3D Convolutional Deep Learning for Nonlinear Estimation of Body Composition from Whole-Body Morphology

Total and regional body composition are strongly correlated with metabolic syndrome and have been estimated non-invasively from 3D optical scans using linear parameterizations of body shape and linear regression models. Prior works produced accurate and precise predictions on many, but not all, body composition targets relative to the reference dual X-Ray absorptiometry (DXA) measurement. Here, we report the effects of replacing linear models with nonlinear parameterization and regression models on the precision and accuracy of body composition estimation in a novel application of deep 3D convolutional graph networks to human body composition modeling. We assembled an ensemble dataset of 4286 topologically standardized 3D optical scans from four different human body shape databases, DFAUST, CAESAR, Shape Up! Adults, and Shape Up! Kids and trained a parameterized shape model using a graph convolutional 3D autoencoder (3DAE) in lieu of linear PCA. We trained a nonlinear Gaussian process regression (GPR) on the 3DAE parameter space to predict body composition via correlations to paired DXA reference measurements from the Shape Up! scan subset. We tested our model on a set of 424 randomly withheld test meshes and compared the effects of nonlinear computation against prior linear models. Nonlinear GPR produced up to 20% reduction in prediction error and up to 30% increase in precision over linear regression for both sexes in 10 tested body composition variables. Deep shape features produced 6-8% reduction in prediction error over linear PCA features for males only and a 4-14% reduction in precision error for both sexes. Our best performing nonlinear model predicting body composition from deep features outperformed prior work using linear methods on all tested body composition prediction metrics in both precision and accuracy. All coefficients of determination (R2) for all predicted variables were above 0.86. We show that GPR is a more precise and accurate method for modeling body composition mappings from body shape features than linear regression. Deep 3D features learned by a graph convolutional autoencoder only improved male body composition accuracy but improved precision in both sexes. Our work achieved lower estimation RMSEs than all previous work on 10 metrics of body composition.

A DYNAMIC SPATIAL FILTERING APPROACH TO MITIGATE UNDERESTIMATION BIAS IN FIELD CALIBRATED LOW-COST SENSOR AIR POLLUTION DATA

Low-cost air pollution sensors, offering hyper-local characterization of pollutant concentrations, are becoming increasingly prevalent in environmental and public health research. However, low-cost air pollution data can be noisy, biased by environmental conditions, and usually need to be field-calibrated by collocating low-cost sensors with reference-grade instruments. We show, theoretically and empirically, that the common procedure of regression-based calibration using collocated data systematically underestimates high air pollution concentrations, which are critical to diagnose from a health perspective. Current calibration practices also often fail to utilize the spatial correlation in pollutant concentrations. We propose a novel spatial filtering approach to collocation-based calibration of low-cost networks that mitigates the underestimation issue by using an inverse regression. The inverse-regression also allows for incorporating spatial correlations by a second-stage model for the true pollutant concentrations using a conditional Gaussian Process. Our approach works with one or more collocated sites in the network and is dynamic, leveraging spatial correlation with the latest available reference data. Through extensive simulations, we demonstrate how the spatial filtering substantially improves estimation of pollutant concentrations, and measures peak concentrations with greater accuracy. We apply the methodology for calibration of a low-cost PM2.5 network in Baltimore, Maryland, and diagnose air pollution peaks that are missed by the regression-calibration.

Sparse Depth-Guided Image Enhancement Using Incremental GP with Informative Point Selection

We propose an online dehazing method with sparse depth priors using an incremental Gaussian Process (iGP). Conventional approaches focus on achieving single image dehazing by using multiple channels. In many robotics platforms, range measurements are directly available, except in a sparse form. This paper exploits direct and possibly sparse depth data in order to achieve efficient and effective dehazing that works for both color and grayscale images. The proposed algorithm is not limited to the channel information and works equally well for both color and gray images. However, efficient depth map estimations (from sparse depth priors) are additionally required. This paper focuses on a highly sparse depth prior for online dehazing. For efficient dehazing, we adopted iGP for incremental depth map estimation and dehazing. Incremental selection of the depth prior was conducted in an information-theoretic way by evaluating mutual information (MI) and other information-based metrics. As per updates, only the most informative depth prior was added, and haze-free images were reconstructed from the atmospheric scattering model with incrementally estimated depth. The proposed method was validated using different scenarios, color images under synthetic fog, real color, and grayscale haze indoors, outdoors, and underwater scenes.

Bayesian Inferences on Neural Activity in EEG-Based Brain-Computer Interface

A brain-computer interface (BCI) is a system that translates brain activity into commands to operate technology. A common design for an electroencephalogram (EEG) BCI relies on the classification of the P300 event-related potential (ERP), which is a response elicited by the rare occurrence of target stimuli among common non-target stimuli. Few existing ERP classifiers directly explore the underlying mechanism of the neural activity. To this end, we perform a novel Bayesian analysis of the probability distribution of multi-channel real EEG signals under the P300 ERP-BCI design. We aim to identify relevant spatial temporal differences of the neural activity, which provides statistical evidence of P300 ERP responses and helps design individually efficient and accurate BCIs. As one key finding of our single participant analysis, there is a 90% posterior probability that the target ERPs of the channels around visual cortex reach their negative peaks around 200 milliseconds post-stimulus. Our analysis identifies five important channels (PO7, PO8, Oz, P4, Cz) for the BCI speller leading to a 100% prediction accuracy. From the analyses of nine other participants, we consistently select the identified five channels, and the selection frequencies are robust to small variations of bandpass filters and kernel hyper-parameters.

3D Geometric Shape Reconstruction for Revision TKA and UKA Knees Using Gaussian Process Regression

Revision knee surgery is complicated by distortion of previous components and removal of additional bone, potentially causing misalignment and inappropriate selection of implants. In this study, we reconstructed the native femoral and tibial surface shapes in simulated total/unicompartmental knee arthroplasty (TKA/UKA) for 20 femurs and 20 tibias using a statistical inference method based on Gaussian Process regression. Compared to the true geometry, the average absolute errors (mean absolute distances) in the prediction of resected femur bones in TKA, medial UKA, and lateral UKA were 1.0 ± 0.3 mm, 1.0 ± 0.3 mm, and 0.8 ± 0.2 mm, respectively, while those in the prediction of tibia resections in the corresponding surgeries were 1.0 ± 0.4 mm, 0.8 ± 0.2 mm, and 0.7 ± 0.2 mm, respectively. Furthermore, it was found that the prediction accuracy depends on the size and gender of the resected bone. For example, the prediction accuracy for UKA cuts was significantly better than that for TKA cuts (p < 0.05). The female and male cuts were often overfit and underfit, respectively. The data indicated that this reconstruction approach can be a viable option for planning of revision surgeries, especially when contralateral anatomy is pathological or cannot be available.

Large Vessel Occlusion Prediction in the Emergency Department with National Institutes of Health Stroke Scale Components: A Machine Learning Approach

OBJECTIVE

To determine the feasibility of using a machine learning algorithm to screen for large vessel occlusions (LVO) in the Emergency Department (ED).

MATERIALS AND METHODS

A retrospective cohort of consecutive ED stroke alerts at a large comprehensive stroke center was analyzed. The primary outcome was diagnosis of LVO at discharge. Components of the National Institutes of Health Stroke Scale (NIHSS) were used in various clinical methods and machine learning algorithms to predict LVO, and the results were compared with the baseline method (aggregate NIHSS score with threshold of 6). The Area-Under-Curve (AUC) was used to measure the overall performance of the models. Bootstrapping (n = 1000) was applied for the statistical analysis.

RESULTS

Of 1133 total patients, 67 were diagnosed with LVO. A Gaussian Process (GP) algorithm significantly outperformed other methods including the baseline methods. AUC score for the GP algorithm was 0.874 ± 0.025, compared with the simple aggregate NIHSS score, which had an AUC score of 0.819 ± 0.024. A dual-stage GP algorithm is proposed, which offers flexible threshold settings for different patient populations, and achieved an overall sensitivity of 0.903 and specificity of 0.626, in which sensitivity of 0.99 was achieved for high-risk patients (defined as initial NIHSS score > 6).

CONCLUSION

Machine learning using a Gaussian Process algorithm outperformed a clinical cutoff using the aggregate NIHSS score for LVO diagnosis. Future studies would be beneficial in exploring prospective interventions developed using machine learning in screening for LVOs in the emergent setting.

ADAPTIVE IMPORTANCE SAMPLING VIA AUTO-REGRESSIVE GENERATIVE MODELS AND GAUSSIAN PROCESSES

The quality of importance distribution is vital to adaptive importance sampling, especially in high dimensional sampling spaces where the target distributions are sparse and hard to approximate. This requires that the proposal distributions are expressive and easily adaptable. Because of the need for weight calculation, point evaluation of the proposal distributions is also needed. The Gaussian process has been proven to be a highly expressive non-parametric model for conditional density estimation whose training process is also straightforward. In this paper, we introduce a class of adaptive importance sampling methods where the proposal distribution is constructed in a way that Gaussian processes are combined autoregressively. By numerical experiments of sampling from a high dimensional target distribution, we demonstrate that the method is accurate and efficient compared to existing methods.

Motor Cortex Mapping using Active Gaussian Processes

One important application of transcranial magnetic stimulation (TMS) is to map cortical motor topography by spatially sampling the motor cortex, and recording motor evoked potentials (MEP) with surface electromyography. Standard approaches to TMS mapping involve repetitive stimulations at different loci spaced on a (typically 1 cm) grid on the scalp. These mappings strategies are time consuming and responsive sites are typically sparse. Furthermore, the long time scale prevents measurement of transient cortical changes, and is poorly tolerated in clinical populations. An alternative approach involves using the TMS mapper expertise to exploit the map's sparsity through the use of feedback of MEPs to decide which loci to stimulate. In this investigation, we propose a novel active learning method to automatically infer optimal future stimulus loci in place of user expertise. Specifically, we propose an active Gaussian Process (GP) strategy with loci selection criteria such as entropy and mutual information (MI). The proposed method twists the usual entropy- and MI-based selection criteria by modeling the estimated MEP field, i.e., the GP mean, as a Gaussian random variable itself. By doing so, we include MEP amplitudes in the loci selection criteria which would be otherwise completely independent of the MEP values. Experimental results using real data shows that the proposed strategy can greatly outperform competing methods when the MEP variations are mostly conned in a sub-region of the space.

Optimally Balanced Gaussian Process Propensity Scores for Estimating Treatment Effects

Propensity scores are commonly employed in observational study settings where the goal is to estimate average treatment effects. This paper introduces a flexible propensity score modeling approach, where the probability of treatment is modeled through a Gaussian process framework. To evaluate the effectiveness of the estimated propensity score, a metric of covariate imbalance is developed that quantifies the discrepancy between the distributions of covariates in the treated and control groups. It is demonstrated that this metric is ultimately a function of the hyperparameters of the covariance matrix of the Gaussian process and therefore it is possible to select the hyperparameters to optimize the metric and minimize overall covariate imbalance. The effectiveness of the GP method is compared in a simulation against other methods of estimating the propensity score and the method is applied to data from Dehejia and Wahba (1999) to demonstrate benchmark performance within a relevant policy application.

A parametric interpretation of Bayesian Nonparametric Inference from Gene Genealogies: Linking ecological, population genetics and evolutionary processes

Using a nonparametric Bayesian approach Palacios and Minin (2013) dramatically improved the accuracy, precision of Bayesian inference of population size trajectories from gene genealogies. These authors proposed an extension of a Gaussian Process (GP) nonparametric inferential method for the intensity function of non-homogeneous Poisson processes. They found that not only the statistical properties of the estimators were improved with their method, but also, that key aspects of the demographic histories were recovered. The authors' work represents the first Bayesian nonparametric solution to this inferential problem because they specify a convenient prior belief without a particular functional form on the population trajectory. Their approach works so well and provides such a profound understanding of the biological process, that the question arises as to how truly "biology-free" their approach really is. Using well-known concepts of stochastic population dynamics, here I demonstrate that in fact, Palacios and Minin's GP model can be cast as a parametric population growth model with density dependence and environmental stochasticity. Making this link between population genetics and stochastic population dynamics modeling provides novel insights into eliciting biologically meaningful priors for the trajectory of the effective population size. The results presented here also bring novel understanding of GP as models for the evolution of a trait. Thus, the ecological principles foundation of Palacios and Minin (2013)'s prior adds to the conceptual and scientific value of these authors' inferential approach. I conclude this note by listing a series of insights brought about by this connection with Ecology.