An Ensemble CNN for Subject-Independent Classification of Motor Imagery-based EEG.

Deep learning methods, and in particular Convolutional Neural Networks (CNNs), have shown breakthrough performance in a wide variety of classification applications, including electroencephalogram-based Brain Computer Interfaces (BCIs). Despite the advances in the field, BCIs are still far from the subject-independent decoding of brain activities, primarily due to substantial inter-subject variability. In this study, we examine the potential application of an ensemble CNN classifier to integrate the capabilities of CNN architectures and ensemble learning for decoding EEG signals collected in motor imagery experiments. The results prove the superiority of the proposed ensemble CNN in comparison with the average base CNN classifiers, with an improvement up to 9% in classification accuracy depending on the test subject. The results also show improvement with respect to the performance of a number of state-of-the-art methods that have been previously used for subject-independent classification in the same datasets used here (i.e., BCI Competition IV 2A and 2B datasets).

Is Riemannian Geometry Better than Euclidean in Averaging Covariance Matrices for CSP-based Subject-Independent Classification of Motor Imagery?

Common Spatial Pattern (CSP) is a popular feature extraction algorithm used for electroencephalogram (EEG) data classification in brain-computer interfaces. One of the critical operations used in CSP is taking the average of trial covariance matrices for each class. In this regard, the arithmetic mean, which minimizes the sum of squared Euclidean distances to the data points, is conventionally used; however, this operation ignores the Riemannian geometry in the manifold of covariance matrices. To alleviate this problem, Fréchet mean determined using different Riemannian distances have been used. In this paper, we are primarily concerned with the following question: Does using the Fréchet mean with Riemannian distances instead of arithmetic mean in averaging CSP covariance matrices improve the subject-independent classification of motor imagery (MI)? To answer this question we conduct a comparative study using the largest MI dataset to date, with 54 subjects and a total of 21,600 trials of left-and right-hand MI. The results indicate a general trend of having a statistically significant better performance when the Riemannian geometry is used.

On the Classification of Electromyography Signals to Control a Four Degree-Of-Freedom Prosthetic Device.

This study investigates the applicability of Electromyography (EMG) signal classification algorithms with relatively low training time to control prosthetic devices. The perceived quality of control depends on many factors, such as the 1) accuracy of the algorithm, 2) the complexity of the control, and 3) the ability to compensate for the error. The high granularity of control in the time domain reduces the perceived effect of error but also limits the classification accuracy. This work aims to find the borderline for the accuracy of algorithms to be selected as a control strategy for hand prosthetic devices and thus shorten the gap between laboratory devices and commercially available devices. In particular, we compared five classification algorithms and selected one for real-time testing. The results from a test conducted on four subjects showed that the EMG-based control strategy has comparable performances with an IMU-based controller.

Learning Discriminative Spatiospectral Features of ERPs for Accurate Brain-Computer Interfaces.

Constructing accurate predictive models is at the heart of brain-computer interfaces (BCIs) because these models can ultimately translate brain activities into communication and control commands. The majority of the previous work in BCI use spatial, temporal, or spatiotemporal features of event-related potentials (ERPs). In this study, we examined the discriminatory effect of their spatiospectral features to capture the most relevant set of neural activities from electroencephalographic recordings that represent users' mental intent. In this regard, we model ERP waveforms using a sum of sinusoids with unknown amplitudes, frequencies, and phases. The effect of this signal modeling step is to represent high-dimensional ERP waveforms in a substantially lower dimensionality space, which includes their dominant power spectral contents. We found that the most discriminative frequencies for accurate decoding of visual attention modulated ERPs lie in a spectral range less than 6.4 Hz. This was empirically verified by treating dominant frequency contents of ERP waveforms as feature vectors in the state-of-the-art machine learning techniques used herein. The constructed predictive models achieved remarkable performance, which for some subjects was as high as 94% as measured by the area under curve. Using these spectral contents, we further studied the discriminatory effect of each channel and proposed an efficient strategy to choose subject-specific subsets of channels that generally led to classifiers with comparable performance.

Design and Optimization of an EEG-Based Brain Machine Interface (BMI) to an Upper-Limb Exoskeleton for Stroke Survivors.

This study demonstrates the feasibility of detecting motor intent from brain activity of chronic stroke patients using an asynchronous electroencephalography (EEG)-based brain machine interface (BMI). Intent was inferred from movement related cortical potentials (MRCPs) measured over an optimized set of EEG electrodes. Successful intent detection triggered the motion of an upper-limb exoskeleton (MAHI Exo-II), to guide movement and to encourage active user participation by providing instantaneous sensory feedback. Several BMI design features were optimized to increase system performance in the presence of single-trial variability of MRCPs in the injured brain: (1) an adaptive time window was used for extracting features during BMI calibration; (2) training data from two consecutive days were pooled for BMI calibration to increase robustness to handle the day-to-day variations typical of EEG, and (3) BMI predictions were gated by residual electromyography (EMG) activity from the impaired arm, to reduce the number of false positives. This patient-specific BMI calibration approach can accommodate a broad spectrum of stroke patients with diverse motor capabilities. Following BMI optimization on day 3, testing of the closed-loop BMI-MAHI exoskeleton, on 4th and 5th days of the study, showed consistent BMI performance with overall mean true positive rate (TPR) = 62.7 ± 21.4% on day 4 and 67.1 ± 14.6% on day 5. The overall false positive rate (FPR) across subjects was 27.74 ± 37.46% on day 4 and 27.5 ± 35.64% on day 5; however for two subjects who had residual motor function and could benefit from the EMG-gated BMI, the mean FPR was quite low (< 10%). On average, motor intent was detected -367 ± 328 ms before movement onset during closed-loop operation. These findings provide evidence that closed-loop EEG-based BMI for stroke patients can be designed and optimized to perform well across multiple days without system recalibration.

A Novel Experimental and Analytical Approach to the Multimodal Neural Decoding of Intent During Social Interaction in Freely-behaving Human Infants.

Understanding typical and atypical development remains one of the fundamental questions in developmental human neuroscience. Traditionally, experimental paradigms and analysis tools have been limited to constrained laboratory tasks and contexts due to technical limitations imposed by the available set of measuring and analysis techniques and the age of the subjects. These limitations severely limit the study of developmental neural dynamics and associated neural networks engaged in cognition, perception and action in infants performing "in action and in context". This protocol presents a novel approach to study infants and young children as they freely organize their own behavior, and its consequences in a complex, partly unpredictable and highly dynamic environment. The proposed methodology integrates synchronized high-density active scalp electroencephalography (EEG), inertial measurement units (IMUs), video recording and behavioral analysis to capture brain activity and movement non-invasively in freely-behaving infants. This setup allows for the study of neural network dynamics in the developing brain, in action and context, as these networks are recruited during goal-oriented, exploration and social interaction tasks.

EEG Source Imaging in Partial Epilepsy in Comparison with Presurgical Evaluation and Magnetoencephalography.

BACKGROUND AND PURPOSE: The aim of this study was to determine the usefulness of three-dimensional (3D) scalp EEG source imaging (ESI) in partial epilepsy in comparison with the results of presurgical evaluation, magnetoencephalography (MEG), and electrocorticography (ECoG). METHODS: The epilepsy syndrome of 27 partial epilepsy patients was determined by presurgical evaluations. EEG recordings were made using 70 scalp electrodes, and the 3D coordinates of the electrodes were digitized. ESI images of individual and averaged spikes were analyzed by Curry software with a boundary element method. MEG and ECoG were performed in 23 and 9 patients, respectively. RESULTS: ESI and MEG source imaging (MSI) results were well concordant with the results of presurgical evaluations (in 96.3% and 100% cases for ESI and MSI, respectively) at the lobar level. However, there were no spikes in the MEG recordings of three patients. The ESI results were well concordant with MSI results in 90.0% of cases. Compared to ECoG, the ESI results tended to be localized deeper than the cortex, whereas the MSI results were generally localized on the cortical surface. ESI was well concordant with ECoG in 8 of 9 (88.9%) cases, and MSI was also well concordant with ECoG in 4 of 5 (80.0%) cases. The EEG single dipoles in one patient with mesial temporal lobe epilepsy were tightly clustered with the averaged dipole when a 3 Hz high-pass filter was used. CONCLUSIONS: The ESI results were well concordant with the results of the presurgical evaluation, MSI, and ECoG. The ESI analysis was found to be useful for localizing the seizure focus and is recommended for the presurgical evaluation of intractable epilepsy patients.

Cortical activation pattern for grasping during observation, imagery, execution, FES, and observation-FES integrated BCI: an fNIRS pilot study.

Passive movement, action observation and motor imagery as well as motor execution have been suggested to facilitate the motor function of human brain. The purpose of this study is to investigate the cortical activation patterns of these four modes using a functional near-infrared spectroscopy (fNIRS) system. Seven healthy volunteers underwent optical brain imaging by fNIRS. Passive movements were provided by a functional electrical stimulation (FES). Results demonstrated that while all movement modes commonly activated premotor cortex, there were considerable differences between modes. The pattern of neural activation in motor execution was best resembled by passive movement, followed by motor imagery, and lastly by action observation. This result indicates that action observation may be the least preferred way to activate the sensorimotor cortices. Thus, in order to show the feasibility of motor facilitation by a brain computer interface (BCI) for an extreme case, we paradoxically adopted the observation as a control input of the BCI. An observation-FES integrated BCI activated sensorimotor system stronger than observation but slightly weaker than FES. This limitation should be overcome to utilize the observation-FES integrated BCI as an active motor training method.

Minimizing inter-subject variability in fNIRS-based brain-computer interfaces via multiple-kernel support vector learning.

Brain signal variation across different subjects and sessions significantly impairs the accuracy of most brain-computer interface (BCI) systems. Herein, we present a classification algorithm that minimizes such variation, using linear programming support-vector machines (LP-SVM) and their extension to multiple kernel learning methods. The minimization is based on the decision boundaries formed in classifiers' feature spaces and their relation to BCI variation. Specifically, we estimate subject/session-invariant features in the reproducing kernel Hilbert spaces (RKHS) induced with Gaussian kernels. The idea is to construct multiple subject/session-dependent RKHS and to perform classification with LP-SVMs. To evaluate the performance of the algorithm, we applied it to oxy-hemoglobin data sets acquired from eight sessions and seven subjects as they performed two different mental tasks. Results show that our classifiers maintain good performance when applied to random patterns across varying sessions/subjects.

Classification of frontal cortex haemodynamic responses during cognitive tasks using wavelet transforms and machine learning algorithms.

Recent advances in neuroimaging demonstrate the potential of functional near-infrared spectroscopy (fNIRS) for use in brain-computer interfaces (BCIs). fNIRS uses light in the near-infrared range to measure brain surface haemoglobin concentrations and thus determine human neural activity. Our primary goal in this study is to analyse brain haemodynamic responses for application in a BCI. Specifically, we develop an efficient signal processing algorithm to extract important mental-task-relevant neural features and obtain the best possible classification performance. We recorded brain haemodynamic responses due to frontal cortex brain activity from nine subjects using a 19-channel fNIRS system. Our algorithm is based on continuous wavelet transforms (CWTs) for multi-scale decomposition and a soft thresholding algorithm for de-noising. We adopted three machine learning algorithms and compared their performance. Good performance can be achieved by using the de-noised wavelet coefficients as input features for the classifier. Moreover, the classifier performance varied depending on the type of mother wavelet used for wavelet decomposition. Our quantitative results showed that CWTs can be used efficiently to extract important brain haemodynamic features at multiple frequencies if an appropriate mother wavelet function is chosen. The best classification results were obtained by a specific combination of input feature type and classifier.

Decision support algorithm for diagnosis of ADHD using electroencephalograms.

Attention deficit hyperactivity disorder is a complex brain disorder which is usually difficult to diagnose. As a result many literature reports about the increasing rate of misdiagnosis of ADHD disorder with other types of brain disorder. There is also a risk of normal children to be associated with ADHD if practical diagnostic criteria are not supported. To this end we propose a decision support system in diagnosing of ADHD disorder through brain electroencephalographic signals. Subjects of 10 children participated in this study, 7 of them were diagnosed with ADHD disorder and remaining 3 children are normal group. Our main goal of this sthudy is to present a supporting diagnostic tool that uses signal processing for feature selection and machine learning algorithms for diagnosis.Particularly, for a feature selection we propose information theoretic which is based on entropy and mutual information measure. We propose a maximal discrepancy criterion for selecting distinct (most distinguishing) features of two groups as well as a semi-supervised formulation for efficiently updating the training set. Further, support vector machine classifier trained and tested for identification of robust marker of EEG patterns for accurate diagnosis of ADHD group. We demonstrate that the applicability of the proposed approach provides higher accuracy in diagnostic process of ADHD disorder than the few currently available methods.

A new QRS detection method using wavelets and artificial neural networks.

We present a new method for detection and classification of QRS complexes in ECG signals using continuous wavelets and neural networks. Our wavelet method consists of four wavelet basis functions that are suitable in detection of QRS complexes within different QRS morphologies in the signal and thresholding technique for denoising and feature extraction. The results demonstrate that the proposed method is not only efficient for normal ECG signal analysis but also for various types of arrhythmic cardiac signals embedded in noise. For the classification stage, a feedforward neural network was trained with standard backpropagation algorithm. The classifier input features consisted of compact wavelet coefficients of QRS complexes that resulted in higher classification rates. We demonstrate the efficiency of our method with the average accuracy 97.2% in classification of normal and abnormal QRS complexes.

Seizure detection in temporal lobe epileptic EEGs using the best basis wavelet functions.

In this paper, we propose a novel method using best basis wavelet functions and double thresholding that are well suited for detecting and localization of important epileptic events from noisy recorded seizure EEG signals. Our technique is based on dyadic wavelet decomposition and is mainly concerned detection of single epileptic transients within the observation sequence, such as ictal and interictal epochs of EEG. In our experiment we use temporal lobe epileptic data recorded during 84 h from four patients diagnosed with epilepsy. We have achieved promising results that demonstrate efficiency and simplicity that can be used in clinical studies as an automatic decision support tool. Thus reduce the physician's workload and provide accurate diagnosis of epileptic seizures.