Yun Chen

Self-organized neural network for the quality control of 12-lead ECG signals

Telemedicine is very important for the timely delivery of health care to cardiovascular patients, especially those who live in the rural areas of developing countries. However, there are a number of uncertainty factors inherent to the mobile-phone-based recording of electrocardiogram (ECG) signals such as personnel with minimal training and other extraneous noises. PhysioNet organized a challenge in 2011 to develop efficient algorithms that can assess the ECG signal quality in telemedicine settings. This paper presents our efforts in this challenge to integrate multiscale recurrence analysis with a self-organizing map for controlling the ECG signal quality. As opposed to directly evaluating the 12-lead ECG, we utilize an information-preserving transform, i.e. Dower transform, to derive the 3-lead vectorcardiogram (VCG) from the 12-lead ECG in the first place. Secondly, we delineate the nonlinear and nonstationary characteristics underlying the 3-lead VCG signals into multiple time-frequency scales. Furthermore, a self-organizing map is trained, in both supervised and unsupervised ways, to identify the correlations between signal quality and multiscale recurrence features. The efficacy and robustness of this approach are validated using real-world ECG recordings available from PhysioNet. The average performance was demonstrated to be 95.25% for the training dataset and 90.0% for the independent test dataset with unknown labels.

Spatiotemporal Differentiation of Myocardial Infarctions

Myocardial infarction (MI), also known as a heart attack, is the leading cause of death in the U.S. It often occurs due to the occlusion of coronary arteries, thereby leading to insufficient blood and oxygen supply that damage cardiac muscle cells. Because blood vessels are branching throughout the heart, MI occurs at different spatial locations (e.g., anterior and inferior portions) of the heart. The spatial location of the diseased is rupts normal excitation and propagation of cardiac electrical activity in space and time. Most previous studies focused on the relationships between disease and time-domain biomarkers from 12-lead ECG signals (e.g., Q wave, QT interval, ST elevation/depression, T wave). Few, if any, previous approaches investigated how the spatial location of diseases will alter cardiac vectorcardiogram (VCG) signals in both space and time. This paper presents a novel spatiotemporal warping approach to quantify the dissimilarity of disease-altered patterns in 3-lead spatiotemporal VCG signals. The hypothesis testing shows that there are significant spatiotemporal differences between healthy control, MI-anterior, MI-anterior-septal, MI-anterior-lateral, MI-inferior, and MI-inferior-lateral. Furthermore, we optimize the embedding of each functional recording as a feature vector in the high-dimensional space that preserves the dissimilarity distance matrix. This novel spatial embedding approach facilitates the construction of classification models and yields an averaged accuracy of 95.1% for separating MIs and Healthy Controls (HCs) and an averaged accuracy of 95.8% in identifying anterior-related MIs and inferior-related MIs.

Heterogeneous recurrence monitoring and control of nonlinear stochastic processes

Recurrence is one of the most common phenomena in natural and engineering systems. Process monitoring of dynamic transitions in nonlinear and nonstationary systems is more concerned with aperiodic recurrences and recurrence variations. However, little has been done to investigate the heterogeneous recurrence variations and link with the objectives of process monitoring and anomaly detection. Notably, nonlinear recurrence methodologies are based on homogeneous recurrences, which treat all recurrence states in the same way as black dots, and non-recurrence is white in recurrence plots. Heterogeneous recurrences are more concerned about the variations of recurrence states in terms of state properties (e.g., values and relative locations) and the evolving dynamics (e.g., sequential state transitions). This paper presents a novel approach of heterogeneous recurrence analysis that utilizes a new fractal representation to delineate heterogeneous recurrence states in multiple scales, including the recurrences of both single states and multi-state sequences. Further, we developed a new set of heterogeneous recurrence quantifiers that are extracted from fractal representation in the transformed space. To that end, we integrated multivariate statistical control charts with heterogeneous recurrence analysis to simultaneously monitor two or more related quantifiers. Experimental results on nonlinear stochastic processes show that the proposed approach not only captures heterogeneous recurrence patterns in the fractal representation but also effectively monitors the changes in the dynamics of a complex system.

Mobile sensing and network analytics for realizing smart automated systems towards health Internet of Things

Internet of Things (IoT) provides an unprecedented opportunity to realize smart automated systems such as smart manufacturing, smart city and smart home in the past few years. Pervasive sensing and mobile technology deployed in large-scale IoT systems lead to the accumulation of big data. In particular, wearable biosensing accelerates human-centered computing for smart health management. However, limited work has been done to develop advanced IoT technologies for smart monitoring and control of heart health. There is an urgent need to develop a new IoT technology specific to the heart, namely Internet of Hearts (IOH) that will enable and assist (1) the acquisition of electrocardiogram (ECG) signals pertinent to space-time cardiac dynamics at anytime anywhere; (2) real-time management and compact representation of multi-sensor signals; (3) big data analytics in large-scale IoT contexts. This paper presents a new technology of Mobile and E-Network Smart Health (MESH), which is composed of 4 components as follows: 1) Mobile-based ECG sensing device; 2) Space-time representation of cardiac electrical activity; 3) Optimal model-based representation of ECG signals; 4) Dynamic network embedding for disease pattern recognition. Our preliminary experimental results demonstrated that network analytics is efficient and effective for smart health management in IoT contexts. The MESH technology shows strong potentials to provide an indispensable and enabling tool for realizing smart heart health and wellbeing for the population worldwide.

Heterogeneous postsurgical data analytics for predictive modeling of mortality risks in intensive care units.

The rapid advancements of biomedical instrumentation and healthcare technology have resulted in data-rich environments in hospitals. However, the meaningful information extracted from rich datasets is limited. There is a dire need to go beyond current medical practices, and develop data-driven methods and tools that will enable and help (i) the handling of big data, (ii) the extraction of data-driven knowledge, (iii) the exploitation of acquired knowledge for optimizing clinical decisions. This present study focuses on the prediction of mortality rates in Intensive Care Units (ICU) using patient-specific healthcare recordings. It is worth mentioning that postsurgical monitoring in ICU leads to massive datasets with unique properties, e.g., variable heterogeneity, patient heterogeneity, and time asyncronization. To cope with the challenges in ICU datasets, we developed the postsurgical decision support system with a series of analytical tools, including data categorization, data pre-processing, feature extraction, feature selection, and predictive modeling. Experimental results show that the proposed data-driven methodology outperforms traditional approaches and yields better results based on the evaluation of real-world ICU data from 4000 subjects in the database. This research shows great potentials for the use of data-driven analytics to improve the quality of healthcare services.The rapid advancements of biomedical instrumentation and healthcare technology have resulted in data-rich environments in hospitals. However, the meaningful information extracted from rich datasets is limited. There is a dire need to go beyond current medical practices, and develop data-driven methods and tools that will enable and help (i) the handling of big data, (ii) the extraction of data-driven knowledge, (iii) the exploitation of acquired knowledge for optimizing clinical decisions. This present study focuses on the prediction of mortality rates in Intensive Care Units (ICU) using patient-specific healthcare recordings. It is worth mentioning that postsurgical monitoring in ICU leads to massive datasets with unique properties, e.g., variable heterogeneity, patient heterogeneity, and time asyncronization. To cope with the challenges in ICU datasets, we developed the postsurgical decision support system with a series of analytical tools, including data categorization, data pre-processing, feature extraction, feature selection, and predictive modeling. Experimental results show that the proposed data-driven methodology outperforms traditional approaches and yields better results based on the evaluation of real-world ICU data from 4000 subjects in the database. This research shows great potentials for the use of data-driven analytics to improve the quality of healthcare services.

Sparse Modeling and Recursive Prediction of Space–Time Dynamics in Stochastic Sensor Networks

Wireless sensor network has emerged as a key technology for monitoring space-time dynamics of complex systems, e.g., environmental sensor network, battlefield surveillance network, and body area sensor network. However, sensor failures are not uncommon in traditional sensing systems. As such, we propose the design of stochastic sensor networks to allow a subset of sensors at varying locations within the network to transmit dynamic information intermittently. Realizing the full potential of stochastic sensor network hinges on the development of novel information-processing algorithms to support the design and exploit the uncertain information for decision making. This paper presents a new approach of sparse particle filtering to model spatiotemporal dynamics of big data in the stochastic sensor network. Notably, we developed a sparse kernel-weighted regression model to achieve a parsimonious representation of spatial patterns. Further, the parameters of spatial model are transformed into a reduced-dimension space, and thereby sequentially updated with the recursive Bayesian estimation when new sensor observations are available over time. Therefore, spatial and temporal processes closely interact with each other. Experimental results on real-world data and different scenarios of stochastic sensor networks (i.e., spatially, temporally, and spatiotemporally dynamic networks) demonstrated the effectiveness of sparse particle filtering to support the stochastic design and harness the uncertain information for modeling space-time dynamics of complex systems.

Model-driven parametric monitoring of high-dimensional nonlinear functional profiles

In order to cope with system complexity and dynamic environments, modern industries are investing in a variety of sensor networks and data acquisition systems to increase information visibility. Multi-sensor systems bring the proliferation of high-dimensional functional profiles that capture rich information on the evolving dynamics of natural and engineered processes. This provides an unprecedented opportunity for online monitoring of operational quality and integrity of complex systems. However, the classical methodology of statistical process control is not concerned about high-dimensional sensor signals and is limited in the capability to perform multi-sensor fault diagnostics. It is not uncommon that multi-dimensional sensing capabilities are not fully utilized for decision making. This paper presents a new model-driven parametric monitoring strategy for the detection of dynamic fault patterns in high-dimensional functional profiles that are nonlinear and nonstationary. First, we developed a sparse basis function model of high-dimensional functional profiles, thereby reducing the large amount of data to a parsimonious set of model parameters (i.e., weight, shifting and scaling factors) while preserving the information. Further, we utilized the lasso-penalized logistic regression model to select a low-dimensional set of sensitive predictors for fault diagnostics. Experimental results on real-world data from patient monitoring showed that the proposed methodology outperforms traditional methods and effectively identify a sparse set of sensitive features from high-dimensional datasets for process monitoring and fault diagnostics.

Heterogeneous recurrence T 2 charts for monitoring and control of nonlinear dynamic processes

Many real-world systems are evolving over time and exhibit dynamical behaviors. Real-time sensing brings the proliferation of big data (i.e., dynamic, nonlinear, nonstationary, high dimensional) that contains rich information on nonlinear dynamic processes. Nonetheless, limited work on studying nonlinear dynamics underlying sensing data for quality control has been reported. This paper presents a new approach of heterogeneous recurrence T2 control chart for online monitoring and anomaly detection in nonlinear dynamic processes. A partition scheme, named Q-tree indexing, is firstly introduced to delineate local recurrence regions in the multidimensional continuous state space. Further, we designed a new fractal representation of state transitions, among recurrence regions, and then develop new measures to quantify heterogeneous recurrence patterns. Finally, we developed a multivariate Hotelling T2 Chart for on-line monitoring and predictive control of process recurrences. Case studies show that the proposed approach not only captures heterogeneous recurrence patterns in the transformed space, but also provides an effective online control charts to monitor and detect dynamical transitions in the underlying nonlinear process.

A Novel Information-Theoretic Approach for Variable Clustering and Predictive Modeling Using Dirichlet Process Mixtures

In the era of big data, there are increasing interests on clustering variables for the minimization of data redundancy and the maximization of variable relevancy. Existing clustering methods, however, depend on nontrivial assumptions about the data structure. Note that nonlinear interdependence among variables poses significant challenges on the traditional framework of predictive modeling. In the present work, we reformulate the problem of variable clustering from an information theoretic perspective that does not require the assumption of data structure for the identification of nonlinear interdependence among variables. Specifically, we propose the use of mutual information to characterize and measure nonlinear correlation structures among variables. Further, we develop Dirichlet process (DP) models to cluster variables based on the mutual-information measures among variables. Finally, orthonormalized variables in each cluster are integrated with group elastic-net model to improve the performance of predictive modeling. Both simulation and real-world case studies showed that the proposed methodology not only effectively reveals the nonlinear interdependence structures among variables but also outperforms traditional variable clustering algorithms such as hierarchical clustering.

Whole heart modeling — Spatiotemporal dynamics of electrical wave conduction and propagation

Cardiac electrical activities are varying in both space and time. Human heart consists of a fractal network of muscle cells, Purkinje fibers, arteries and veins. Whole-heart modeling of electrical wave conduction and propagation involves a greater level of complexity. Our previous work developed a computer model of the anatomically realistic heart and simulated the electrical conduction with the use of cellular automata. However, simplistic assumptions and rules limit its ability to provide an accurate approximation of real-world dynamics on the complex heart surface, due to sensitive dependence of nonlinear dynamical systems on initial conditions. In this paper, we propose new reaction-diffusion methods and pattern recognition tools to simulate and model spatiotemporal dynamics of electrical wave conduction and propagation on the complex heart surface, which include (i) whole heart model; (ii) 2D isometric graphing of 3D heart geometry; (iii) reaction-diffusion modeling of electrical waves in 2D graph, and (iv) spatiotemporal pattern recognition. Experimental results show that the proposed numerical solution has strong potentials to model the space-time dynamics of electrical wave conduction in the whole heart, thereby achieving a better understanding of disease-altered cardiac mechanisms.