Adaptive filtering of evoked potentials with radial-basis-function neural network prefilter

MI-based BCI with accurate real-time three-class classification processing and light control application

The use of brain-computer interfaces (BCIs) to control intelligent devices is a current and future research direction. However, the challenges of low accuracy of real-time recognition and the need for multiple electroencephalographic channels are yet to be overcome. While a number of research teams have proposed many ways to improve offline classification accuracy, the potential problems in real-time experiments are often overlooked. In this study, we proposed a label-based channel diversion preprocessing to solve the problem of low real-time classification accuracy. The Tikhonov regularised common spatial-pattern algorithm (TRCSP) and one vs rest support vector machine (OVR-SVM) were used for feature extraction and pattern classification. High accuracy was achieved in real-time three-class classification using only three channels (average real-time accuracy of 87.46%, with a maximum of 90.33%). In addition, the stability and reliability of the system were verified through lighting control experiments in a real environment. Using the autonomy of MI and real-time feedback of light brightness, we have built a fully autonomous interactive system. The improvement in the real-time classification accuracy in this study is of great significance to the industrialisation of BCI.

Cost-Sensitive Radial Basis Function Neural Network Classifier for Software Defect Prediction

Effective prediction of software modules, those that are prone to defects, will enable software developers to achieve efficient allocation of resources and to concentrate on quality assurance activities. The process of software development life cycle basically includes design, analysis, implementation, testing, and release phases. Generally, software testing is a critical task in the software development process wherein it is to save time and budget by detecting defects at the earliest and deliver a product without defects to the customers. This testing phase should be carefully operated in an effective manner to release a defect-free (bug-free) software product to the customers. In order to improve the software testing process, fault prediction methods identify the software parts that are more noted to be defect-prone. This paper proposes a prediction approach based on conventional radial basis function neural network (RBFNN) and the novel adaptive dimensional biogeography based optimization (ADBBO) model. The developed ADBBO based RBFNN model is tested with five publicly available datasets from the NASA data program repository. The computed results prove the effectiveness of the proposed ADBBO-RBFNN classifier approach with respect to the considered metrics in comparison with that of the early predictors available in the literature for the same datasets.

Altered intensity coding in the salicylate-overdose animal model of tinnitus

Tinnitus is one of the leading disorders of hearing with no effective cure as its pathophysiological mechanisms remain unclear. While the sensitivity to sound is well-known to be affected, exactly how intensity coding per se is altered remains unclear. To address this issue, we used a salicylate-overdose animal model of tinnitus to measure auditory cortical evoked potentials at various stimulus levels, and analyzed on single-trial basis the response strength and its variance for the computation of the lower bound of Fisher information. Based on Fisher information profiles, we compared the precision or efficiency of intensity coding before and after salicylate-treatment. We found that after salicylate treatment, intensity coding was unexpectedly improved, rather than impaired. Also, the improvement varied in a sound-dependent way. The observed changes are likely due to some central compensatory mechanisms that are activated during tinnitus to bring out the full capacity of intensity coding which is expressed only in part under normal conditions.

Additive and multiplicative noise reduction by back propagation neural network

A novel filter is proposed by applying back propagation neural network (BPNN) ensemble where the noisy signal and the reference one are the same. The neural network(NN) ensemble filter not only well reduces additive and multiplicative white noise inside signals, but also preserves signals' characteristics. It is proved that while power of noise is larger, the reduction of noise using NN ensemble filter is better than the improved ¿ nonlinear filter and single NN filter, and compared with the improved ¿ nonlinear filter, degradation o the capability for reduction of noise by NN ensemble due to the increase of noise power is much suppressed. Furthermore, it is presented of the relationship between noise reduction and bandwidth of noises. The performance of the NN ensemble filter is demonstrated in computer simulations and actual electroencephalogram (EEG) signals processing.

EEG features extraction for motor imagery

Motor imagery is the mental simulation of a motor act that includes preparation for movement, passive observations of action and mental operations of motor representations implicitly or explicitly. Motor imagery as preparation for immediate movement likely involves the motor executive brain regions. Implicit mental operations of motor representations are considered to underlie cognitive functions. Another problem concerning neuro-imaging studies on motor imagery is that the performance of imagination is very difficult to control. The ability of an individual to control its EEG may enable him to communicate without being able to control their voluntary muscles. Communication based on EEG signals does not require neuromuscular control and the individuals who have neuromuscular disorders and who may have no more control over any of their conventional communication abilities may still be able to communicate through a direct brain-computer interface. A brain-computer interface replaces the use of nerves and muscles and the movements they produce with electrophysiological signals and is coupled with the hardware and software that translate those signals into physical actions. One of the most important components of a brain-computer interface is the EEG feature extraction procedure. This paper presents an approach that uses self-organizing fuzzy neural network based time series prediction that performs EEG feature extraction in the time domain only. EEG is recorded from two electrodes placed on the scalp over the motor cortex. EEG signals from each electrode are predicted by a single fuzzy neural network. Features derived from the mean squared error of the predictions and from the mean squared of the predicted signals are extracted from EEG data by means of a sliding window. The architecture of the two auto-organizing fuzzy neural networks is a network with multi inputs and single output.

Development and evaluation of the sparse decomposition method with mixed over-complete dictionary for evoked potential estimation

A new method is developed to decompose a physiological signal into a summation of transient and oscillatory components, referred to as mixed over-complete dictionary based sparse component decomposition algorithm (MOSCA). Based on the characteristics of the transient evoked potential (EP) and the background noise, the mixed dictionary is constructed with an over-complete wavelet dictionary and an over-complete discrete cosine (DC) function dictionary, and the signal is separated by learning in this mixed dictionary with a matching pursuit (MP) algorithm. MOSCA is designed specifically for the separation of a desired transient EP from the existing spontaneous EEG or other background noise. The method was evaluated with several simulation tests in which EPs or simulated EPs were deeply masked in different strong noise backgrounds, and the recovered signal is similar to the original assumed EP with a high and stable correlation coefficient (CC). The method was then applied to estimate event related potential (ERP) in the classical oddball experiment, and the results confirmed that the trial number for a reliable ERP estimation might be greatly reduced by MOSCA.

Single-trial extraction of cognitive evoked potentials by combination of third-order correlation and wavelet denoising

The application of a recently proposed denoising implementation for obtaining cognitive evoked potentials (CEPs) at the single-trial level is shown. The aim of this investigation is to develop the technique of extracting CEPs by combining both the third-order correlation and the wavelet denoising methods. First, the noisy CEPs was passed through a finite impulse response filter whose impulse response is matched with the shape of the noise-free signal. It was shown that it is possible to estimate the filter impulse response on basis of a select third-order correlation slice (TOCS) of the input noisy CEPs. Second, the output from the third-order correlation filter is decomposed with bi-orthogonal splines at 5 levels. The CEPs is reconstructed by wavelet final approximation a<inf>5</inf>. We study its performance in simulated data as well as in cognitive evoked potentials of normal rat and Alzheimer's disease (AD) model rat. For the simulated data, the method gives a significantly better reconstruction of the single-trial cognitive evoked potentials responses in comparison with the simulated data. Moreover, with this approach we obtain a significantly better estimation of the amplitudes and latencies of the simulated CEPs. For the real data, the method clearly improves the visualization of single-trial CEPs. This allows the calculation of better averages as well as the study of systematic or unsystematic variations between trials.

Chaotic Dynamics in Tracing BAEP and its Application on Investigating Brainstem Malfunction

The purpose of this study is to investigate the chaotic dynamics of tracing (dynamic) brainstem auditory evoked potential (BAEP), then to investigate the brainstem malfunction via dynamic characteristics of tracing BAEP. The radial basis function neural network (RBFNN) technique in combination with the moving window average (MWA) method was employed in order to extract the tracing BAEP from strong noise background. Data of noisy BAEP were collected from the Infantile spasm (IS) group with brainstem malfunction and the tester group respectively. The chaotic dynamics of tracing BAEP were analyzed using phase projection and correlation dimension techniques. The results of this study have demonstrated: (1) there is a much stronger determination in BAEP than in noisy BAEP shown by more deterministic phase projections and lower D<inf>2</inf>in BAEP; (2) trajectories of BAEP never repeat and the value of correlation dimension is fractal; (3) the phase projection of BAEP for brainstem malfunction group shows more chaotic and has higher D<inf>2</inf>than those for the tester group. The conclusions of this investigation suggest that BAEP is chaotic not deterministic and there is rich dynamics in BAEP; tracing BAEP is much more useful than noisy BAEP in describing brainstem malfunction because BAEP shows determination corresponding to brainstem function; BAEP in brainstem malfunction has more non-order dynamics than that in the tester.

Single trial somatosensory evoked potential extraction with ARX filtering for a combined spinal cord intraoperative neuromonitoring technique

Background

When spinal cord functional integrity is at risk during surgery, intraoperative neuromonitoring is recommended. Tibial Single Trial Somatosensory Evoked Potentials (SEPs) and H-reflex are here used in a combined neuromonitoring method: both signals monitor the spinal cord status, though involving different nervous pathways. However, SEPs express a trial-to-trial variability that is difficult to track because of the intrinsic low signal-to-noise ratio. For this reason single trial techniques are needed to extract SEPs from the background EEG.

Methods

The analysis is performed off line on data recorded in eight scoliosis surgery sessions during which the spinal cord was simultaneously monitored through classical SEPs and H-reflex responses elicited by the same tibial nerve electrical stimulation. The single trial extraction of SEPs from the background EEG is here performed through AutoRegressive filter with eXogenous input (ARX). The electroencephalographic recording can be modeled as the sum of the background EEG, which can be described as an autoregressive process not related to the stimulus, and the evoked potential (EP), which can be viewed as a filtered version of a reference signal related to the stimulus. The choice of the filter optimal orders is based on the Akaike Information Criterion (AIC). The reference signal used as exogenous input in the ARX model is a weighted average of the previous SEPs trials with exponential forgetting behavior.

Results

The moving average exponentially weighted, used as reference signal for the ARX model, shows a better sensibility than the standard moving average in tracking SEPs fast inter-trial changes. The ability to promptly detect changes allows highlighting relations between waveform changes and surgical maneuvers. It also allows a comparative study with H-reflex trends: in particular, the two signals show different fall and recovery dynamics following stressful conditions for the spinal cord.

Conclusion

The ARX filter showed good performances in single trial SEP extraction, enhancing the available information concerning the current spinal cord status. Moreover, the comparison between SEPs and H-reflex showed that the two signals are affected by the same surgical maneuvers, even if they monitor the spinal cord through anatomically different pathways.

Adaptive signal enhancement of somatosensory evoked potential for spinal cord compression detection: an experimental study

The objective of this study was to assess the efficacy of adaptive signal enhancement (ASE) as a means of indicating intraoperative spinal cord impingement. ASE technique was used to determine the changes in the somatosensory evoked potential (SEP) elicited from eighteen rats with varying levels of spinal cord compression. ASE technique was found to be able to effectively extract SEP signals for the detection of spinal cord injury. Furthermore, while the traditional ensemble averaging (EA) technique requires more than 500 trials for meaningful signal processing in severe noisy SEP recordings, the ASE method required only 50 trials to provide similar information. Because of its fast and reliable SEP detection, the ASE method is ideal for spinal cord monitoring in the clinical setting.

Real-Time Data-Reusing Adaptive Learning of a Radial Basis Function Network for Tracking Evoked Potentials

Tracking variations in both the latency and amplitude of evoked potential (EP) is important in quantifying properties of the nervous system. Adaptive filtering is a powerful tool for tracking such variations. In this paper, a data-reusing non-linear adaptive filtering method, based on a radial basis function network (RBFN), is implemented to estimate EP. The RBFN consists of an input layer of source nodes, a single hidden layer of non-linear processing units and an output layer of linear weights. It has built-in nonlinear activation functions that allow learning of function mappings. Moreover, it produces satisfactory estimates of signals against a background noise without a priori knowledge of the signal, provided that the signal and noise are independent. In clinical situations where EP responses change rapidly, the convergence rate of the algorithm becomes a critical factor. A carefully designed data-reusing RBFN can accelerate the convergence rate markedly and, thus, enhance its performance. Both theoretical analysis and simulation results support the improved performance of our new algorithm.

Adaptive filtering of evoked potentials using higher-order adaptive signal enhancer with genetic-type variable step-size prefilter

An adaptive signal enhancer based on third-order statistics with a genetic-type, variable step-size prefilter is introduced to recover evoked potentials (EPs). EPs are usually embedded in the ongoing electroencephalogram with a very low signal-to-noise ratio (SNR). As a higher-order statistics technique has a natural tolerance to Gaussian noise, it is applicable for filtering EPs. An adaptive signal enhancer based on third-order statistics was used as the major filter in this study. However, the efficiency of the adaptive signal enhancer was reduced when the total power of uncorrelated noises was large. To improve the performance for EPs under poor SNR, a low-noise signal is required. Therefore a prefilter with a genetic-type, variable step-size algorithm was employed to enhance the SNR of the signal in this study. The fundamental idea of a genetic-type, variable step-size algorithm is that its step-sizes are regularly readjusted to optimum. Therefore this algorithm can be used as a prefilter with different noise levels. Experimental results showed that, for filtering EPs, the proposed scheme is superior to the adaptive signal enhancer with a normalised least mean square algorithm.

Real-time ocular artifact suppression using recurrent neural network for electro-encephalogram based brain-computer interface

The paper presents an adaptive noise canceller (ANC) filter using an artificial neural network for real-time removal of electro-oculogram (EOG) interference from electro-encephalogram (EEG) signals. Conventional ANC filters are based on linear models of interference. Such linear models provide poorer prediction for biomedical signals. In this work, a recurrent neural network was employed for modelling the interference signals. The eye movement and eye blink artifacts were recorded by the placing of an electrode on the forehead above the left eye and an electrode on the left temple. The reference signal was then generated by the data collected from the forehead electrode being added to data recorded from the temple electrode. The reference signal was also contaminated by the EEG. To reduce the EEG interference, the reference signal was first low-pass filtered by a moving averaged filter and then applied to the ANC. Matlab Simulink was used for real-time data acquisition, filtering and ocular artifact suppression. Simulation results show the validity and effectiveness of the technique with different signal-to-noise ratios (SNRs) of the primary signal. On average, a significant improvement in SNR up to 27 dB was achieved with the recurrent neural network. The results from real data demonstrate that the proposed scheme removes ocular artifacts from contaminated EEG signals and is suitable for real-time and short-time EEG recordings.

Multi-adaptive filtering technique for surface somatosensory evoked potentials processing

Somatosensory evoked potential (SEP) testing has been widely applied to diagnosis of various neurological disorders. However, SEP recorded using surface electrodes is buried in noises, which makes the signal-to-noise ratio (SNR) very poor. Conventional averaging method usually requires up to thousands of raw SEP input trials to increase the SNR so that an identifiable waveform can be produced for latency and amplitude measurement. In this study, a multi-adaptive filtering (MAF) technique, emerging from the combination of well-developed adaptive noise canceller and adaptive signal enhancer, is introduced for fast and accurate surface SEP extraction. The MAF technique first processes the raw surface recorded SEP by the Canceller with a reference noise channel of background noise for adaptive subtraction before entering the Enhancer. The MAF was verified by filtering simulated SEP signals in which electroencephalography and Gaussian noise of different SNRs were added. It was found that the MAF could effectively suppress the noise and enhance the SEP components such that the SNR of the SEP is improved. Results showed that MAF with 50 input trials could provide similar performance in SEP detection to those extracted by the conventional averaging method with 1000 trials even at an SNR of -20 dB.

Validation of an Adaptive Signal Enhancer in Intraoperative Somatosensory Evoked Potentials Monitoring

The conventional approach of ensemble averaging in intraoperative somatosensory evoked potentials (SEP) monitoring requires more than 500 trials to extract a reliable waveform for neurologic diagnosis. Previous studies showed that an adaptive signal enhancer (ASE) could increase the signal-to-noise ratio of input signals. This study assessed the accuracy and efficiency of the ASE in the extraction of neurologic normal human and abnormal rat SEP. Cortical and subcortical SEP were taken from 16 subjects undergoing scoliosis surgery. SEP extracted by ASE were compared with those obtained with 500-trial averaging in terms of peak latency, amplitude, and waveforms using correlation coefficients. An animal study composed of 18 rats was used to test the ASE in detecting abnormal SEP changes due to spinal cord compression. The results demonstrate the accuracy of ASE by showing very high correlations between ASE-processed SEP and ensemble averaging-processed SEP in waveforms, peak latencies, and amplitudes. The results also show the efficiency of the ASE in extracting SEP waveforms from 50 input trials, which provided waveforms of sufficiently high quality and latency/amplitude measurements equivalent to those obtained in 500 trials of conventional ensemble averaging. Because of its fast extraction ability, adaptive signal enhancement could be an appropriate alternative to conventional ensemble averaging in intraoperative spinal cord monitoring.

Free radical EPR spectroscopy analysis using blind source separation

In this paper, we propose a novel approach for electron paramagnetic resonance (EPR) mixture spectra analysis based on blind source separation (BSS) technique. EPR spectrum of a free radical is often superimposed by overlapping spectra of other species. It is important and challenging to accurately identify and quantify the 'pure' spectra from such mixtures. In this study, an automated BSS method implementing independent component analysis is used to extract the components from mixed EPR spectra that contain overlapping components of different paramagnetic centers. To apply this method, there is no requirement to know the component spectra or the number of components in advance. The method is applied to analyze free radical EPR spectra which are collected from standard chemical system, cultured cell suspense, and ex vivo rat kidneys by spin trapping EPR technique. Results show that the BSS method proposed here is capable of identifying the component EPR spectra from mixtures with unknown compositions. The BSS technique can offer powerful aids in resolving spectral overlapping problems in general EPR spectroscopy analysis.