基于自适应动态特征库的P300诱发脑电单次提取算法研究

目的 诱发电位的单次提取技术一直是脑电信息处理领域的难题之一,为进一步提高单次提取算法的时间准确性和特征精度,针对体感诱发脑电数据信噪比低、试次间参数变化大的特点,研究诱发脑电参数单次提取新算法,保留试次间诱发脑电的动态特性,并提高估计准确率.方法 基于小波滤波和多元线性分析技术,加入自适应动态特征库并由此提出的诱发脑电P300参数单次提取新方法.随机选取4组小波滤波(WF)后诱发脑电数据,分别叠加平均后进行主成分分析(PCA)组成特征库.单次提取时,针对每试次数据从特征库中选择与当次诱发脑电信号相关系数最高的成分作为自变量开展多元线性回归分析,由回归分析结构重构出单次诱发电位信号并自动提取潜伏期和幅值等关键特征.结果 与专家判定的基准数值相比,新算法预测的P300成分潜伏期与幅值参数更准确,两者的平均差值分别为(11.16±8.60) ms和(1.40±1.34)μV;与常用的叠加平均法结果亦更为接近,平均差值分别为(23.26±25.76) ms和(2.52±2.50) μV,新算法相比传统多元线性回归分析算法具有显著优势.结论 将动态更新的诱发脑电数据主成分样本库应用于小波滤波与多元线性回归方法,能有效保留单次诱发脑电数据中的动态特征,从而提升参数估计的准确率.     

Non-invasive estimation of myocardial infarction by means of a heart-model-based imaging approach: A simulation study

In the study, a new myocardial infarction (MI) estimation method was developed for estimating Ml in the three-dimensional myocardium by means of a heart-model-based inverse approach. The site and size of Ml are estimated from body surface electrocardiograms by minimising multiple objective functions of the measured body surface potential maps (BSPMs) and the heart-model-generated BSPMs. Computer simulations were conducted to evaluate the performance of the developed method, using a single-site Ml and dual-site Ml protocols. The simulation results show that, for the single-site Ml, the averaged spatial distance (SD) between the weighting centres of the ‘true’ and estimated Mls, and the averaged relative error (RE) between the numbers of the ‘true’ and estimated infarcted units are 3.0±0.6/3.6±0.6 mm and 0.11±0.02/0.14±0.02, respectively, when 5μV/10μV Gaussian white noise was added to the body surface potentials. For the dual-site Ml, the averaged SD between the weighting centres of the ‘true’ and estimated Mls, and the averaged RE between the numbers of the ‘trus’ and estimated infarcted units are 3.8±0.7/3.9±0.7 mm and 0.12±0.02/0.14±0.03, respectively, when 5μV/10μV Gaussian white noise was added to the body surface potentials. The simulation results suggest the feasibility of applying the heart-model-based imaging approach to the estimation of myocardial infarction from body surface potentials.     

Taking into account latency, amplitude, and morphology: improved estimation of single-trial ERPs by wavelet filtering and multiple linear regression

Across-trial averaging is a widely used approach to enhance the signal-to-noise ratio (SNR) of event-related potentials (ERPs). However, across-trial variability of ERP latency and amplitude may contain physiologically relevant information that is lost by across-trial averaging. Hence, we aimed to develop a novel method that uses 1) wavelet filtering (WF) to enhance the SNR of ERPs and 2) a multiple linear regression with a dispersion term (MLR(d)) that takes into account shape distortions to estimate the single-trial latency and amplitude of ERP peaks. Using simulated ERP data sets containing different levels of noise, we provide evidence that, compared with other approaches, the proposed WF+MLR(d) method yields the most accurate estimate of single-trial ERP features. When applied to a real laser-evoked potential data set, the WF+MLR(d) approach provides reliable estimation of single-trial latency, amplitude, and morphology of ERPs and thereby allows performing meaningful correlations at single-trial level. We obtained three main findings. First, WF significantly enhances the SNR of single-trial ERPs. Second, MLR(d) effectively captures and measures the variability in the morphology of single-trial ERPs, thus providing an accurate and unbiased estimate of their peak latency and amplitude. Third, intensity of pain perception significantly correlates with the single-trial estimates of N2 and P2 amplitude. These results indicate that WF+MLR(d) can be used to explore the dynamics between different ERP features, behavioral variables, and other neuroimaging measures of brain activity, thus providing new insights into the functional significance of the different brain processes underlying the brain responses to sensory stimuli.     

Non-constraining sleep/wake monitoring system using bed actigraphy

This paper introduces a new method, bed actigraphy (BACT) for user-friendly sleep-wake monitoring. BACT provides a non-intrusive acquisition of activity data, and in particular does not require that sensors be attached to the subject’s body. The system consists of four load-sensing cells supporting the bed, an A/D converter, and a microcontroller with appropriate software. The performance of BACT was compared to that of standard polysomnography (PSG) recordings and wrist-worn actigraphy (ACT). Ten normal volunteers underwent overnight PSG recordings and were examined simultaneously with BACT and ACT. An automatic scoring algorithm scored each 30-s epoch of the BACT recordings for either ‘Wake’ or ‘Sleep.’ A sleep specialist manually scored the PSG recordings, and the results were divided into ‘Wake’ and ‘Sleep’ categories. The three methods showed a significant correlation when compared with in the contingency test. The mean epoch-by-epoch agreements between the BACT and PSG, ACT and PSG, and BACT and ACT recordings were 95.2, 92.9, and 94.3%, respectively. The mean absolute differences in sleep percentage (SP) between them were 1.8 ± 0.82, 3.4 ± 1.45, and 1.9 ± 1.16 %, respectively. BACT differentiation of the ‘Wake’ and ‘Sleep’ stages proved to be sufficiently robust, and its results were comparable to PSG analysis. This finding supports the experimental and clinical value of bed-activity monitoring during sleep.     

Observing social gestures: an fMRI study

We investigated the effects of social content of gestures on brain activation patterns. We used a 3 × 3 × 3 factorial design in an event-related functional magnetic resonance imaging experiment with participants observing gestures varied by type (fascist salute, wave, or arm lift), number of images shown at a time, and face frequency. We sought to determine whether increasing the social content of the gesture spreads activation from traditional sensorimotor regions engaged in mirror neuron activity to prefrontal regions concerned with social behavior. Results indicate that viewing a highly provocative gesture (fascist salute) compared to a less provocative but still socially meaningful gesture (wave) reveals activation in prefrontal and limbic areas. In addition, as expected there was more inferior frontal gyrus activation when participants observed a greater number of gesturing actors. Additionally, the psychological characteristics of shame and defeat affected activation in the inferior parietal lobe, which is part of the mirror neuron system, for the fascist salute compared to the wave contrast. We conclude that observing social gestures activates social- and emotion-processing areas of the brain, and the activation varies depending on the observer’s psychological characteristics.     

Age Differences in Genetic and Environmental Variations in Stress-Coping During Adulthood: A Study of Female Twins

The way people cope with stressors of day to day living has an important influence on health. The aim of the present study was to explore whether genetic and environmental variations in stress-coping differ over time during adulthood. The brief COPE was mailed to a large sample of the UK female twins (N = 4,736) having a wide range of age (20–87 years). Factor analyses of the items of the brief COPE yielded three coping scales: ‘Problem-Solving’, ‘Support Seeking’, and ‘Avoidance’. Monozygotic and dizygotic twin correlations tended to become lower with age for all three scales, suggesting that unique environmental factors may become more important with age during adulthood. Model-fitting results showed that relative influences of unique environmental factors increased from 60 % at age 20 years to 74% at age 87 years for ‘Problem-Solving’ and 56 % at age 20 years to 76% at age 87 years for ‘Avoidance’. During the same age period, genetic factors decreased from 40 to 26 % for ‘Problem-Solving’ and from 44 to 24 % for ‘Avoidance’. For ‘Seeking Support’, the magnitude of genetic and unique environmental factors was not significantly different across the adulthood. For all three scales, shared environmental effects were negligible. Overall, our findings implicate that the effects of environment that stem from idiosyncratic experience of stressful life events accumulate and become increasingly important in adulthood.     

Correspondence of Visual Evoked Potentials with FMRI Signals in Human Visual Cortex

Functional magnetic resonance imaging (FMRI) and event related potentials (ERPs) are tools that can be used to image brain activity with relatively good spatial and temporal resolution, respectively. Utilizing both of these methods is therefore desirable in neuroimaging studies to explore the spatio-temporal characteristics of brain function. While several studies have investigated the relationship between EEG and positive (+) BOLD (activation), little is known about the relationship between EEG signals and negative (−) BOLD (deactivation) responses. In this study, we used a visual stimuli designed to shift cortical activity from anterior to posterior regions of the visual cortex. Using EEG and FMRI, we investigated how shifts in +BOLD and −BOLD location were correlated to shifts in the N75 and P100 visual evoked potential (VEP) dipolar sources. The results show that the N75 dipole along with +BOLD, were indeed shifted from posterior to anterior regions of the visual cortex. The P100 VEP component, along with the −BOLD were not shifted to the same extent, indicating that N75 is better correlated to +BOLD than to −BOLD. These findings indicate how different components of the EEG signal are related to the positive and negative BOLD responses, which may aid in interpreting the relationship between visually evoked EEG and FMRI signals. An erratum to this article can be found at     

Gait control system for functional electrical stimulation using neural networks

In functional electrical stimulation (FES) systems for restoring walking in spinal cord injured (SCI) individuals, hand switches are the preferred method for controlling stimulation timing. Through practice the user becomes an ‘expert’ in determining when stimulation should be applied. Neural networks have been used to ‘clone’ this expertise but these applications have used small numbers of sensors, and their structure has used a binary output, giving rise to possible controller oscillations. It was proposed that a threelayer structure neural network with continuous function, using a larger number of sensors, including ‘virtual’ sensors, can be used to ‘clone’ this expertise to produce good controllers. Using a sensor set of ten force sensors and another of 13 ‘virtual’ kinematic sensors, a good FES control system was constructed using a three-layer neural network with five hidden nodes. The sensor set comprising three sensors showed the best performance. The accuracy of the optimum three-sensor set for the force sensors and the virtual kinematic sensors was 90% and 93%, respectively, compared with 81% and 77% for a heel switch. With 32 synchronised sensors, binary neural networks and continuous neural networks were constructed and compared. The networks using continuous function had significantly fewer oscillations. Continuous neural networks offer the ability to generate good FES controllers.     

Effects of single-trial averaging on spatial extent of brain activation detected by fMRI are subject and task dependent

Aim

The effects of single-trial averaging on the spatial extent of event-related fMRI activation may vary between subjects and tasks. The purpose of this study was to evaluate this variability using a visual task and a word generation task.

Patients, materials, and methods

Five Chinese right-handed male volunteers participated in the experiment. Experiments were conducted using a 1.5 T clinical MRI scanner with a T2*-weighted single-shot gradient-echo EPI sequence. Each task contained 150 trials that were separated into 5 runs. For each voxel, time courses averaged across different numbers of randomly selected trials, were obtained. They were applied for determining the voxels with significant activations, using a students’ t-test (p<0.001, uncorrected).

Results

Consistent with previous findings, the number of the activated voxels increased monotonically with the number of trials combined. The ascending rate and the maximum number of the activated voxels were different, however, between tasks and among subjects.

Conclusions

The effects of single-trial averaging were found to vary significantly between tasks and subjects. Therefore, we strongly advise to carefully consider such variability when using the spatial extent of activation as a measure in a group or a task comparison.     

Bilateral parietal cortex function during motor imagery

The aim of this study was to investigate the involvement of the parietal cortex during motor imagery (MI). In experiment one, participants imagined a sequence of upper limb movements during FMRI scanning. Statistical parametric mapping revealed a network of activation consistent with previous MI research, including activation in right and left inferior and superior parietal cortex. In experiment two, participants imagined a sequence of upper limb movements while real or sham single-pulse TMS was delivered over the scalp area corresponding to each individual’s left or right superior parietal cortex. At the end of each trial, participants moved their upper limbs to the position that would result from executing the sequence of movements. TMS degraded accuracy of MI compared to sham stimulation, and both accuracy and confidence decreased with real and sham stimulation later in the MI sequence. The effects of TMS were similar when delivered to either hemisphere. The results of this study provide evidence of the crucial role of SPL in MI, and may have implications for rehabilitation from brain injury.     

Evidence of break-points in breathing pattern at the gas-exchange thresholds during incremental cycling in young, healthy subjects

The present study investigated whether ‘break-points’ in breathing pattern correspond to the first ( G_\textEX1 G_{{{\text{EX}}_{1} }} ) and second gas-exchange thresholds ( G_{\textEX 2} G_{{{\text{EX}}_{ 2} }} ) during incremental cycling. We used polynomial spline smoothing to detect accelerations and decelerations in pulmonary gas-exchange data, which provided an objective means of ‘break-point’ detection without assumption of the number and shape of said ‘break-points’. Twenty-eight recreational cyclists completed the study, with five individuals excluded from analyses due to low signal-to-noise ratios and/or high risk of ‘pseudo-threshold’ detection. In the remaining participants (n = 23), two separate and distinct accelerations in respiratory frequency (f _R) during incremental work were observed, both of which demonstrated trivial biases and reasonably small ±95% limits of agreement (LOA) for the G_\textEX1 G_{{{\text{EX}}_{1} }} (0.2 ± 3.0 ml O₂ kg⁻¹ min⁻¹) and G_{\textEX 2} G_{{{\text{EX}}_{ 2} }} (0.0 ± 2.4 ml O₂ kg⁻¹ min⁻¹), respectively. A plateau in tidal volume (V _T) data near the G_\textEX1 G_{{{\text{EX}}_{1} }} was identified in only 14 individuals, and yielded the most unsatisfactory mean bias ±LOA of all comparisons made (−0.4 ± 5.3 ml O₂ kg⁻¹ min⁻¹). Conversely, 18 individuals displayed V _T-plateau in close proximity to the G_{\textEX 2} G_{{{\text{EX}}_{ 2} }} evidenced by a mean bias ± LOA of 0.1 ± 3.1 ml O₂ kg⁻¹ min⁻¹. Our findings suggest that both accelerations in f _R correspond to the gas-exchange thresholds, and a plateau (or decline) in V _T at the G_{\textEX 2} G_{{{\text{EX}}_{ 2} }} is a common (but not universal) feature of the breathing pattern response to incremental cycling.     

The Brain Computer Interface Using Flash Visual Evoked Potential and Independent Component Analysis

In this study flashing stimuli, such as digits or letters, are displayed on a LCD screen to induce flash visual evoked potentials (FVEPs). The aim of the proposed interface is to generate desired strings while one stares at target stimulus one after one. To effectively extract visually-induced neural activities with superior signal-to-noise ratio, independent component analysis (ICA) is employed to decompose the measured EEG and task-related components are subsequently selected for data reconstruction. In addition, all the flickering sequences are designed to be mutually independent in order to remove the contamination induced by surrounding non-target stimuli from the ICA-recovered signals. Since FVEPs are time-locked and phase-locked to flash onsets of gazed stimulus, segmented epochs from ICA-recovered signals based on flash onsets of gazed stimulus will be sharpen after averaging whereas those based on flash onsets of non-gazed stimuli will be suppressed after averaging. The stimulus inducing the largest averaged FVEPs is identified as the gazed target and corresponding digit or letter is sent out. Five subjects were asked to gaze at each stimulus. The mean detection accuracy resulted from averaging 15 epochs was 99.7%. Another experiment was to generate a specified string ‘0287513694E’. The mean accuracy and information transfer rates were 83% and 23.06 bits/min, respectively.     

Representing actions through their sound

Since the discovery of ‘mirror neurons’ in the monkey premotor and parietal cortex, an increasing body of evidence in animals and humans alike has supported the notion of the inextricable link between action execution and action perception. Although research originally focused on the relationship between performed and viewed actions, more recent studies highlight the importance of representing the actions of others through audition. In the first part of this article, we discuss animal studies, which provide direct evidence that action is inherently linked to multi-sensory cues, as well as the studies carried out on healthy subjects by using state-of-the-art cognitive neuroscience techniques such as functional magnetic resonance imaging (fMRI), event-related potentials (ERP), magnetoencephalography (MEG), and transcranial magnetic stimulation (TMS). In the second section, we review the lesion analysis studies in brain-damaged patients demonstrating the link between ‘resonant’ fronto-parieto-temporal networks and the ability to represent an action by hearing its sound. Moreover, we examine the evidence in favour of somatotopy as a possible representational rule underlying the auditory mapping of actions and consider the links between language and audio-motor action mapping. We conclude with a discussion of some outstanding questions for future research on the link between actions and the sounds they produce.     

独立分量分析的拟牛顿迭代算法及其在视觉诱发电位特征提取中的应用

少次叠加平均处理后的视觉诱发电位（VEP）中仍含有一定的背景噪声．对其进行进一步的提取与处理有重要的实用价值。独立分量分析（ICA）能够从混合信号中分离出最独立的成分，有效抑制噪声。本文尝试采用ICA的拟牛顿迭代算法进行VEP特征提取，介绍该方法的原理、实验和结果，并与采用牛顿迭代准则的快速独立孕量分析（Fast ICA）算法进行了比较。结果表明，基于拟牛顿法的ICA可以有效增强信号，从少次叠加平均的结果中提取出易于辨识的VEP的P300信号，具有较高的应用价值。     

Non-linear function model of voice pitch dependency on physical and mental load

The present work describes that under increasing physical load the voice fundamental frequency (voice pitch) remains on a given level as long as the physical load is well tolerated by the subject, whereas heart rate and blood pressure continuously increase during increasing physical load. This voice pitch level was compared to voice pitch levels under mental load. Using a word recognition system, 11 well trained, young male subjects had to solve 2 moderate mental load tasks. Before, during and after each task, there were structured relaxation phases. The physical load protocol was a standard bicycle stress test. In each protocol phase the subjects had to count from 1 to 10 in order to provide a standardized speech sample. Heart rate and blood pressure were recorded in all phases. Voice frequency was at average 106 ± 5.2 Hz in the relaxation phases (‘rest level’) and was increased under mental load (115.9 ± 5.7 Hz, Pillais-P = 0.037). During physical stress testing, voice pitch remained unchanged (‘tolerated load level’) between 100 and 200 W (117.4 ± 4.8 Hz) and increased shortly before physical exhaustion (‘exhaustion level’, 275–350 W, 142.9 ± 5.6 Hz, Pillais-P = 0.007). In contrast, heart rate and blood pressure increased continuously with the physical load. Three voice pitch levels could be verified also individually for each subject. For the practical monitoring of emotional stress the individual anchor frequencies for these levels must be assessed. These data indicate that the relationship between both types of load and voice pitch is non-linear with multiple plateaus and transition functions between them.     

基于共空间模式和支持向量机算法的食物偏好脑电分类研究

研究不同食物偏好图片刺激所引起的脑电信号差异,提出一套用于分类识别的方法。18名受试者对5类食物进行评分后,参与oddball事件相关电位(ERP)实验。根据受试者评分决定其喜爱程度高和喜爱程度低的食物,并用相应图片刺激,采集受试者面对两类刺激时所诱发的ERP信号,分析两类信号在ERP成分上是否存在显著差异。利用共空间模式算法(CSP)进行特征提取,随后使用支持向量机(SVM)进行单试次脑电分类识别,并使用留一法进行交叉验证。结果表明,喜爱程度高和低食物图片刺激诱发的P3幅值在统计学上存在显著差异(P<0.05),振幅分别为(16.74±5.85)μV和(14.67±4.37)μV,相关分析结果表明食物评分与脑电P3幅值存在正相关。分类识别上,在使用4次单试次ERP信号下,CSP+SVM分类正确率可达93.16%±8.93%,取得了较好的分类识别结果。实验结果表明,脑电信号可作为一种新的食物偏好分析工具,CSP+SVM可对食物偏好脑电进行较好地识别,可为食物评价和厌食症辅助治疗等提供新的解决思路。     

Manipulating time-to-plan alters patterns of brain activation during the Fitts’ task

Fitts’ law predicts that there is an essential trade-off between speed and accuracy during movement. Past investigations of Fitts’ law have not characterized whether advance planning of upcoming fast and accurate movements impacts either behavior or patterns of brain activation. With an event-related functional magnetic resonance imaging (fMRI) paradigm, we investigated the neural correlates of advance planning and movement difficulty of rapid, goal-directed aimed movements using a discrete version of the classic Fitts’ task. Our behavioral data revealed strong differences in response time, initial movement velocity, and end-point accuracy based on manipulation of both time to plan movements and response difficulty. We discovered a modulation of the neural network associated with executing the Fitts’ task that was dependent on the availability of time to plan the upcoming movement and motor difficulty. Specifically, when time to plan for the upcoming movement was available, medial frontal gyrus (BA 10), pre-SMA (BA 6), putamen and cerebellar lobule VI were uniquely active to plan movements. Further, their activation correlated with behavioral measures of movement. In contrast, manipulating movement difficulty invoked a different pattern of brain activations in regions that are known to participate in motor control, including supplementary motor area (BA 6), sensory motor cortex (BA 4, 3, 2) and putamen. Our finding that medial frontal gyrus (BA 10) was important for discrete, fast and accurate movements expands the known role of this brain region, which in the past has been identified as a cognitive processing system supporting stimulus-oriented attending. We now extend this conceptualization to include motor functions such as those employed for processing for rapid, goal-directed aimed movements.

Potential roles of force cues in human stance control

Human stance is inherently unstable. A small deviation from upright body orientation is enough to yield a gravitational component in the ankle joint torque, which tends to accelerate the body further away from upright (‘gravitational torque’; magnitude is related to body-space lean angle). Therefore, to maintain a given body lean position, a corresponding compensatory torque must be generated. It is well known that subjects use kinematic sensory information on body-space lean from the vestibular system for this purpose. Less is known about kinetic cues from force/torque receptors. Previous work indicated that they are involved in compensating external contact forces such as a pull or push having impact on the body. In this study, we hypothesized that they play, in addition, a role when the vestibular estimate of the gravitational torque becomes erroneous. Reasons may be sudden changes in body mass, for instance by a load, or an impairment of the vestibular system. To test this hypothesis, we mimicked load effects on the gravitational torque in normal subjects and in patients with chronic bilateral vestibular loss (VL) with eyes closed. We added/subtracted extra torque to the gravitational torque by applying an external contact force (via cable winches and a body harness). The extra torque was referenced to body-space lean, using different proportionality factors. We investigated how it affected body-space lean responses that we evoked using sinusoidal tilts of the support surface (motion platform) with different amplitudes and frequencies (normals ±1°, ±2°, and ±4° at 0.05, 0.1, 0.2, and 0.4 Hz; patients ±1° and ±2° at 0.05 and 0.1 Hz). We found that added/subtracted extra torque scales the lean response in a systematic way, leading to increase/decrease in lean excursion. Expressing the responses in terms of gain and phase curves, we compared the experimental findings to predictions obtained from a recently published sensory feedback model. For the trials in which the extra torque tended to endanger stance control, predictions in normals were better when the model included force cues than without these cues. This supports our notion that force cues provide an automatic ‘gravitational load compensation’ upon changes in body mass in normals. The findings in the patients support our notion that the presumed force cue mechanism provides furthermore vestibular loss compensation. Patients showed a body-space stabilization that cannot be explained by ankle angle proprioception, but must involve graviception, most likely by force cues. Our findings suggest that force cues contribute considerably to the redundancy and robustness of the human stance control system.     

Electrotactile stimuli delivered across fingertips inducing the Cutaneous Rabbit Effect

Previous studies have been unable to induce the Cutaneous Rabbit Effect (CRE) when the most likely perceived location of the illusory stimulus is on a non-continuous skin area. To determine whether the CRE could be elicited when each of the delivered stimuli were on non-continuous skin areas, we developed a new electrotactile stimulation paradigm attempting to induce the CRE across the fingertips. Though our stimulation paradigm differed from classic reduced CRE paradigms through the use of electrotactile stimuli, focusing the subject attention to a ‘likely’ illusory site, and the inclusion of a fourth stimulation site (two stimuli after the illusory stimulus), these factors were not the cause of the illusory effect we observed. Experiments conducted on the forearm validated that our paradigm elicited similar results to those reported in previous CRE studies that used either 3-stimulation-point mechanical or electrotactile stimuli with subject attention focused on the ‘likely’ illusory site. Across the fingertips, we observed an increase in stimulus mislocalization onto the middle fingertip, the ‘likely’ perceived location of the illusory stimuli, under Illusory Rabbit Trains compared to the Motion Bias Trains. Because the Motion Bias Trains should not induce a perceived location shift of the illusory stimulus but stimulates the adjacent digits in a similar way to the Illusory Rabbit Trains, differences observed between their mislocalization rates between these trains indicate that the CRE can be induced across the fingertips. These results provide the first evidence that the CRE can ‘jump’ when the stimuli occur across non-continuous skin areas.     

针对快速序列视觉呈现脑电信号的时空混合特征提取方法

快速序列视觉呈现脑机接口（RSVP-BCI）是目前基于人脑对目标进行早期发现任务中最为常用的技术,该技术能够获取人脑对环境的快速感知。脑电信号（EEG）具有非平稳和信噪比较低的特点,因此在单次试验中准确解码大脑活动较为困难。为解决RSVP-BCI技术中单次试验分类准确率不高的问题,本文提出一种基于时空域混合特征提取的新方法。该方法充分考虑大脑活动的时空模式,分别在时域和空域采用主成分分析（PCA）和共空间模式（CSP）对EEG信号进行特征提取,构成时空混合CSP-PCA(STHCP)方法,通过时域、空域两次特征提取最大化目标类与非目标类之间的判别距离,有效地降低特征维数。STHCP的单试次解码曲线下面积（AUC）较三种基准算法[空间加权费希尔（Fisher）线性判决-PCA(SWFP)、CSP及PCA算法]分别提高了17.9%、22.2%及29.2%,为利用RSVP-BCI技术进行快速高效的目标检测提供了新方法。