A Generalizable Brain-Computer Interface (BCI) Using Machine Learning for Feature Discovery

This work describes a generalized method for classifying motor-related neural signals for a brain-computer interface (BCI), based on a stochastic machine learning method. The method differs from the various feature extraction and selection techniques employed in many other BCI systems. The classifier does not use extensive a-priori information, resulting in reduced reliance on highly specific domain knowledge. Instead of pre-defining features, the time-domain signal is input to a population of multi-layer perceptrons (MLPs) in order to perform a stochastic search for the best structure. The results showed that the average performance of the new algorithm outperformed other published methods using the Berlin BCI IV (2008) competition dataset and was comparable to the best results in the Berlin BCI II (2002–3) competition dataset. The new method was also applied to electroencephalography (EEG) data recorded from five subjects undertaking a hand squeeze task and demonstrated high levels of accuracy with a mean classification accuracy of 78.9% after five-fold cross-validation. Our new approach has been shown to give accurate results across different motor tasks and signal types as well as between subjects.     

Proprioceptive Feedback and Brain Computer Interface (BCI) Based Neuroprostheses

Brain computer interface (BCI) technology has been proposed for motor neurorehabilitation, motor replacement and assistive technologies. It is an open question whether proprioceptive feedback affects the regulation of brain oscillations and therefore BCI control. We developed a BCI coupled on-line with a robotic hand exoskeleton for flexing and extending the fingers. 24 healthy participants performed five different tasks of closing and opening the hand: (1) motor imagery of the hand movement without any overt movement and without feedback, (2) motor imagery with movement as online feedback (participants see and feel their hand, with the exoskeleton moving according to their brain signals, (3) passive (the orthosis passively opens and closes the hand without imagery) and (4) active (overt) movement of the hand and rest. Performance was defined as the difference in power of the sensorimotor rhythm during motor task and rest and calculated offline for different tasks. Participants were divided in three groups depending on the feedback receiving during task 2 (the other tasks were the same for all participants). Group 1 (n = 9) received contingent positive feedback (participants'' sensorimotor rhythm (SMR) desynchronization was directly linked to hand orthosis movements), group 2 (n = 8) contingent “negative” feedback (participants'' sensorimotor rhythm synchronization was directly linked to hand orthosis movements) and group 3 (n = 7) sham feedback (no link between brain oscillations and orthosis movements). We observed that proprioceptive feedback (feeling and seeing hand movements) improved BCI performance significantly. Furthermore, in the contingent positive group only a significant motor learning effect was observed enhancing SMR desynchronization during motor imagery without feedback in time. Furthermore, we observed a significantly stronger SMR desynchronization in the contingent positive group compared to the other groups during active and passive movements. To summarize, we demonstrated that the use of contingent positive proprioceptive feedback BCI enhanced SMR desynchronization during motor tasks.     

Unsupervised Learning and Adaptation in a Model of Adult Neurogenesis

Adult neurogenesis has long been documented in the vertebrate brain and recently even in humans. Although it has been conjectured for many years that its functional role is related to the renewing of memories, no clear mechanism as to how this can be achieved has been proposed. Using the mammalian olfactory bulb as a paradigm, we present a scheme in which incorporation of new neurons proceeds at a constant rate, while their survival is activity-dependent and thus contingent on new neurons establishing suitable connections. We show that a simple mathematical model following these rules organizes its activity so as to maximize the difference between its responses and can adapt to changing environmental conditions in unsupervised fashion, in agreement with current neurophysiological data.     

Simultaneous Detection of P300 and Steady-State Visually Evoked Potentials for Hybrid Brain-Computer Interface

Objective

We study the feasibility of a hybrid Brain-Computer Interface (BCI) combining simultaneous visual oddball and Steady-State Visually Evoked Potential (SSVEP) paradigms, where both types of stimuli are superimposed on a computer screen. Potentially, such a combination could result in a system being able to operate faster than a purely P300-based BCI and encode more targets than a purely SSVEP-based BCI.

Approach

We analyse the interactions between the brain responses of the two paradigms, and assess the possibility to detect simultaneously the brain activity evoked by both paradigms, in a series of 3 experiments where EEG data are analysed offline.

Main Results

Despite differences in the shape of the P300 response between pure oddball and hybrid condition, we observe that the classification accuracy of this P300 response is not affected by the SSVEP stimulation. We do not observe either any effect of the oddball stimulation on the power of the SSVEP response in the frequency of stimulation. Finally results from the last experiment show the possibility of detecting both types of brain responses simultaneously and suggest not only the feasibility of such hybrid BCI but also a gain over pure oddball- and pure SSVEP-based BCIs in terms of communication rate.     

脑机接口专利情报研究

脑机接口是在无外周神经系统和肌肉组织参与的条件下,通过计算机等电子设备输出控制信号,进而与外界环境进行交流的全新通讯和控制技术。它的发展依赖于神经科学、心理学、工程学、康复医学和计算机等学科专家间的密切合作,具有非常重要的学术价值、科学意义和广阔的应用前景,是当今世界研究的热点。本文利用新一代专利分析平台和工具Innography,结合专利情报分析理论,在对世界范围内脑机接口技术的专利进行申请趋势分析、区域分析、IPC分析、专利权人分析、诉讼专利和核心专利分析以及重点技术文本聚类分析的基础上,了解国内该技术发展态势,并尝试为脑机接口领域的发展提供有用的竞争情报参考。     

Towards an Optimization of Stimulus Parameters for Brain-Computer Interfaces Based on Steady State Visual Evoked Potentials

Efforts to construct an effective brain-computer interface (BCI) system based on Steady State Visual Evoked Potentials (SSVEP) commonly focus on sophisticated mathematical methods for data analysis. The role of different stimulus features in evoking strong SSVEP is less often considered and the knowledge on the optimal stimulus properties is still fragmentary. The goal of this study was to provide insight into the influence of stimulus characteristics on the magnitude of SSVEP response. Five stimuli parameters were tested: size, distance, colour, shape, and presence of a fixation point in the middle of each flickering field. The stimuli were presented on four squares on LCD screen, with each square highlighted by LEDs flickering with different frequencies. Brighter colours and larger dimensions of flickering fields resulted in a significantly stronger SSVEP response. The distance between stimulation fields and the presence or absence of the fixation point had no significant effect on the response. Contrary to a popular belief, these results suggest that absence of the fixation point does not reduce the magnitude of SSVEP response. However, some parameters of the stimuli such as colour and the size of the flickering field play an important role in evoking SSVEP response, which indicates that stimuli rendering is an important factor in building effective SSVEP based BCI systems.     

Decoding Sensorimotor Rhythms during Robotic-Assisted Treadmill Walking for Brain Computer Interface (BCI) Applications

Locomotor malfunction represents a major problem in some neurological disorders like stroke and spinal cord injury. Robot-assisted walking devices have been used during rehabilitation of patients with these ailments for regaining and improving walking ability. Previous studies showed the advantage of brain-computer interface (BCI) based robot-assisted training combined with physical therapy in the rehabilitation of the upper limb after stroke. Therefore, stroke patients with walking disorders might also benefit from using BCI robot-assisted training protocols. In order to develop such BCI, it is necessary to evaluate the feasibility to decode walking intention from cortical patterns during robot-assisted gait training. Spectral patterns in the electroencephalogram (EEG) related to robot-assisted active and passive walking were investigated in 10 healthy volunteers (mean age 32.3±10.8, six female) and in three acute stroke patients (all male, mean age 46.7±16.9, Berg Balance Scale 20±12.8). A logistic regression classifier was used to distinguish walking from baseline in these spectral EEG patterns. Mean classification accuracies of 94.0±5.4% and 93.1±7.9%, respectively, were reached when active and passive walking were compared against baseline. The classification performance between passive and active walking was 83.4±7.4%. A classification accuracy of 89.9±5.7% was achieved in the stroke patients when comparing walking and baseline. Furthermore, in the healthy volunteers modulation of low gamma activity in central midline areas was found to be associated with the gait cycle phases, but not in the stroke patients. Our results demonstrate the feasibility of BCI-based robotic-assisted training devices for gait rehabilitation.     

Prediction of Auditory and Visual P300 Brain-Computer Interface Aptitude

Objective

Brain-computer interfaces (BCIs) provide a non-muscular communication channel for patients with late-stage motoneuron disease (e.g., amyotrophic lateral sclerosis (ALS)) or otherwise motor impaired people and are also used for motor rehabilitation in chronic stroke. Differences in the ability to use a BCI vary from person to person and from session to session. A reliable predictor of aptitude would allow for the selection of suitable BCI paradigms. For this reason, we investigated whether P300 BCI aptitude could be predicted from a short experiment with a standard auditory oddball.

Methods

Forty healthy participants performed an electroencephalography (EEG) based visual and auditory P300-BCI spelling task in a single session. In addition, prior to each session an auditory oddball was presented. Features extracted from the auditory oddball were analyzed with respect to predictive power for BCI aptitude.

Results

Correlation between auditory oddball response and P300 BCI accuracy revealed a strong relationship between accuracy and N2 amplitude and the amplitude of a late ERP component between 400 and 600 ms. Interestingly, the P3 amplitude of the auditory oddball response was not correlated with accuracy.

Conclusions

Event-related potentials recorded during a standard auditory oddball session moderately predict aptitude in an audiory and highly in a visual P300 BCI. The predictor will allow for faster paradigm selection.

Significance

Our method will reduce strain on patients because unsuccessful training may be avoided, provided the results can be generalized to the patient population.     

Investigating the Effect of Flickering Frequency Pair and Mother Wavelet Selection in Steady-State Visually-Evoked Potentials on Two-Command Brain-Computer Interfaces

IntroductionSteady-state visually evoked potentials (SSVEPs) have become popular in brain-computer interface (BCI) applications in addition to many other applications on clinical neuroscience (neurodegenerative disorders, schizophrenia, epilepsy, etc.), cognitive (visual attention, working memory, brain rhythms, etc.), and use of engineering researches. Among available methods to measure brain activities, SSVEPs have advantages like higher information transfer rate, simplicity in structure, and short training time. SSVEP-based BCIs use flickering stimuli at different frequencies to discriminate distinct commands in real life. Some features are extracted from the SSVEP signals before these commands are classified. The wavelet transform (WT) has attracted researchers among feature extraction methods since it utilizes the non-stationary signals well. In the WT, a sample function (named mother wavelet) represents the SSVEP signal in both time and frequency domains. Unfortunately, there is no universal mother wavelet function that fits all the signals. Therefore, choosing an appropriate mother wavelet function may be a challenge in WT-related studies. Although there are such studies in three- and seven-command SSVEP-based studies, there is no study for two-command systems in our knowledge.Materials and MethodsIn this study, two user commands flickered at the combinations of seven different frequencies were tested to determine which frequency pairs give the highest performance. For this purpose, three well-known wavelet features (energy, entropy, and variance) were calculated for each of derived EEG frequency bands from the discrete WT coefficients of SSVEP signals. The WT was repeated for six different mother wavelet functions (Haar, Db4, Sym4, Coif1, Bior3.5, and Rbior2.8). Then, four feature sets (every three features, and all together) were applied to seven commonly-used machine learning algorithms (Decision Tree, Discriminant Analysis, Logistic Regression, Naive Bayes, Support Vector Machines, Nearest Neighbors, and Ensemble Classifiers).Results and DiscussionWe achieved 100% accuracies among these 3,528 runs (7 classifiers x 4 feature sets x 6 mother wavelets x 21 flickering frequency pairs) using the mother wavelet function of Haar and the Ensemble Learner classifier. The highest classifier performances are 100% when two commands have the flickering frequency pairs of (6.0 and 10 Hz), (6.5 and 8.2 Hz), or (6.5 and 10.0 Hz).ConclusionWe obtained three main outcomes from this study. First, the most representative mother wavelet function was Haar, while the worst one was Symlet 4. Second, the Ensemble Learner classifier gave the maximum classifier performance in a two-command SSVEP-based BCI system. Besides, two user commands from SSVEP should be one of the frequency pairs of (6.0 and 10.0 Hz), (6.5 and 8.2 Hz), and (6.5 and 10.0 Hz) to achieve the maximum accuracy.     

Automated Glioblastoma Segmentation Based on a Multiparametric Structured Unsupervised Classification

Automatic brain tumour segmentation has become a key component for the future of brain tumour treatment. Currently, most of brain tumour segmentation approaches arise from the supervised learning standpoint, which requires a labelled training dataset from which to infer the models of the classes. The performance of these models is directly determined by the size and quality of the training corpus, whose retrieval becomes a tedious and time-consuming task. On the other hand, unsupervised approaches avoid these limitations but often do not reach comparable results than the supervised methods. In this sense, we propose an automated unsupervised method for brain tumour segmentation based on anatomical Magnetic Resonance (MR) images. Four unsupervised classification algorithms, grouped by their structured or non-structured condition, were evaluated within our pipeline. Considering the non-structured algorithms, we evaluated K-means, Fuzzy K-means and Gaussian Mixture Model (GMM), whereas as structured classification algorithms we evaluated Gaussian Hidden Markov Random Field (GHMRF). An automated postprocess based on a statistical approach supported by tissue probability maps is proposed to automatically identify the tumour classes after the segmentations. We evaluated our brain tumour segmentation method with the public BRAin Tumor Segmentation (BRATS) 2013 Test and Leaderboard datasets. Our approach based on the GMM model improves the results obtained by most of the supervised methods evaluated with the Leaderboard set and reaches the second position in the ranking. Our variant based on the GHMRF achieves the first position in the Test ranking of the unsupervised approaches and the seventh position in the general Test ranking, which confirms the method as a viable alternative for brain tumour segmentation.     

Glaucomatous Patterns in Frequency Doubling Technology (FDT) Perimetry Data Identified by Unsupervised Machine Learning Classifiers

Purpose

The variational Bayesian independent component analysis-mixture model (VIM), an unsupervised machine-learning classifier, was used to automatically separate Matrix Frequency Doubling Technology (FDT) perimetry data into clusters of healthy and glaucomatous eyes, and to identify axes representing statistically independent patterns of defect in the glaucoma clusters.

Methods

FDT measurements were obtained from 1,190 eyes with normal FDT results and 786 eyes with abnormal FDT results from the UCSD-based Diagnostic Innovations in Glaucoma Study (DIGS) and African Descent and Glaucoma Evaluation Study (ADAGES). For all eyes, VIM input was 52 threshold test points from the 24-2 test pattern, plus age.

Results

FDT mean deviation was −1.00 dB (S.D. = 2.80 dB) and −5.57 dB (S.D. = 5.09 dB) in FDT-normal eyes and FDT-abnormal eyes, respectively (p<0.001). VIM identified meaningful clusters of FDT data and positioned a set of statistically independent axes through the mean of each cluster. The optimal VIM model separated the FDT fields into 3 clusters. Cluster N contained primarily normal fields (1109/1190, specificity 93.1%) and clusters G₁ and G₂ combined, contained primarily abnormal fields (651/786, sensitivity 82.8%). For clusters G₁ and G₂ the optimal number of axes were 2 and 5, respectively. Patterns automatically generated along axes within the glaucoma clusters were similar to those known to be indicative of glaucoma. Fields located farther from the normal mean on each glaucoma axis showed increasing field defect severity.

Conclusions

VIM successfully separated FDT fields from healthy and glaucoma eyes without a priori information about class membership, and identified familiar glaucomatous patterns of loss.     

Studies on Oxidation-Reduction Potentials (ORP) of Microbial Cultures

At maximum production of l-glutamic acid, the oxidation-reduction potential of the culture broth in l-glutamic acid fermentation showed a stable value of 9.0 to 9.6 as rH value. When biotin concentration in the medium was high (40γ/liter), the production of l-glutamic acid decreased, and the rH was 8.0 and it was out of accordance with that of the control (biotin-poor; 2γ/liter). Under “less-aerobic” conditions, its rH rose to 10.4.

From these results, it was concluded that the rH during maximum production of l-glutamic acid showed a stable value affected actively by the redox system, l-glutamic acid/α-ketoglutaric acid and      

Studies on Oxidation-Reduction Potentials (ORP) of Microbial Cultures

Aerobacter aerogenes No. 505 isolated from soil by Uyeda produced l-valine extracellularly by an aerobic shaking culture. Under anaerobic conditions the production of this amino acid was inhibited while lactic acid as well as a small amount of alanine were produced. The changes in ORP during the incubation under both conditions were investigated. When l-valine was the main product under aerobic conditions the ORP showed a constant value (rH 8.0) from 16 to 40 hr after inoculation. But when lactic acid was the main product and alanine was produced as the only amino acid under anaerobic conditions, the ORP drifted to rH 0 (zero). The phenomenon of the conversion of fermentation was shown clearly by the ORP of the culture broth.

The endpotentiai of lactic acid fermentation by Rhizopus G-36 was rH 13 to 14 when measured in the presence of trace amounts of redox dye mixtures. Without dyes, the rH was 18 to 22 and this fungal culture was slower in reaching endpotentials than bacterial cultures. It was postulated that the amount of redox substances exhibiting electromotive activity was not sufficient in this culture.

rH value 13 to 14 was not obtained under such conditions that lactic acid was not produced; that is in a medium with higher concentration of the nitrogen source in the presence of Fe²⁺ and Zn²⁺, or in a medium containing acetate in place of glucose as the carbon source.

Mycelia of Rhizopus G-36 after 36 hr-culture produced lactic acid even in the absence of oxygen. But unexpectedly, the ORP under anaerobic secondary culture was exactly the same as that in the aerobic shaking culture (rH 13.2).

A method for homogenization of the culture without secondary oxidation was improved. The ORP of anaerobically homogenized cultures was rH 11, and was thought to be due to the activities of all redox systems in the mycelium.

The respiration system of this strain was switched from cytochrome system to flavin system at the point of change in KGN-sensitivity. The ORP of this strain may be influenced by respiration through the flavin system.     

A P300-based Brain-Computer Interface with Stimuli on Moving Objects: Four-Session Single-Trial and Triple-Trial Tests with a Game-Like Task Design

Brain-computer interfaces (BCIs) are tools for controlling computers and other devices without using muscular activity, employing user-controlled variations in signals recorded from the user’s brain. One of the most efficient noninvasive BCIs is based on the P300 wave of the brain’s response to stimuli and is therefore referred to as the P300 BCI. Many modifications of this BCI have been proposed to further improve the BCI’s characteristics or to better adapt the BCI to various applications. However, in the original P300 BCI and in all of its modifications, the spatial positions of stimuli were fixed relative to each other, which can impose constraints on designing applications controlled by this BCI. We designed and tested a P300 BCI with stimuli presented on objects that were freely moving on a screen at a speed of 5.4°/s. Healthy participants practiced a game-like task with this BCI in either single-trial or triple-trial mode within four sessions. At each step, the participants were required to select one of nine moving objects. The mean online accuracy of BCI-based selection was 81% in the triple-trial mode and 65% in the single-trial mode. A relatively high P300 amplitude was observed in response to targets in most participants. Self-rated interest in the task was high and stable over the four sessions (the medians in the 1st/4th sessions were 79/84% and 76/71% in the groups practicing in the single-trial and triple-trial modes, respectively). We conclude that the movement of stimulus positions relative to each other may not prevent the efficient use of the P300 BCI by people controlling their gaze, e.g., in robotic devices and in video games.     

Adaptation and gain normalization: a comment on Ullman & Schechtman (1982)

A well-known process for adaptation and gain normalization is compared with the process described by S. Ullman and G. Schechtman (Proc. R. Soc. Lond. B 216, 299-313 (1982)). A neural interpretation of this process in terms of transmitter gating, slow accumulation, and release is described. Applications to a wide variety of problems in perception, cognition, and motivated behaviour can be made by embedding the gating process into opponent processes, notably shunting on-centre off-surround networks, to form a network module called a gated dipole field.     

The Efficiency of the Brain-Computer Interfaces Based on Motor Imagery with Tactile and Visual Feedback

In this study we compared tactile and visual feedbacks for the motor imagery-based brain–computer interface (BCI) in five healthy subjects. A vertical green bar from the center of the fixing cross to the edge of the screen was used as visual feedback. Vibration motors that were placed on the forearms of the right and the left hands and on the back of the subject’s neck were used as tactile feedback. A vibration signal was used to confirm the correct classification of the EEG patterns of the motor imagery of right and left hand movements and the rest task. The accuracy of recognition in the classification of the three states (right hand movement, left hand movement, and rest) in the BCI without feedback exceeded the random level (33% for the three states) for all the subjects and was rather high (67.8% ± 13.4% (mean ± standard deviation)). Including the visual and tactile feedback in the BCI did not significantly change the mean accuracy of recognition of mental states for all the subjects (70.5% ± 14.8% for the visual feedback and 65.9% ± 12.4% for the tactile feedback). The analysis of the dynamics of the movement imagery skill in BCI users with the tactile and visual feedback showed no significant differences between these types of feedback. Thus, it has been found that the tactile feedback can be used in the motor imagery-based BCI instead of the commonly used visual feedback, which greatly expands the possibilities of the practical application of the BCI.     

Adaptation to the complexity of a task (a P300 study)

In numerous studies the P300 component of the event-related brain potential (ERP) has been shown to occur in connection with stimulus evaluation processes. 10 healthy right-handed volunteers (3 women, 7 men) aged from 25 to 30 years (mean age 27.8 years) participated in the experiments. One of 5 equiprobably occurring two-letter strings appeared on the screen always at the same central location. The strings informed the subjects about the difficulty of subsequently presented mental arithmetic tasks. After the letter strings vanished from the screen the subjects were to press the space-bar whereby a mental arithmetic task was presented corresponding in difficulty to the preceding message. The EEG was recorded by means of Ag/AgCl electrodes from frontal (F zeta), central (C zeta) and parietal (P zeta) midline electrodes referenced to linked earlobes. EEG and EOG were sampled 1200 ms, starting 200 ms prior to string onset. P300 peak latencies, peak amplitudes and areas in the time range 300 to 900 ms were measured in ERPs averaged selectively for the 5 strings. The main finding was that the P300 amplitude in ERPs to the 5 different strings varied in a U-shaped trend as a function of announced task difficulty. This result gives further evidence that the P300 amplitude reflects distance between incoming information and current adaptation level at the inferred internal dimension, i.e. task difficulty in this experiment.     

以病例为基础的讨论式教学模式在临床教学中的应用

目的:探讨以病例为基础讨论式学习(case-based discussion and learning,CBDL)在临床教学中的应用效果。方法:将40位临床医学本科专业学生随机分为两组,其中20人采用CBDL教学,20人采用传统教学,并且通过理论基础考试和临床技能考试的形式评价学生的学习效果,通过问卷的形式调查学生对教学方法的满意度。结果:CBDL教学组的理论考试、临床技能考试及总成绩得分均较传统教学组明显提高(P0.01),CBDL教学组学生对教学的满意度分别为:教学方法 100%,教学效果90%,小组功能95%,教师评价95%。结论:CBDL教学法较传统教学方法更能提高学生的基础和临床能力,为临床工作中独立解决问题打下了基础。     

PBL结合LBL教学法在老年医学教学中的应用

目的:通过研究PBL结合LBL教学法在老年医学教学中的应用,观察与传统的LBL教学方法比较,PBL结合LBL教学法对学生考核成绩提高的作用,并探讨PBL结合LBL教学法加强对学生自主学习、综合实践、团队协作等各方面能力的提升。方法:选择2014年9月-2015年5月第三军医大学临床医学系2013级学生145名,随机分为实验组(PBL结合LBL教学法,75名)和对照组(传统的LBL教学法,70名)作为研究对象,通过考核与问卷调查评估结果。结果:以冠心病相关知识作为考核内容,与对照组比较,实验组考核成绩高于对照组(91.50±4.36 vs.83.24±4.12,P0.01);对实验组学生75名发放调查问卷,87%以上学生认为该教学法对提高自身学习能力、激发学习兴趣、增强表达能力等方面有较大帮助。结论:PBL结合LBL的教学法有助于提高老年医学教学效果。