling a gap in an institution's portfolio of ways to improve the efficiency of translating healthcare innovations from academia into practice.

To advance the development of point-of-care technology (POCT), the National Institute of Biomedical Imaging and Bioengineering established the POCT Research Network (POCTRN), comprised of Centers that emphasize multidisciplinary partnerships and close facilitation to move technologies from an early stage of development into clinical testing and patient use. This paper describes the POCTRN and the three currently funded Centers as examples of academic-based organizations that support collaborations across disciplines, institutions, and geographic regions to successfully drive innovative solutions from concept to patient care., This paper describes the National Institute of Biomedical Imaging and Bioengineering's Point-of-Care Technologies Research Network, three Centers that, through multidisciplinary partnerships and close facilitation, are working to move point-of-care technologies from an early stage of development into clinical testing and patient use.

Energy disaggregation algorithms decompose building-level energy data into device-level information. We conduct a head-to-head comparison of energy disaggregation techniques across multiple metrics and data sets. Our framework for analyzing the performance of a complete energy disaggregation system includes event detection, classification, and power assignment. We use receiver operating characteristics (ROCs) to evaluate event detection performance, and we introduce a technique to evaluate device-level event detection. We use confusion matrices to compare classification performance across several classifiers, and evaluate the resulting power assignments using several assignment metrics that are commonly used in the literature to demonstrate the varying strengths of the techniques that were considered. We apply this framework to several publicly available datasets and demonstrate how system performance varies with sampling frequency and the inclusion of reactive power. Our results suggest that (1) disaggregation performance varies considerably across data sets (2) increased data sampling rate improves disaggregation performance, and (3) additional features such as reactive power yields disaggregation performance improvements.

The quantity of rooftop solar photovoltaic (PV) installations has grown rapidly in the US in recent years. There is a strong interest among decision makers in obtaining high quality information about rooftop PV, such as the locations, power capacity, and energy production of existing rooftop PV installations. Solar PV installations are typically connected directly to local power distribution grids, and therefore it is important for the reliable integration of solar energy to have information at high geospatial resolutions: by county, zip code, or even by neighborhood. Unfortunately, traditional means of obtaining this information, such as surveys and utility interconnection filings, are limited in availability and geospatial resolution. In this work a new approach is investigated where a computer vision algorithm is used to detect rooftop PV installations in high resolution color satellite imagery and aerial photography. It may then be possible to use the identified PV images to estimate power capacity and energy production for each array of panels, yielding a fast, scalable, and inexpensive method to obtain rooftop PV estimates for regions of any size. The aim of this work is to investigate the feasibility of the first step of the proposed approach: detecting rooftop PV in satellite imagery. Towards this goal, a collection of satellite rooftop images is used to develop and evaluate a detection algorithm. The results show excellent detection performance on the testing dataset and that, with further development, the proposed approach may be an effective solution for fast and scalable rooftop PV information collection.

A regionalization system delineates the geographical landscape into spatially contiguous, homogeneous units for landscape ecology research and applications. In this study, we investigated a quantitative approach for developing a regional-ization system using constrained clustering algorithms. Unlike conventional clustering, constrained clustering uses domain constraints to help guide the clustering process towards finding a desirable solution. For region delineation, the adjacency relationship between neighboring spatial units can be provided as constraints to ensure that the resulting regions are geographically connected. However, using a large-scale terrestrial ecology data set as our case study, we showed that incorporating such constraints into existing constrained clustering algorithms is not that straightforward. First, the algorithms must carefully balance the trade-off between spatial contiguity and landscape homogeneity of the regions. Second, the effectiveness of the algorithms strongly depends on how the spatial constraints are represented and incorporated into the clustering framework. In this paper, we introduced a truncated exponential kernel to represent spatial contiguity constraints for region delineation using constrained spectral clustering. We also showed that a Hadamard product approach that combines the kernel with landscape feature similarity matrix can produce regions that are more spatially contiguous compared to other baseline algorithms.

A sheet of high dielectric constant surrounding the head has been used in MRI scans to improve signal-to-noise ratio and lower the power transmitted by the array of coils. In a study presented at the URSI General Assembly and Scientific Symposium, Beijing, China, August 2014, we have presented a two-dimensional model in which the human head is simulated by a cylinder with appropriate uniform dielectric constant and conductivity, surrounded by a dielectric sheet of high permittivity and illuminated by an array of line sources. We proved that under appropriate excitation of the array elements, the electric field is zero and the magnetic field is locally maximum and circularly polarized on the axis of the structure.

We investigated how high permittivity materials affect the S-parameters and transmit efficiency of a head-sized birdcage coil. We also studied whether re-tuning and matching are necessary for improved performance. While adjusting the S-parameters is possible during coil development and for prototype coils, it is often unfeasible to re-tune and match commercially manufactured coils, and poses a challenge after adding dielectric material to such coils. Our work demonstrates the change in the S-parameters and the transmit efficiency after placing a dielectric sleeve within a head-sized birdcage coil, and reports that coil performance can be improved after re-tuning and matching the coil in the presence of the dielectric material.

We present an analytical approach to justify the behavior of the magnetic field observed when dielectric material is used to enhance magnetic resonance imaging (MRI) of the human head.

We present a novel tracking algorithm for an autonomous vehicle equipped with a single camera. Given only monocular visual data, our algorithm utilizes projective geometry to compute concise features of the environment. Using these features, road markings are identified by a multi-class classifier. The classification results are then used with a Rao-Blackwellized particle filter to track the vehicle as it moves back and forth across the road. The resulting position tracker is part of a complete, simulated autonomous driving system. The realistic driving video game Need for Speed: Hot Pursuit was used as a vehicle simulation platform, and the autonomous system is shown to perform competitively against the game's automated opponents.

Cigarette smoking remains a major public health issue. Despite a variety of treatment options, existing intervention protocols intended to support attempts to quit smoking have low success rates. An emerging treatment framework, referred to as adaptive interventions in behavioral health, addresses the chronic, relapsing nature of behavioral health disorders by tailoring the composition and dosage of intervention components to an individual’s changing needs over time. An important component of a rapid and effective adaptive smoking intervention is an understanding of the behavior change relationships that govern smoking behavior and an understanding of intervention components’ dynamic effects on these behavioral relationships. As traditional behavior models are static in nature, they cannot act as an effective basis for adaptive intervention design. In this article, behavioral data collected daily in a smoking cessation clinical trial is used in development of a dynamical systems model that describes smoking behavior change during cessation as a self-regulatory process. Drawing from control engineering principles, empirical models of smoking behavior are constructed to reflect this behavioral mechanism and help elucidate the case for a control-oriented approach to smoking intervention design.

We present Haptics Assisted Training (HAT) System, a force feedback workstation for transferring and improving handwriting skill. The HAT system is the first practical handwriting training system whose usability was tested both in typical and special education classrooms of a local school district. Simulating the role of occupational therapists (OT), the HAT system guides the user's hand along the sequence of strokes of the reference handwriting recorded by an expert or teacher. Since the handwriting data is formatted as a set of ordered paths, it can teach any form of pen strokes, such as words, shapes, and even drawings. Training tasks are presented as a 3D game which captures children's attention and more effective visual motor integration (VMI) is anticipated than traditional paper materials. In this paper we describe the design and implementation of the prototype. We also report a comparative study on handwriting performance of twenty participants grouped into four categories by education types (special vs. typical) and haptic devices (Phantom Omni™ vs. Novint Falcon™). As a result, we observed that the children quickly adjusted themselves to the system, and they sustained engagement at the word-writing tasks without any intervention. Results also showed improvements in tracing precision on the HAT system over time. Details of the results and lessons learned from the study are discussed.