Single-trial Classification Accuracy Research Articles

Towards single-trial classification of invasively recorded auditory evoked potentials in cochlear implant users

Objective. One promising approach towards further improving cochlear implants (CI) is to use brain signals controlling the device in order to close the auditory loop. Initial electroencephalography (EEG) studies have already shown promising results. However, they are based on noninvasive measurements, whereas implanted electrodes are expected to be more convenient in terms of everyday-life usability. If additional measurement electrodes were implanted during CI surgery, then invasive recordings should be possible. Furthermore, implantation will provide better signal quality, higher robustness to artefacts, and thus enhanced classification accuracy. Approach. In an initial project, three additional epidural electrodes were temporarily implanted during the surgical procedure. After surgery, different auditory evoked potentials (AEPs) were recorded both invasively (epidural) and using surface electrodes, with invasively recorded signals demonstrated as being markedly superior. In this present analysis, cortical evoked response audiometry (CERA) signals recorded in seven patients were used for single-trial classification of sounds with different intensities. For classification purposes, we used shrinkage-regularized linear discriminant analysis (sLDA). Clinical speech perception scores were also investigated. Main results. Analysis of CERA data from different subjects showed single-trial classification accuracies of up to 99.2% for perceived vs. non-perceived sounds. Accuracies of up to 89.1% were achieved in classification of sounds perceived at different intensities. Highest classification accuracies were achieved by means of epidural recordings. Required loudness differences seemed to correspond to speech perception in noise. Significance. The proposed epidural recording approach showed good classification accuracy into sound perceived and not perceived when the best-performing electrodes were selected. Classifying different levels of sound stimulation accurately proved more challenging. At present, the methods explored in this study would not be sufficiently reliable to allow automated closed-loop control of CI parameters. However, our findings are an important initial contribution towards improving applicability of closed auditory loops and for next-generation automatic fitting approaches.

Riemannian classification of single-trial surface EEG and sources during checkerboard and navigational images in humans.

Different visual stimuli are classically used for triggering visual evoked potentials comprising well-defined components linked to the content of the displayed image. These evoked components result from the average of ongoing EEG signals in which additive and oscillatory mechanisms contribute to the component morphology. The evoked related potentials often resulted from a mixed situation (power variation and phase-locking) making basic and clinical interpretations difficult. Besides, the grand average methodology produced artificial constructs that do not reflect individual peculiarities. This motivated new approaches based on single-trial analysis as recently used in the brain-computer interface field. We hypothesize that EEG signals may include specific information about the visual features of the displayed image and that such distinctive traits can be identified by state-of-the-art classification algorithms based on Riemannian geometry. The same classification algorithms are also applied to the dipole sources estimated by sLORETA. We show that our classification pipeline can effectively discriminate between the display of different visual items (Checkerboard versus 3D navigational image) in single EEG trials throughout multiple subjects. The present methodology reaches a single-trial classification accuracy of about 84% and 93% for inter-subject and intra-subject classification respectively using surface EEG. Interestingly, we note that the classification algorithms trained on sLORETA sources estimation fail to generalize among multiple subjects (63%), which may be due to either the average head model used by sLORETA or the subsequent spatial filtering failing to extract discriminative information, but reach an intra-subject classification accuracy of 82%.

Benchmarking cEEGrid and Solid Gel-Based Electrodes to Classify Inattentional Deafness in a Flight Simulator.

Transfer from experiments in the laboratory to real-life tasks is challenging due notably to the inability to reproduce the complexity of multitasking dynamic everyday life situations in a standardized lab condition and to the bulkiness and invasiveness of recording systems preventing participants from moving freely and disturbing the environment. In this study, we used a motion flight simulator to induce inattentional deafness to auditory alarms, a cognitive difficulty arising in complex environments. In addition, we assessed the possibility of two low-density EEG systems a solid gel-based electrode Enobio (Neuroelectrics, Barcelona, Spain) and a gel-based cEEGrid (TMSi, Oldenzaal, Netherlands) to record and classify brain activity associated with inattentional deafness (misses vs. hits to odd sounds) with a small pool of expert participants. In addition to inducing inattentional deafness (missing auditory alarms) at much higher rates than with usual lab tasks (34.7% compared to the usual 5%), we observed typical inattentional deafness-related activity in the time domain but also in the frequency and time-frequency domains with both systems. Finally, a classifier based on Riemannian Geometry principles allowed us to obtain more than 70% of single-trial classification accuracy for both mobile EEG, and up to 71.5% for the cEEGrid (TMSi, Oldenzaal, Netherlands). These results open promising avenues toward detecting cognitive failures in real-life situations, such as real flight.

Prediction of Navigational Decisions in the Real-World: A Visual P300 Event-Related Potentials Brain-Computer Interface

ABSTRACT Despite the widespread availability of portable neuroimaging systems, current applications of brain-computer interfaces (BCI) have largely remained confined to laboratory and clinical settings. In order to develop BCI that will assist individuals across a wide range of everyday life behaviors, it is critical to test paradigms that have been successful in highly controlled setups within the frame of real-world environments. In the present study, a visual P300 paradigm was used to elicit neural responses reflective of decision making within the frame of real-world navigation. Participants (n = 8) were equipped with a tablet and a mobile EEG system while navigating through university corridors. Upon reaching an intersection, participants were presented with left and right arrows flashing on the tablet (and additional non-directional stimuli in the distractor condition). Neural responses elicited by the presentation of such stimuli (Event-Related Potentials) were recorded along with the participant’s navigational decisions. The participants completed two separate sessions to collect data for both the training and testing of an offline BCI system aimed at predicting their navigational decisions based on the extraction of event-related potentials features. Single-trial classification accuracy reached 59.6% (up to 72.3% when classifying group averages of six trials). Individual classification results were contrasted with single-trial analysis. Results are discussed in terms of their implications for the design of real-world BCI applications, and several recommendations to improve experimental protocols are proposed.

Typical somatomotor physiology of the hand is preserved in a patient with an amputated arm: An ECoG case study

Electrophysiological signals in the human motor system may change in different ways after deafferentation, with some studies emphasizing reorganization while others propose retained physiology. Understanding whether motor electrophysiology is retained over longer periods of time can be invaluable for patients with paralysis (e.g. ALS or brainstem stroke) when signals from sensorimotor areas may be used for communication or control over neural prosthetic devices. In addition, a maintained electrophysiology can potentially benefit the treatment of phantom limb pains through prolonged use of these signals in a brain-machine interface (BCI).Here, we were presented with the unique opportunity to investigate the physiology of the sensorimotor cortex in a patient with an amputated arm using electrocorticographic (ECoG) measurements. While implanted with an ECoG grid for clinical evaluation of electrical stimulation for phantom limb pain, the patient performed attempted finger movements with the contralateral (lost) hand and executed finger movements with the ipsilateral (healthy) hand.The electrophysiology of the sensorimotor cortex contralateral to the amputated hand remained very similar to that of hand movement in healthy people, with a spatially focused increase of high-frequency band (65–175 Hz; HFB) power over the hand region and a distributed decrease in low-frequency band (15–28 Hz; LFB) power. The representation of the three different fingers (thumb, index and little) remained intact and HFB patterns could be decoded using support vector learning at single-trial classification accuracies of >90%, based on the first 1–3 s of the HFB response. These results indicate that hand representations are largely retained in the motor cortex. The intact physiological response of the amputated hand, the high distinguishability of the fingers and fast temporal peak are encouraging for neural prosthetic devices that target the sensorimotor cortex.

Emotion-Inducing Imagery Versus Motor Imagery for a Brain-Computer Interface.

Neural correlates of intentionally induced human emotions may offer alternative imagery strategies to control brain-computer interface (BCI) applications. In this paper, a novel BCI control strategy i.e., imagining fictional or recalling mnemonic sad and happy events, emotion-inducing imagery (EII), is compared to motor imagery (MI) in a study involving multiple sessions using a two-class electroencephalogram (EEG)-based BCI paradigm with 12 participants. The BCI setup enabled online continuous visual feedback presentation in a game involving one-dimensional control of a game character. MI and EII are compared across different signal-processing frameworks which are based on neural-time-series-prediction-preprocessing (NTSPP), filter bank common spatial patterns (FBCSP) and hemispheric asymmetry (ASYM). Online single-trial classification accuracies (CA) results indicate that MI performance across all participants is 77.54% compared to EII performance of 68.78% ( ). The results show that an ensemble of the NTSPP, FBCSP and ASYM frameworks maximizes performance for EII with average CA of 71.64% across all participants. Furthermore, the participants' subjective responses indicate that they preferred MI over emotion-inducing imagery (EII) in controlling the game character, and MI was perceived to offer most control over the game character. The results suggest that EII is not a viable alternative to MI for the majority of participants in this study but may be an alternative imagery for a subset of BCI users based on acceptable EII performance (CA >70%) observed for some participants.

Monitoring Pilot's Mental Workload Using ERPs and Spectral Power with a Six-Dry-Electrode EEG System in Real Flight Conditions.

Recent technological progress has allowed the development of low-cost and highly portable brain sensors such as pre-amplified dry-electrodes to measure cognitive activity out of the laboratory. This technology opens promising perspectives to monitor the “brain at work” in complex real-life situations such as while operating aircraft. However, there is a need to benchmark these sensors in real operational conditions. We therefore designed a scenario in which twenty-two pilots equipped with a six-dry-electrode EEG system had to perform one low load and one high load traffic pattern along with a passive auditory oddball. In the low load condition, the participants were monitoring the flight handled by a flight instructor, whereas they were flying the aircraft in the high load condition. At the group level, statistical analyses disclosed higher P300 amplitude for the auditory target (Pz, P4 and Oz electrodes) along with higher alpha band power (Pz electrode), and higher theta band power (Oz electrode) in the low load condition as compared to the high load one. Single trial classification accuracy using both event-related potentials and event-related frequency features at the same time did not exceed chance level to discriminate the two load conditions. However, when considering only the frequency features computed over the continuous signal, classification accuracy reached around 70% on average. This study demonstrates the potential of dry-EEG to monitor cognition in a highly ecological and noisy environment, but also reveals that hardware improvement is still needed before it can be used for everyday flight operations.

Investigation of the mirrored-word reading paradigm for BCI implementation.

Brain-computer interface (BCI) applications such as keyboard control and vehicular navigation present significant assistive merit for disabled individuals. However, there are limitations associated with BCI paradigms which restrict a wider adoption of BCI technology. For example, rapid serial visual presentation (RSVP) paradigms can induce seizures in photosensitive epileptic subjects. This paper evaluates the novel mirrored-word reading paradigm (MWRP) for BCI implementation using an offline experimental study. The offline study obtained an average single-trial classification accuracy of 74.10%. The results also demonstrate that the use of multiple trials for classification can increase the accuracy as is common with BCIs. The developed MWRP-based BCI also utilized a low presentation frequency which averts the possibility of paradigm induced photosensitivity. However, there are multiple avenues for future work. The MWRP can be implemented in the online format for real-time device control. For example, a vehicular application platform can be used where the word orientation represents directions for travel. The MWRP can also be investigated across a wider range of stimulus presentation parameters such as timing, color and stimulus size. Such studies can be used to suggest further improvements to the paradigm which can enhance its applicability for online device control.

Hyperdimensional Computing for Blind and One-Shot Classification of EEG Error-Related Potentials

The mathematical properties of high-dimensional (HD) spaces show remarkable agreement with behaviors controlled by the brain. Computing with HD vectors, referred to as “hypervectors,” is a brain-inspired alternative to computing with numbers. HD computing is characterized by generality, scalability, robustness, and fast learning, making it a prime candidate for utilization in application domains such as brain–computer interfaces. We describe the use of HD computing to classify electroencephalography (EEG) error-related potentials for noninvasive brain–computer interfaces. Our algorithm naturally encodes neural activity recorded from 64 EEG electrodes to a single temporal–spatial hypervector without requiring any electrode selection process. This hypervector represents the event of interest, can be analyzed to identify the most discriminative electrodes, and is used for recognition of the subject’s intentions. Using the full set of training trials, HD computing achieves on average 5% higher single-trial classification accuracy compared to a conventional machine learning method on this task (74.5% vs. 69.5%) and offers further advantages: (1) Our algorithm learns fast: using only 34% of training trials it achieves an average accuracy of 70.5%, surpassing the conventional method. (2) Conventional method requires prior domain expert knowledge, or a separate process, to carefully select a subset of electrodes for a subsequent preprocessor and classifier, whereas our algorithm blindly uses all 64 electrodes, tolerates noises in data, and the resulting hypervector is intrinsically clustered into HD space; in addition, most preprocessing of the electrode signal can be eliminated while maintaining an average accuracy of 71.7%.

Separability of motor imagery of the self from interpretation of motor intentions of others at the single trial level: an EEG study

BackgroundWe aimed to investigate the separability of the neural correlates of 2 types of motor imagery, self and third person (actions owned by the participant himself vs. another individual). If possible this would allow for the development of BCI interfaces to train disorders of action and intention understanding beyond simple imitation, such as autism.MethodsWe used EEG recordings from 20 healthy participants, as well as electrocorticography (ECoG) in one, based on a virtual reality setup. To test feasibility of discrimination between each type of imagery at the single trial level, time-frequency and source analysis were performed and further assessed by data-driven statistical classification using Support Vector Machines.ResultsThe main observed differences between self-other imagery conditions in topographic maps were found in Frontal and Parieto-Occipital regions, in agreement with the presence of 2 independent non μ related contributions in the low alpha frequency range. ECOG corroborated such separability. Source analysis also showed differences near the temporo-parietal junction and single-trial average classification accuracy between both types of motor imagery was 67 ± 1%, and raised above 70% when 3 trials were used. The single-trial classification accuracy was significantly above chance level for all the participants of this study (p < 0.02).ConclusionsThe observed pattern of results show that Self and Third Person MI use distinct electrophysiological mechanisms detectable at the scalp (and ECOG) at the single trial level, with separable levels of involvement of the mirror neuron system in different regions.These observations provide a promising step to develop new BCI training/rehabilitation paradigms for patients with neurodevelopmental disorders of action understanding beyond simple imitation, such as autism, who would benefit from training and anticipation of the perceived intention of others as opposed to own intentions in social contexts.

Decoding motor responses from the EEG during altered states of consciousness induced by propofol

Objective. Patients undergoing general anesthesia may awaken and become aware of the surgical procedure. Due to neuromuscular blocking agents, patients could be conscious yet unable to move. Using brain–computer interface (BCI) technology, it may be possible to detect movement attempts from the EEG. However, it is unknown how an anesthetic influences the brain response to motor tasks. Approach. We tested the offline classification performance of a movement-based BCI in 12 healthy subjects at two effect-site concentrations of propofol. For each subject a second classifier was trained on the subject’s data obtained before sedation, then tested on the data obtained during sedation (‘transfer classification’). Main results. At concentration 0.5 μg ml−1, despite an overall propofol EEG effect, the mean single trial classification accuracy was 85% (95% CI 81%–89%), and 83% (79%–88%) for the transfer classification. At 1.0 μg ml−1, the accuracies were 81% (76%–86%), and 72% (66%–79%), respectively. At the highest propofol concentration for four subjects, unlike the remaining subjects, the movement-related brain response had been largely diminished, and the transfer classification accuracy was not significantly above chance. These subjects showed a slower and more erratic task response, indicating an altered state of consciousness distinct from that of the other subjects. Significance. The results show the potential of using a BCI to detect intra-operative awareness and justify further development of this paradigm. At the same time, the relationship between motor responses and consciousness and its clinical relevance for intraoperative awareness requires further investigation.

P300 variability during target detection in natural images: implications for single-trial classification

References The P300 is one of the most prominent and well-studied event-related potentials (ERPs) in the literature. The P300 is also a primary discriminant signal for many brain-computer interface (BCI) systems. It has been well established that the P300 amplitude and latency, along with reaction time, are directly linked to target discriminability. However, since the majority of P300 studies using an odd-ball or target-detection paradigm have a fixed or unquantified level of target discriminability, it remains unclear how the P300 is systematically modulated by stimulus properties affecting target detection. In this study we quantified visual properties of target objects within a large ensemble of natural images. Using a rapid serial visual presentation (RSVP) paradigm, we were able to systematically measure the effect of these properties on the behavioral reaction time (RT), P300, and single-trial classification accuracy. ERP & Single-trial Classification Target Size & P300 Group Effects

Spatiotemporal Representations of Rapid Visual Target Detection: A Single-Trial EEG Classification Algorithm

Brain computer interface applications, developed for both healthy and clinical populations, critically depend on decoding brain activity in single trials. The goal of the present study was to detect distinctive spatiotemporal brain patterns within a set of event related responses. We introduce a novel classification algorithm, the spatially weighted FLD-PCA (SWFP), which is based on a two-step linear classification of event-related responses, using fisher linear discriminant (FLD) classifier and principal component analysis (PCA) for dimensionality reduction. As a benchmark algorithm, we consider the hierarchical discriminant component Analysis (HDCA), introduced by Parra, etal. 2007. We also consider a modified version of the HDCA, namely the hierarchical discriminant principal component analysis algorithm (HDPCA). We compare single-trial classification accuracies of all the three algorithms, each applied to detect target images within a rapid serial visual presentation (RSVP, 10Hz) of images from five different object categories, based on single-trial brain responses. We find a systematic superiority of our classification algorithm in the tested paradigm. Additionally, HDPCA significantly increases classification accuracies compared to the HDCA. Finally, we show that presenting several repetitions of the same image exemplars improve accuracy, and thus may be important in cases where high accuracy is crucial.

Semantic Classical Conditioning and Brain-Computer Interface Control: Encoding of Affirmative and Negative Thinking

The aim of the study was to investigate conditioned electroencephalography (EEG) responses to factually correct and incorrect statements in order to enable binary communication by means of a brain-computer interface (BCI). In two experiments with healthy participants true and false statements (serving as conditioned stimuli, CSs) were paired with two different tones which served as unconditioned stimuli (USs). The features of the USs were varied and tested for their effectiveness to elicit differentiable conditioned reactions (CRs). After acquisition of the CRs, these CRs to true and false statements were classified offline using a radial basis function kernel support vector machine. A mean single-trial classification accuracy of 50.5% was achieved for differentiating conditioned “yes” versus “no” thinking and mean accuracies of 65.4% for classification of “yes” and 68.8% for “no” thinking (both relative to baseline) were found using the best US. Analysis of the area under the curve of the conditioned EEG responses revealed significant differences between conditioned “yes” and “no” answers. Even though improvements are necessary, these first results indicate that the semantic conditioning paradigm could be a useful basis for further research regarding BCI communication in patients in the complete locked-in state.

Filter Bank Common Spatial Pattern Algorithm on BCI Competition IV Datasets 2a and 2b

The Common Spatial Pattern (CSP) algorithm is an effective and popular method for classifying 2-class motor imagery electroencephalogram (EEG) data, but its effectiveness depends on the subject-specific frequency band. This paper presents the Filter Bank Common Spatial Pattern (FBCSP) algorithm to optimize the subject-specific frequency band for CSP on Datasets 2a and 2b of the Brain-Computer Interface (BCI) Competition IV. Dataset 2a comprised 4 classes of 22 channels EEG data from 9 subjects, and Dataset 2b comprised 2 classes of 3 bipolar channels EEG data from 9 subjects. Multi-class extensions to FBCSP are also presented to handle the 4-class EEG data in Dataset 2a, namely, Divide-and-Conquer (DC), Pair-Wise (PW), and One-Versus-Rest (OVR) approaches. Two feature selection algorithms are also presented to select discriminative CSP features on Dataset 2b, namely, the Mutual Information-based Best Individual Feature (MIBIF) algorithm, and the Mutual Information-based Rough Set Reduction (MIRSR) algorithm. The single-trial classification accuracies were presented using 10 × 10-fold cross-validations on the training data and session-to-session transfer on the evaluation data from both datasets. Disclosure of the test data labels after the BCI Competition IV showed that the FBCSP algorithm performed relatively the best among the other submitted algorithms and yielded a mean kappa value of 0.569 and 0.600 across all subjects in Datasets 2a and 2b respectively.

A new (semantic) reflexive brain–computer interface: In search for a suitable classifier

The goal of the current study is to find a suitable classifier for electroencephalogram (EEG) data derived from a new learning paradigm which aims at communication in paralysis. A reflexive semantic classical (Pavlovian) conditioning paradigm is explored as an alternative to the operant learning paradigms, currently used in most brain–computer interfaces (BCIs). Comparable with a lie-detection experiment, subjects are presented with true and false statements. The EEG activity following true and false statements was classified with the aim to separate covert ‘yes’ from covert ‘no’ responses.Four classification algorithms are compared for classifying off-line data collected from a group of 14 healthy participants: (i) stepwise linear discriminant analysis (SWLDA), (ii) shrinkage linear discriminant analysis (SLDA), (iii) linear support vector machine (LIN-SVM) and (iv) radial basis function kernel support vector machine (RBF-SVM).The results indicate that all classifiers perform at chance level when separating conditioned ‘yes’ from conditioned ‘no’ responses. However, single conditioned reactions could be successfully classified on a single-trial basis (single conditioned reaction against a baseline interval). All of the four investigated classification methods achieve comparable performance, however results with RBF-SVM show the highest single-trial classification accuracy of 68.8%. The results suggest that the proposed paradigm may allow affirmative and negative (disapproving negative) communication in a BCI experiment.

Spatial filtering and selection of optimized components in four class motor imagery EEG data using independent components analysis

Three independent components analysis (ICA) algorithms (Infomax, FastICA and SOBI) have been compared with other preprocessing methods in order to find out whether and to which extent spatial filtering of EEG data can improve single trial classification accuracy. As reference methods, common spatial patterns (CSP) (a supervised method, whereas all ICA algorithms are unsupervised), bipolar derivations and the original raw monopolar data were used. In addition to only performing ICA, the number of components was reduced with PCA before calculating a spatial filter for Infomax and FastICA. The multichannel data (22 channels) of eight subjects, consisting of two sessions recorded on different days, was analyzed. The task was to perform motor imagery of the left hand, right hand, foot or tongue, respectively, during predefined time slices (cued paradigm). For a measure of fitness, classification accuracies for both cross-validated results using data from just one session as well as simulated online results (representing the session-to-session transfer) were calculated. In the latter case, the spatial filters and classifiers were computed for one session and applied to the completely unseen second session. For the data analyzed in this study, Infomax outperformed the other two ICA variants by far, both in the cross-validated as well as in the simulated online case. CSP, on the other hand, yielded significantly lower classification accuracies than Infomax for the cross-validated results, whereas there is no statistically significant difference when it comes to simulated online data. Performing PCA before ICA improved the results in the case of FastICA, whereas the classification accuracies dropped significantly for Infomax.

Spatio-temporal cortical patterns evoked in man by a discrimination task

Magnetoencephalographic (MEG) recordings were obtained from a 37 sensor array during performance of a discrimination task. Pairs of stimuli were presented to the somatosensory and auditory systems which fell into two classes. Subjects were instructed to respond to one class and not the other. A “virtual electrode” estimation procedure was applied to generate time varying waveforms from locations within the brain as if from direct electrodes. Multivariate statistical procedures were then applied to the single trial MEG and virtual electrode estimates. This produces strong statistical inferences from the data from each subject, produces functional timing and localization for each subject, and demonstrates clear differences between subjects. It is an integrated experimental/analytic paradigm which is applicable to the study of perception, cognition, affect, and a wide variety of psychopathologies. Discriminant analysis was applied to the sampled single trial MEG data to (1) determine whether information was extracted and (2) to produce a discriminant function by which the 37 channel data stream can be collapsed to a single data stream to produce functional timing information. For all 5 subjects, these results conform to the following model for the sequence of activities required to perform the discrimination task: 1. (1) Primary sensory processing. 2. (2) Discrimination between stimulus classes, this being the imperative function upon which correct task performance depends. 3. (3) Response preparation. 4. (4) Motor activation. Discriminant analysis was applied to the sampled single trial virtual electrode estimates to (1) determine whether information was extracted and (2) to produce functional localization information by identifying virtual electrode sites which are differentially activated under different experimental conditions. For sensory modality, the parietal cortex was the primary site of differential activation at 36 msec post stimulus. At 54 msec, differential activation had spread anteriorly along the cortical convexity and at 95 msec had spread to the medial cortical surface. For stimulus class, differential activation was seen in the cerebellum and in the parietal, parieto temporal, premotor and anterior temporal cortices. For response type (GO/NOGO), differential activation was seen primarily in the parietal and premotor cortices. Single trial classification accuracies from raw MEG data produced significant p -values for all subjects for sensory modality ( p < 10 −100 ) and for 4 out of 5 subjects for stimulus class and trial type ( p < 10 −4 ). Single trial classification accuracies from virtual electrode estimates produced significant p -values for all subjects for sensory modality ( p < 10 −60 ) and for 4 out of 5 subjects for stimulus class ( p < 10 −4 ) and trial type ( p < 10 −6 ). Sensory modality classification accuracy was significant for 4 out of 5 subjects for the time epoch 400 msec following stimulus presentation ( p < 10 −5 ) and for 1 subject for the time period 1.8 seconds following stimulus presentation ( p < 10 −12 ).