Search

This paper considers a problem of estimating brain effective connectivity from EEG signals using a Granger causality (GC) concept characterized on state-space models. We propose a state-space model for explaining coupled dynamics of the source and EEG signals where EEG is a linear combination of sources according to the characteristics of volume conduction. Our formulation has a sparsity prior on the source output matrix that can further classify active and inactive sources. The scheme is comprised of two main steps: model estimation and model inference to estimate brain connectivity. The model estimation consists of performing a subspace identification and the active source selection based on a group-norm regularized least-squares. The model inference relies on the concept of state-space GC that requires solving a discrete-time Riccati equation for the covariance of estimation error. We verify the performance on simulated data sets that represent realistic human brain activities under several conditions including percentages of active sources, a number of EEG electrodes and the location of active sources. The performance of estimating brain networks is compared with a two-stage approach using source reconstruction algorithms and VAR-based Granger analysis. Our method achieved better performances than the two-stage approach under the assumptions that the true source dynamics are sparse and generated from state-space models. The method is applied to a real EEG SSVEP data set and we found that the temporal lobe played a role of a mediator of connections between temporal and occipital areas, which agreed with findings in previous studies.

We design an algorithm to automatically detect epileptic seizure onsets and offsets from scalp EEGs. The proposed scheme consists of two sequential steps: detecting seizure episodes from long EEG recordings, and determining seizure onsets and offsets of the detected episodes. We introduce a neural network-based model called ScoreNet to carry out the second step by better predicting the seizure probability of pre-detected seizure epochs to determine seizure onsets and offsets. A cost function called log-dice loss with a similar meaning to the F1 score is proposed to handle the natural data imbalance inherent in EEG signals signifying seizure events. ScoreNet is then verified on the CHB-MIT Scalp EEG database in combination with several classifiers including random forest, CNN, and logistic regression. As a result, ScoreNet improves seizure detection performance over lone epoch-based seizure classification methods; F1 scores increase significantly from 16–37% to 53–70%, and false positive rates per hour decrease from 0.53–5.24 to 0.05–0.61. This method provides clinically acceptable latencies of detecting seizure onset and offset of less than 10 seconds. In addition, an effective latency index is proposed as a metric for detection latency whose scoring considers undetected events to provide better insight into onset and offset detection than conventional time-based metrics.

Working memory (WM) encompasses crucial cognitive processes or abilities to retain and manipulate temporary information for immediate execution of complex cognitive tasks in daily functioning such as reasoning and decision-making. The WM of individuals sustaining traumatic brain injury (TBI) was commonly compromised, especially in the domain of WM. The current study investigated the brain responses of WM in a group of participants with mild-moderate TBI compared to their healthy counterparts employing functional magnetic resonance imaging. All consented participants (healthy: n = 26 and TBI: n = 15) performed two variations of the n-back WM task with four load conditions (0-, 1-, 2-, and 3-back). The respective within-group effects showed a right hemisphere-dominance activation and slower reaction in performance for the TBI group. Random-effects analysis revealed activation difference between the two groups in the right occipital lobe in the guided n-back with cues, and in the bilateral occipital lobe, superior parietal region, and cingulate cortices in the n-back without cues. The left middle frontal gyrus was implicated in the load-dependent processing of WM in both groups. Further group analysis identified that the notable activation changes in the frontal gyri and anterior cingulate cortex are according to low and high loads. Though relatively smaller in scale, this study was eminent as it clarified the neural alterations in WM in the mild-moderate TBI group compared to healthy controls. It confirmed the robustness of the phenomenon in TBI with the reproducibility of the results in a heterogeneous non-Western sample.

This paper considers joint learning of multiple sparse Granger graphical models to discover underlying common and differential Granger causality (GC) structures across multiple time series. This can be applied to drawing group-level brain connectivity inferences from a homogeneous group of subjects or discovering network differences among groups of signals collected under heterogeneous conditions. By recognizing that the GC of a single multivariate time series can be characterized by common zeros of vector autoregressive (VAR) lag coefficients, a group sparse prior is included in joint regularized least-squares estimations of multiple VAR models. Group-norm regularizations based on group- and fused-lasso penalties encourage a decomposition of multiple networks into a common GC structure, with other remaining parts defined in individual-specific networks. Prior information about sparseness and sparsity patterns of desired GC networks are incorporated as relative weights, while a non-convex group norm in the penalty is proposed to enhance the accuracy of network estimation in low-sample settings. Extensive numerical results on simulations illustrated our method's improvements over existing sparse estimation approaches on GC network sparsity recovery. Our methods were also applied to available resting-state fMRI time series from the ADHD-200 data sets to learn the differences of causality mechanisms, called effective brain connectivity, between adolescents with ADHD and typically developing children. Our analysis revealed that parts of the causality differences between the two groups often resided in the orbitofrontal region and areas associated with the limbic system, which agreed with clinical findings and data-driven results in previous studies., Comment: 23 pages, 9 figures, 3 tables, Comments: Title changed, Math expression corrected in the formulation and algorithm section, More references and discussions are added to explain difficulties in the convergence analysis

This paper describes the release of the detailed building operation data, including electricity consumption and indoor environmental measurements, of the seven-story 11,700-m2 office building located in Bangkok, Thailand. The electricity consumption data (kW) are that of individual air conditioning units, lighting, and plug loads in each of the 33 zones of the building. The indoor environmental sensor data comprise temperature (°C), relative humidity (%), and ambient light (lux) measurements of the same zones. The entire datasets are available at one-minute intervals for the period of 18 months from July 1, 2018, to December 31, 2019. Such datasets can be used to support a wide range of applications, such as zone-level, floor-level, and building-level load forecasting, indoor thermal model development, validation of building simulation models, development of demand response algorithms by load type, anomaly detection methods, and reinforcement learning algorithms for control of multiple AC units., Measurement(s) electrical energy • temperature of air • humidity • visible spectrum radiation Technology Type(s) Gauge or Meter Device • Sensor Device Factor Type(s) floor • date/time Sample Characteristic - Environment building Machine-accessible metadata file describing the reported data: 10.6084/m9.figshare.12527219

We establish a deep learning-based method to automatically detect the epileptic seizure onsets and offsets in multi-channel electroencephalography (EEG) signals. A convolutional neural network (CNN) is designed to identify occurrences of seizures in EEG epochs from the EEG signals and an onset-offset detector is proposed to determine the seizure onsets and offsets. The EEG signals are considered as inputs and the outputs are the onset and offset. In the CNN, a filter is factorized to separately capture temporal and spatial patterns in EEG epochs. Moreover, we develop an onset-offset detection method based on clinical decision criteria. As a result, verified on the whole CHB-MIT Scalp EEG database, the CNN model correctly detected seizure activities over 90%. Furthermore, combined with the onset-offset detector, this method accomplished F 1 of 64.40% and essentially determined the seizure onset and offset with absolute onset and offset latencies of 5.83 and 10.12 seconds, respectively.

This paper aims to apply machine learning techniques to an automated epileptic seizure detection using EEG signals to help neurologists in a time-consuming diagnostic process. We employ two approaches based on convolution neural networks (CNNs) and artificial neural networks (ANNs) to provide a probability of seizure occurrence in a windowed EEG recording of 18 channels. In order to extract relevant features based on time, frequency, and time-frequency domains for these networks, we consider an improvement of the Bayesian error rate from a baseline. Features of which the improvement rates are higher than the significant level are considered. These dominant features extracted from all EEG channels are concatenated as the input for ANN with 7 hidden layers, while the input of CNN is taken as raw multi-channel EEG signals. Using multi-concept of deep CNN in image processing, we exploit 2D-filter decomposition to handle the signal in spatial and temporal domains. Our experiments based on CHB-MIT Scalp EEG Database showed that both ANN and CNN were able to perform with the overall accuracy of up to 99.07% and F1-score of up to 77.04%. ANN with dominant features is more capable of detecting seizure events than CNN whereas CNN requiring no feature extraction is slightly better than ANN in classification accuracy.

Since the manual detection of electrographic seizures in continuous electroencephalogram (EEG) monitoring is very time-consuming and requires a trained expert, attempts to develop automatic seizure detection are diverse and ongoing. Machine learning approaches are intensely being applied to this problem due to their ability to classify seizure conditions from a large amount of data, and provide pre-screened results for neurologists. Several features, data transformations, and classifiers have been explored to analyze and classify seizures via EEG signals. In the literature, some jointly-applied features used in the classification may have shared similar contributions, making them redundant in the learning process. Therefore, this paper aims to comprehensively summarize feature descriptions and their interpretations in characterizing epileptic seizures using EEG signals, as well as to review classification performance metrics. To provide meaningful information of feature selection, we conducted an experiment to examine the quality of each feature independently. The Bayesian error and non-parametric probability distribution estimation were employed to determine the significance of the individual features. Moreover, a redundancy analysis using a correlation-based feature selection was applied. The results showed that the following features --variance, energy, nonlinear energy, and Shannon entropy computed on a raw EEG signal, as well as variance, energy, kurtosis, and line length calculated on wavelet coefficients-- were able to significantly capture the seizures. An improvement of 4.77--13.51% in the Bayesian error from the baseline was obtained., Comment: 28 pages, 4 figures

Functional magnetic resonance imaging (fMRI) technique allows us to capture activities occurring in a human brain via signals from blood flow, known as BOLD (blood oxygen level-dependent) signals. Exploring a relationship among brain regions inside human brains from fMRI data is an active and challenging research topic. Relationships or associations between brain regions are commonly referred to as brain connectivity or brain network. This connectivity can be divided into two groups, the functional connectivity which describes the statistical information among brain regions and the effective connectivity which specifies how one region interacts with others by a causal model. This survey paper provides a review on learning brain connectivities via fMRI data, mathematical definitions or dependence measures of such connectivities. These well-known measures include correlation, partial correlation, conditional independence, dynamical causal modeling, Granger causality, and structural equation modeling, which all can be translated in terms of mathematical conditions of model parameters. We also discusses about relevant estimation techniques that have been widely used in the problems of fMRI modeling. Understanding a rigorous definition on relationships in human brain allows us to interpret or compare the results in the context of learning brain network more clearly.

Electroencephalography (EEG) signals are known to be generated from the current source signals occurring inside human brains and these sources may or may not be active concurrently at a certain time. This paper aims to classify active and inactive sources from the information that can be inferred from parameters of a dynamical model that captures characteristics of EEG time series. We propose a state-space model for explaining coupled dynamics of the source and EEG signals where EEG is a linear combination of sources according to the characteristics of volume conduction. Our model has a structure that the sparsity pattern of the model output matrix can indicate the position of active and inactive sources. With this assumption, the proposed estimation method consists of two steps. Firstly, a subspace identification method is performed to estimate the dynamic matrix of the model and the mapping matrix from the state variable to EEG output. Secondly, the estimation of the output matrix in the state-space model from the mapping matrix is solved by a group lasso problem to promote a sparsity pattern. As a result, nonzero rows of the output matrix represent active source that corresponding to EEG data. We verify the performance of our method on randomly generated data sets that represent realistic human brain activities in a fair setting. An acceptable accuracy of 95 – 98% is obtained with a suitable selection of a problem parameter and a thresholding process to discard small magnitudes of the output matrix.

This paper considers an estimation of state-space system matrices of a building temperature system. The building model has two rooms equipped with air-conditioning (AC) system where they contain heat interactions with walking path and the outdoor. Only one-dimensional heat transfer and internal energy change have been concerned while neglecting the effects caused by humidity and air leakage. Cooling energy from the AC system is expressed via the coefficient of performance (COP) and is a nonlinear function of room temperatures. We propose an approximate linear model with an assumption that the COP is a constant obtained by the air-conditioner specifications or it can be estimated. Temperature data and air-conditioners electrical input data can be collected by the Chulalongkorn University Building Energy Management System (CUBEMS) and therefore allow us to estimate the model parameters using the least-squares approach. Our results show that the model with a structure on the input matrix that agrees with physically-based dynamical equations is recommended and including the solar radiation as an input is not necessary.

This paper presents two statistical methods for refining predicted solar irradiance from Numerical Weather Prediction (NWP). Firstly, the spatial averaging is used to reduce the variance of solar irradiance prediction over a spatial area of interest. Secondly, a model output statistics method which is a linear regression model, is applied to explain relationships between solar irradiance and relevant weather variables. In this study, we perform a variable selection based on three statistical methods: partial correlation, stepwise regression, and subset regression, to specify the important meteorological variables for solar irradiance prediction in Thailand. The results show that spatial averaging of NWP forecasts over $7 \times 7$ grid points provides a reduction of 3.42% Normalized Root Mean Square Error (NRMSE) from NWP forecasts. Model output statistics with and without variable selection can further improve output from spatial averaging by 41.32% and 41.89% NRMSE respectively. We conclude that highly relevant variables including irradiance measurements, ambient temperature, relative humidity, UV index, and NWP forecasts are sufficient for a significant improvement of NWP forecasts in Thailand.

Path analysis is a model class of structural equation modeling (SEM), which it describes causal relations among measured variables in the form of a multiple linear regression. This paper presents two estimation formulations, one each for confirmatory and exploratory SEM, where a zero pattern of the estimated path coefficient matrix can explain a causality structure of the variables. The original nonlinear equality constraints of the model parameters were relaxed to an inequality, allowing the transformation of the original problem into a convex framework. A regularized estimation formulation was then proposed for exploratory SEM using an l1-type penalty of the path coefficient matrix. Under a condition on problem parameters, our optimal solution is low rank and provides a useful solution to the original problem. Proximal algorithms were applied to solve our convex programs in a large-scale setting. The performance of this approach was demonstrated in both simulated and real data sets, and in comparison with an existing method. When applied to two real application results (learning causality among climate variables in Thailand and examining connectivity differences in autism patients using fMRI time series from ABIDE data sets) the findings could explain known relationships among environmental variables and discern known and new brain connectivity differences, respectively., Comment: 24 pages, 19 figures

This paper aims to explain relationships between time series by using the Granger causality (GC) concept through autoregressive (AR) models and to assure the model stability. Examining such GC relationship is performed on the model parameters using the Wald test and the model stability is guaranteed by the infinity-norm constraint on the dynamic matrix of the AR process. The proposed formulation is a least-squares estimation with Granger causality and stability constraints which is a convex program with a quadratic objective subject to linear equality and inequality norm constraints. We show by simulations that various typical factors could lead to unstable estimated models when using an unconstrained method. Estimated models from our approach are guaranteed to be stable but the model fitting error could be conservatively increased due to the selected stability condition.

This paper considers a solar irradiance forecasting using seasonal autoregressive integrated moving average models (SARIMA) which finds an application in solar power prediction for energy management. A time series prediction typically requires a complete set of historical data while the global horizontal irradiance (GHI) data in Thailand are frequently missing for consecutive days. This paper proposes a data-missing imputation technique using the mean value of GHI averaged over the data corresponding to the same weather type. The required weather classification consists of two steps: a seasonal segmentation based on detecting changes of monotonic properties of temperature and humidity time series, and a nonlinear support vector machine (SVM) that uses weather labels from the previous seasonal segmentation. Experimental results on the GHI data in Bangkok, Thailand shows that the proposed method significantly provides decreased imputation errors and leads to a better solar forecasting performance compared to other imputation techniques.

Patellar resurfacing during total knee arthroplasty (TKA) depends on surgeon's decisions and has been discussed if resurfacing is better than non-resurfacing. This paper aims to discuss the similarities and differences of the vibroarthrographic (VAG) signal of the knee underwent TKA with either resurfaced or non-resurfaced patella in time-frequency domain. Motion trends and noises were filtered and then processed with Ensemble Empirical Model Decomposition (EEMD) and Detrended Fluctuation Analysis (DFA), which decomposes a non-stationary signal into a set of modes and discards modes that are uncorrelated. However, time-frequency analysis of processed signals suggests no significant differences between both classes and the results rather depend on the individual knee condition.

Path analysis is a special problem in Structural Equation Modeling (SEM) where its model describes causal relations among measured variables in a form of multiple linear regression. This work presents an alternative estimation formulation for a special case problem in path analysis as a convex framework by relaxing an equality constraint of the original problem. Under a condition on problem parameters, we show that, our optimal solution is low rank and provides an estimate of the path matrix of the original problem. To solve our estimation problem in a convex framework, we apply the alternating direction method of multipliers (ADMM) algorithm which is suitable for large-scale implementation. The performance of this algorithm is demonstrated in numerical experiments.

This paper proposes a formulation of estimating a sparse image signal where the sparsity only occurs on the high frequency components. The formulation is for compressed sensing reconstruction when a compressed measurement signal is contaminated by impulsive noise. The approach is based on two prior well-known assumptions. First, adding an L 1 -norm penalty on a compressible signal to the cost objective of the estimation problem will promote a sparsity in the signal. Secondly, an image signal should be sparse in high frequency domain, while it is dense in low frequency domain where most of its important information lies. Our approach is therefore to impose an L 1 -norm penalty only on the signal components that need to be sparse while we allow the important low frequency part to be dense. The obtained formulation is a convex optimization which can be effectively solved by many available algorithms. The experiment demonstrated a superior impulsive noise tolerance of the proposed penalty function to a conventional scheme where all signal components are penalized by L 1 -norm. At the noise probability of less than 0.1, the reconstruction was better both quantitatively and qualitatively.

We propose a convex optimization method for maximum likelihood estimation of autoregressive models, subject to conditional independence constraints. This problem is an extension to times series of the classical covariance selection problem in graphical modeling. The conditional independence constraints impose quadratic equalities on the autoregressive model parameters, which makes the maximum likelihood estimation problem nonconvex and difficult to solve. We formulate a convex relaxation and prove that it is exact when the sample covariance matrix is block-Toeplitz. We also observe experimentally that in practice the relaxation is exact under much weaker conditions. We discuss applications to topology selection in graphical models of time series, by enumerating all possible topologies, and ranking them using information-theoretic model selection criteria. The method is illustrated by an example of air pollution data.

This paper describes the dynamic models of a double rotary inverted pendulum, which has been developed for the laboratory experiments. Two different-length rigid pendulums are connected to a horizontally rotating disc which is attached directly to a DC motor. The derivation of the dynamical equations and the linearized model are described. Finally, the time responses of open-loop system are shown and compared with the experimental data to verify the model validity.

This paper concerns dynamics of a servo-driven conveyor system, which has been developed for the laboratory-scale experiments. This system consists of a belt lying on the iron plate with two shafts at the end. A DC motor is used to drive the belt through the shaft. The belt is considered to behave like a nonlinear spring. We include the Lugre friction in this model. Then, the dynamics of the conveyor system are derived based on certain specified assumptions. At the same time, the linear and linearized model is used to analyze the behaviors for small signals. Finally, the time responses of these models are shown to verify the modelling analysis.