Enhanced SpO2 estimation using explainable machine learning and neck photoplethysmography.

Reflectance-based photoplethysmogram (PPG) sensors provide flexible options of measuring sites for blood oxygen saturation (SpO2) measurement. But they are mostly limited by accuracy, especially when applied to different subjects, due to the diverse human characteristics (skin colors, hair density, etc.) and usage conditions of different sensor settings. This study addresses the estimation of SpO2 at non-standard measuring sites employing reflectance-based sensors. It proposes an automated construction of subject inclusion-exclusion criteria for SpO2 measuring devices, using a combination of unsupervised clustering, supervised regression, and model explanations. This is perhaps among the first adaptation of SHAP to explain the clusters gleaned from unsupervised learning methods. As a wellness application case study, we developed a pillow-based wearable device to collect reflectance PPGs from both the brachiocephalic and carotid arteries around the neck. The experiment was conducted on 33 subjects, each under totally 80 different sensor settings. The proposed approach addressed the variations of humans and devices, as well as the heterogeneous mapping between signals and SpO2 values. It identified effective device settings and characteristics of their applicable subject groups (i.e., subject inclusion-exclusion criteria). Overall, it reduced the root mean squared error (RMSE) by 16%, compared to an empirical formula and a plain SpO2 estimation model.

Toward standardizing the clinical testing protocols of point-of-care devices for obstructive sleep apnea diagnosis.

PURPOSE: In recent years, point-of-care (POC) devices, especially smart wearables, have been introduced to provide a cost-effective, comfortable, and accessible alternative to polysomnography (PSG)-the current gold standard-for the monitoring, screening, and diagnosis of obstructive sleep apnea (OSA). Thorough validation and human subject testing are essential steps in the translation of these device technologies to the market. However, every device development group tests their device in their own way. No standard guidelines exist for assessing the performance of these POC devices. The purpose of this paper is to critically distill the key aspects of the various protocols reported in the literature and present a protocol that unifies the best practices for testing wearable and other POC devices for OSA. METHODS: A limited review and graphical descriptive analytics of literature-including journal articles, web sources, and clinical manuscripts by authoritative agencies in sleep medicine-are performed to glean the testing and validation methods employed for POC devices, specifically for OSA. RESULTS: The analysis suggests that the extent of heterogeneity of the demographics, the performance metrics, subject survey, hypotheses, and statistical analyses need to be carefully considered in a systematic protocol for testing POC devices for OSA. CONCLUSION: We provide a systematic method and list specific recommendations to extensively assess various performance criteria for human subject testing of POC devices. A rating scale of 1-3 is provided to encourage studies to put a focus on addressing the key elements of a testing protocol.

Surface plastic flow in polishing of rough surfaces.

We present experimental evidence for a new mechanism for how smooth surfaces emerge during repetitive sliding contacts, as in polishing. Electron microscopy observations of Ti-6Al-4V surface with a spherical asperity structure-realized via additive manufacturing-during successive polishing stages suggest that asperity-abrasive contacts exhibit viscous behavior, where the asperity material flows in the form of thin (1-10 µm) fluid-like layers. Subsequent bridging of these layers among neighboring asperities results in progressive surface smoothening. Using analytical asperity-abrasive contact temperature modeling and microstructural characterization, we show that the sliding contacts encounter flash temperatures of the order of 700-900 K which is in the range of the dynamic recrystallization temperature of the material considered, thus supporting the experimental observations. Besides providing a new perspective on the long-held mechanism of polishing, our observations provide a novel approach based on graph theory to quantitatively characterize the evolution of surface morphology. Results suggest that the graph representation offers a more efficient measure to characterize the surface morphology emerging at various stages of polishing. The research findings and observations are of broad relevance to the understanding of plastic flow behavior of sliding contacts ubiquitous in materials processing, tribology, and natural geological processes as well as present unique opportunities to tailor the microstructures by controlling the thermomechanics of the asperity-abrasive contacts.

Correction: Nonlinear Dynamics Forecasting of Obstructive Sleep Apnea Onsets.

[This corrects the article DOI: 10.1371/journal.pone.0164406.].

Nonlinear Dynamics Forecasting of Obstructive Sleep Apnea Onsets.

Recent advances in sensor technologies and predictive analytics are fueling the growth in point-of-care (POC) therapies for obstructive sleep apnea (OSA) and other sleep disorders. The effectiveness of POC therapies can be enhanced by providing personalized and real-time prediction of OSA episode onsets. Previous attempts at OSA prediction are limited to capturing the nonlinear, nonstationary dynamics of the underlying physiological processes. This paper reports an investigation into heart rate dynamics aiming to predict in real time the onsets of OSA episode before the clinical symptoms appear. A prognosis method based on a nonparametric statistical Dirichlet-Process Mixture-Gaussian-Process (DPMG) model to estimate the transition from normal states to an anomalous (apnea) state is utilized to estimate the remaining time until the onset of an impending OSA episode. The approach was tested using three datasets including (1) 20 records from 14 OSA subjects in benchmark ECG apnea databases (Physionet.org), (2) records of 10 OSA patients from the University of Dublin OSA database and (3) records of eight subjects from previous work. Validation tests suggest that the model can be used to track the time until the onset of an OSA episode with the likelihood of correctly predicting apnea onset in 1 min to 5 mins ahead is 83.6 ± 9.3%, 80 ± 8.1%, 76.2 ± 13.3%, 66.9 ± 15.4%, and 61.1 ± 16.7%, respectively. The present prognosis approach can be integrated with wearable devices, enhancing proactive treatment of OSA and real-time wearable sensor-based of sleep disorders.

Random Forests Are Able to Identify Differences in Clotting Dynamics from Kinetic Models of Thrombin Generation.

Current methods for distinguishing acute coronary syndromes such as heart attack from stable coronary artery disease, based on the kinetics of thrombin formation, have been limited to evaluating sensitivity of well-established chemical species (e.g., thrombin) using simple quantifiers of their concentration profiles (e.g., maximum level of thrombin concentration, area under the thrombin concentration versus time curve). In order to get an improved classifier, we use a 34-protein factor clotting cascade model and convert the simulation data into a high-dimensional representation (about 19000 features) using a piecewise cubic polynomial fit. Then, we systematically find plausible assays to effectively gauge changes in acute coronary syndrome/coronary artery disease populations by introducing a statistical learning technique called Random Forests. We find that differences associated with acute coronary syndromes emerge in combinations of a handful of features. For instance, concentrations of 3 chemical species, namely, active alpha-thrombin, tissue factor-factor VIIa-factor Xa ternary complex, and intrinsic tenase complex with factor X, at specific time windows, could be used to classify acute coronary syndromes to an accuracy of about 87.2%. Such a combination could be used to efficiently assay the coagulation system.

Inferring sparse networks for noisy transient processes.

Inferring causal structures of real world complex networks from measured time series signals remains an open issue. The current approaches are inadequate to discern between direct versus indirect influences (i.e., the presence or absence of a directed arc connecting two nodes) in the presence of noise, sparse interactions, as well as nonlinear and transient dynamics of real world processes. We report a sparse regression (referred to as the l1-min) approach with theoretical bounds on the constraints on the allowable perturbation to recover the network structure that guarantees sparsity and robustness to noise. We also introduce averaging and perturbation procedures to further enhance prediction scores (i.e., reduce inference errors), and the numerical stability of l1-min approach. Extensive investigations have been conducted with multiple benchmark simulated genetic regulatory network and Michaelis-Menten dynamics, as well as real world data sets from DREAM5 challenge. These investigations suggest that our approach can significantly improve, oftentimes by 5 orders of magnitude over the methods reported previously for inferring the structure of dynamic networks, such as Bayesian network, network deconvolution, silencing and modular response analysis methods based on optimizing for sparsity, transients, noise and high dimensionality issues.

Detection of sleep apnea events via tracking nonlinear dynamic cardio-respiratory coupling from electrocardiogram signals.

Obstructive sleep apnea (OSA) is a common sleep disorder that causes increasing risk of mortality and affects quality of life of approximately 6.62% of the total US population. Timely detection of sleep apnea events is vital for the treatment of OSA. In this paper, we present a novel approach based on extracting the quantifiers of nonlinear dynamic cardio-respiratory coupling from electrocardiogram (ECG) signals to detect sleep apnea events. The quantifiers of the cardio-respiratory dynamic coupling were extracted based on recurrence quantification analysis (RQA), and a battery of statistical data mining techniques were to enhance OSA detection accuracy. This approach would lead to a cost-effective and convenient means for screening of OSA, compared to traditional polysomnography (PSG) methods. The results of tests conducted using data from PhysioNets Sleep Apnea database suggest excellent quality of the OSA detection based on a thorough comparison of multiple models, using model selection criteria of validation data: Sensitivity (91.93%), Specificity (85.84%), Misclassification (11.94%) and Lift (2.7).

Topology and random-walk network representation of cardiac dynamics for localization of myocardial infarction.

While detection of acute cardiac disorders such as myocardial infarction (MI) from electrocardiogram (ECG) and vectorcardiogram (VCG) has been widely reported, identification of MI locations from these signals, pivotal for timely therapeutic and prognostic interventions, remains a standing issue. We present an approach for MI localization based on representing complex spatiotemporal patterns of cardiac dynamics as a random-walk network reconstructed from the evolution of VCG signals across a 3-D state space. Extensive tests with signals from the PTB database of the PhysioNet databank suggest that locations of MI can be determined accurately (sensitivity of ∼88% and specificity of ∼92%) from tracking certain consistently estimated invariants of this random-walk representation.

Real-time lumped parameter modeling of cardiovascular dynamics using electrocardiogram signals: toward virtual cardiovascular instruments.

We present an approach to deriving a real-time, lumped parameter cardiovascular dynamics model that uses features extracted from online electrocardiogram (ECG) signal recordings to generate certain surrogate hemodynamic signals. The model represents the coupled dynamics of the heart chambers, valves, and pulmonary and systemic blood circulation loops in the form of nonlinear differential equations. The features extracted from ECG signals were used to estimate the timings and amplitudes of the atrioventricular activation input functions as well as other model parameters that capture the effect of cardiac morphological and physiological characteristics. The model was tested using hemodynamic signals from the PhysioNet MGH/MF Waveform database. The results suggest that the model can capture the salient time and frequency patterns of the measured central venous pressure, pulmonary arterial pressure, and respiratory impedance signals (R(2) > 0.65). We have developed a method based on Anderson-Darling statistic and Kullback-Leibler divergence to compare the clinical measures (i.e., systolic and diastolic pressures) estimated from model waveform-extrema with those from actual measurements. The test statistics of the model waveform-extrema were statistically indistinguishable from the measured values with beat-to-beat rejection rates of 10%. The results indicate the potential of a virtual instrument that uses the model-derived signals for clinical diagnosis in lieu of expensive instrumentation.

Wireless Wearable Multisensory Suite and Real-Time Prediction of Obstructive Sleep Apnea Episodes.

Obstructive sleep apnea (OSA) is a common sleep disorder found in 24% of adult men and 9% of adult women. Although continuous positive airway pressure (CPAP) has emerged as a standard therapy for OSA, a majority of patients are not tolerant to this treatment, largely because of the uncomfortable nasal air delivery during their sleep. Recent advances in wireless communication and advanced ("bigdata") preditive analytics technologies offer radically new point-of-care treatment approaches for OSA episodes with unprecedented comfort and afforadability. We introduce a Dirichlet process-based mixture Gaussian process (DPMG) model to predict the onset of sleep apnea episodes based on analyzing complex cardiorespiratory signals gathered from a custom-designed wireless wearable multisensory suite. Extensive testing with signals from the multisensory suite as well as PhysioNet's OSA database suggests that the accuracy of offline OSA classification is 88%, and accuracy for predicting an OSA episode 1-min ahead is 83% and 3-min ahead is 77%. Such accurate prediction of an impending OSA episode can be used to adaptively adjust CPAP airflow (toward improving the patient's adherence) or the torso posture (e.g., minor chin adjustments to maintain steady levels of the airflow).

A low-cost, portable, high-throughput wireless sensor system for phonocardiography applications.

This paper presents the design and testing of a wireless sensor system developed using a Microchip PICDEM developer kit to acquire and monitor human heart sounds for phonocardiography applications. This system can serve as a cost-effective option to the recent developments in wireless phonocardiography sensors that have primarily focused on Bluetooth technology. This wireless sensor system has been designed and developed in-house using off-the-shelf components and open source software for remote and mobile applications. The small form factor (3.75 cm × 5 cm × 1 cm), high throughput (6,000 Hz data streaming rate), and low cost ($13 per unit for a 1,000 unit batch) of this wireless sensor system make it particularly attractive for phonocardiography and other sensing applications. The experimental results of sensor signal analysis using several signal characterization techniques suggest that this wireless sensor system can capture both fundamental heart sounds (S1 and S2), and is also capable of capturing abnormal heart sounds (S3 and S4) and heart murmurs without aliasing. The results of a denoising application using Wavelet Transform show that the undesirable noises of sensor signals in the surrounding environment can be reduced dramatically. The exercising experiment results also show that this proposed wireless PCG system can capture heart sounds over different heart conditions simulated by varying heart rates of six subjects over a range of 60-180 Hz through exercise testing.

Identification of myocardial infarction (MI) using spatio-temporal heart dynamics.

Cardiovascular disorders, such as myocardial infarction (MI) are the leading causes of mortality in the world. This paper presents an approach that uses novel spatio-temporal patterns of the vectorcardiogram (VCG) signals for the identification of various types of MI. In contrast to the traditional electrocardiogram (ECG) approaches, the 3D cardiac VCG signal is partitioned into 8 octants for localized analysis of the heart's electrical activities. The proposed method was tested using the PhysioNet PTB database for 368 MIs and 80 healthy control (HC) recordings, each of which includes 12-lead ECG and 3-lead VCG. Significant differences are found in the VCG spatial distribution between MI and HC groups. Furthermore, classification and regression tree (CART) analysis was used to demonstrate that VCG octant features can distinguish MIs from HCs with a sensitivity (accuracy of MI identification) of 97.28% and a specificity (accuracy of HC identification) of 95.00%, which is promising compared to the previously reported results using other ECG databases. The results indicate that the present approach provides an effective way for monitoring, post-processing, and interpretation of ECG data, and hopefully can impact the current cardiac diagnostic practice.

Practical training framework for fitting a function and its derivatives.

This paper describes a practical framework for using multilayer feedforward neural networks to simultaneously fit both a function and its first derivatives. This framework involves two steps. The first step is to train the network to optimize a performance index, which includes both the error in fitting the function and the error in fitting the derivatives. The second step is to prune the network by removing neurons that cause overfitting and then to retrain it. This paper describes two novel types of overfitting that are only observed when simultaneously fitting both a function and its first derivatives. A new pruning algorithm is proposed to eliminate these types of overfitting. Experimental results show that the pruning algorithm successfully eliminates the overfitting and produces the smoothest responses and the best generalization among all the training algorithms that we have tested.

Forecasting the evolution of nonlinear and nonstationary systems using recurrence-based local Gaussian process models.

An approach based on combining nonparametric Gaussian process (GP) modeling with certain local topological considerations is presented for prediction (one-step look ahead) of complex physical systems that exhibit nonlinear and nonstationary dynamics. The key idea here is to partition system trajectories into multiple near-stationary segments by aligning the boundaries of the partitions with those of the piecewise affine projections of the underlying dynamic system, and deriving nonparametric prediction models within each segment. Such an alignment is achieved through the consideration of recurrence and other local topological properties of the underlying system. This approach was applied for state and performance forecasting in Lorenz system under different levels of induced noise and nonstationarity, synthetic heart-rate signals, and a real-world time-series from an industrial operation known to exhibit highly nonlinear and nonstationary dynamics. The results show that local Gaussian process can significantly outperform not just classical system identification, neural network and nonparametric models, but also the sequential Bayesian Monte Carlo methods in terms of prediction accuracy and computational speed.

Linear affine transformations between 3-lead (Frank XYZ leads) vectorcardiogram and 12-lead electrocardiogram signals.

BACKGROUND: Recent advances in computer graphics and wireless technologies have renewed interest in vectorcardiogram (VCG) signals that use fewer leads than the conventional 12-lead electrocardiogram (ECG) signals for medical diagnostic applications. However, most cardiologists are accustomed to the 12-lead ECG even though some of the leads are either nearly aligned with or derived from the others and consequently contain redundant information. The ability to transform from orthogonal 3-lead VCG to 12-lead ECG enables the use of fewer leads for signal analysis, computer visualization, and wireless transmission of signals. This can also improve mobility, albeit limited, to the patients. MATERIALS AND METHODS: We present a statistical approach to transform 3-lead Frank VCG to 12-lead ECG signals and vice versa, based on Dower's pioneering work on lead transformation. This approach enables compensation of baseline shifts and other constant biases present in long ECG data streams, so that the resulting statistical transforms can be more consistent and accurate. We compare the performance of the affine transform with that of Dower transform (from 3 to 12 and from 12 to 3) using the data from the PhysioNet PTB database. RESULTS: The results show that for both myocardial infarction (MI) and healthy control (HC) subjects, the statistical affine transform presented here maps 3-lead VCG to12-lead ECG more accurately than Dower or other lead transformation matrices of the ECG recordings. DISCUSSION: This investigation also shows the limitations associated with single dipole assumption that underlies Dower's geometric transformation. The results also indicate that lead transformation accuracy can be improved using separate customized transforms to, for example, age or pathologic conditions (here, MI vs HC) than a single statistical or geometric transform. Pertinently, we find that the affine transform coefficients can serve as discriminating features for classification/discrimination of MI patients from HC subjects.