Aleksandra Vuckovic

Automatic recognition of alertness and drowsiness from EEG by an artificial neural network

We present a novel method for classifying alert vs drowsy states from 1 s long sequences of full spectrum EEG recordings in an arbitrary subject. This novel method uses time series of interhemispheric and intrahemispheric cross spectral densities of full spectrum EEG as the input to an artificial neural network (ANN) with two discrete outputs: drowsy and alert. The experimental data were collected from 17 subjects. Two experts in EEG interpretation visually inspected the data and provided the necessary expertise for the training of an ANN. We selected the following three ANNs as potential candidates: (1) the linear network with Widrow-Hoff (WH) algorithm; (2) the non-linear ANN with the Levenberg-Marquardt (LM) rule; and (3) the Learning Vector Quantization (LVQ) neural network. We showed that the LVQ neural network gives the best classification compared with the linear network that uses WH algorithm (the worst), and the non-linear network trained with the LM rule. Classification properties of LVQ were validated using the data recorded in 12 healthy volunteer subjects, yet whose EEG recordings have not been used for the training of the ANN. The statistics were used as a measure of potential applicability of the LVQ: the t-distribution showed that matching between the human assessment and the network output was 94.37+/-1.95%. This result suggests that the automatic recognition algorithm is applicable for distinguishing between alert and drowsy state in recordings that have not been used for the training.

Delta band contribution in cue based single trial classification of real and imaginary wrist movements

The aim of this study was to classify different movements about the right wrist. Four different movements were performed: extension, flexion, pronation and supination. Two-class single trial classification was performed on six possible combinations of two movements (extension–flexion, extension–supination, extension–pronation, flexion–supination, flexion–pronation, pronation–supination). Both real and imaginary movements were analysed. The analysis was done in the joint time–frequency domain using the Gabor transform. Feature selection was based on the Davis-Bouldin Index (DBI) and feature classification was based on Elman’s recurrent neural networks (ENN). The best classification results, near 80% true positive rate, for imaginary movements were achieved for discrimination between extension and any other type of movement. The experiments were run with 10 able-bodied subjects. For some subjects, real movement classification rates higher than 80% were achieved for any combination of movements, though not simultaneously for all six combinations of movements. For classification of the imaginary movements, the results suggest that the type of movement and frequency band play an important role. Unexpectedly, the delta band was found to carry significant class-related information.

A comparative study of three techniques for diameter selective fiber activation in the vagal nerve: anodal block, depolarizing prepulses and slowly rising pulses

The paper shows selective smaller fiber activation in the left and right vagal nerve in in vivo experiments in pigs using three different techniques: anodal block, depolarizing prepulses and slowly rising pulses. All stimulation techniques were performed with the same experimental setup. The techniques have been compared in relation to maximum achievable suppression of nerve activity, maximum required current, maximum achievable stimulation frequency and the required charge per phase. Suppression of the largest fiber activity (expressed as a percentage of the maximum response) was 0-40% for anodal block, 10-25% for depolarizing prepulses and 40-50% for slowly rising pulses (duration up to 5 ms). Incomplete suppression of activation was mainly attributed to the large size of the vagal nerve (3.0-3.5 mA) which resulted in a large difference of the excitation thresholds of nerve fibers at different distances from the electrode, as well as a relatively short duration of slowly rising pulses. The technique of anodal block required the highest currents. The techniques of slowly rising pulses and anodal block required comparable charge per phase that was larger than for the technique of depolarizing prepulses. Depolarizing prepulses were an optimal choice regarding maximum required current and charge per phase but were very sensitive to small changes of the current amplitude. The other two techniques were more robust regarding small changes of stimulation parameters. The maximum stimulation frequency, using typical values of stimulation parameters, was 105 Hz for depolarizing prepulses, 30 Hz for anodal block and 28 Hz for slowly rising pulses. Only a technique of depolarizing prepulses had a charge per phase within the safe limits. For the other two techniques it would be necessary to optimize the shape of a stimulation pulse in order to reduce the charge per phase.

Different pulse shapes to obtain small fiber selective activation by anodal blocking-a simulation study

The aim of this study was to investigate whether it is possible to reduce a charge per pulse, which is needed for selective nerve stimulation. Simulation is performed using a two-part simulation model: a volume conductor model to calculate the electrical potential distribution inside a tripolar cuff electrode and a human fiber model to simulate the fiber response to simulation. Selective stimulation is obtained by anodal block. To obtain anodal block of large fibers, long square pulses (>350 /spl mu/s) with a relatively high currents (1-2.5 mA) are usually required. These pulses might not be safe for a long-term application because of a high charge per pulse. In this study, several pulse shapes are proposed that have less charge per pulse compared with the conventional square pulse and would therefore be safer in a chronic application. Compared with the conventional square pulse, it was possible to reduce the charge with all proposed pulse shapes, but the best results are obtained with a combination of a square depolarizing pulse and a blocking pulse. The charge per pulse was up to 32% less with that pulse shape than with a square pulse. Using a hyperpolarizing anodal prepulse preceding a square pulse, it was not possible to block nerve fibers in a whole nerve bundle and to obtain reduction of a charge per phase. Reduction of the charge could be achieved only with spatially selective blocking. The charge per phase was larger for the combination of a hyperpolarizing anodal prepulse and a two-step pulse than for the two-step pulse alone.

Dynamic oscillatory signatures of central neuropathic pain in spinal cord injury

Central neuropathic pain (CNP) is believed to be accompanied by increased activation of the sensorimotor cortex. Our knowledge of this interaction is based mainly on functional magnetic resonance imaging studies, but there is little direct evidence on how these changes manifest in terms of dynamic neuronal activity. This study reports on the presence of transient electroencephalography (EEG)-based measures of brain activity during motor imagery in spinal cord–injured patients with CNP. We analyzed dynamic EEG responses during imaginary movements of arms and legs in 3 groups of 10 volunteers each, comprising able-bodied people, paraplegic patients with CNP (lower abdomen and legs), and paraplegic patients without CNP. Paraplegic patients with CNP had increased event-related desynchronization in the theta, alpha, and beta bands (16–24 Hz) during imagination of movement of both nonpainful (arms) and painful limbs (legs). Compared to patients with CNP, paraplegics with no pain showed a much reduced power in relaxed state and reduced event-related desynchronization during imagination of movement. Understanding these complex dynamic, frequency-specific activations in CNP in the absence of nociceptive stimuli could inform the design of interventional therapies for patients with CNP and possibly further understanding of the mechanisms involved. Perspective This study compares the EEG activity of spinal cord–injured patients with CNP to that of spinal cord–injured patients with no pain and also to that of able-bodied people. The study shows that the presence of CNP itself leads to frequency-specific EEG signatures that could be used to monitor CNP and inform neuromodulatory treatments of this type of pain.

Modeling EEG fractal dimension changes in wake and drowsy states in humans--a preliminary study.

Aim of this preliminary study was to examine and compare topographic distribution of Higuchis fractal dimension (FD, measure of signal complexity) of EEG signals between states of relaxed wakefulness and drowsiness, as well as their FD differences. The experiments were performed on 10 healthy individuals using a fourteen-channel montage. An explanation is offered on the causes of the detected FD changes. FD values of 60s records belonging to wake (Horis stage 1) and drowsy (Horis stages 2-4) states were calculated for each channel and each subject. In 136 out of 140 epochs an increase in FD was obtained. Relationship between signal FD and its relative alpha amplitude was mathematically modeled and we quantitatively demonstrated that the increase in FD was predominantly due to a reduction in alpha activity. The model was generalized to include other EEG oscillations. By averaging FD values for each channel across 10 subjects, four clusters (O2O1; T6P4T5P3; C3F3F4C4F8F7; T4T3) for the wake and two clusters (O2O1P3T6P4T5; C3C4F4F3F8T4T3F7) for the drowsy state were statistically verified. Topographic distribution of FD values in wakefulness showed a lateral symmetry and a partial fronto-occipital gradient. In drowsiness, a reduction in the number of clusters was detected, due to regrouping of channels T3, T4, O1 and O2. Topographic distribution of absolute FD differences revealed largest values at F7, O1 and F3. Reorganization of channel clusters showed that regionalized brain activity, specific for wakefulness, became more global by entering into drowsiness. Since the global increase in FD during wake-to-drowsy transition correlated with the decrease of alpha power, we inferred that increase of EEG complexity may not necessarily be an index of brain activation.

Similarities between explicit and implicit motor imagery in mental rotation of hands: An EEG study

Chronometric and imaging studies have shown that motor imagery is used implicitly during mental rotation tasks in which subjects for example judge the laterality of human hand pictures at various orientations. Since explicit motor imagery is known to activate the sensorimotor areas of the cortex, mental rotation is expected to do similar if it involves a form of motor imagery. So far, functional magnetic resonance imaging and positron emission tomography have been used to study mental rotation and less attention has been paid to electroencephalogram (EEG) which offers a high time-frequency resolution. The time-frequency analysis is an established method for studying explicit motor imagery. Although hand mental rotation is claimed to involve motor imagery, the time-frequency characteristics of mental rotation have never been compared with those of explicit motor imagery. In this study, time-frequency responses of EEG recorded during explicit motor imagery and during a mental rotation task, inducing implicit motor imagery, were compared. Fifteen right-handed healthy volunteers performed motor imagery of hands in one condition and hand laterality judgement tasks in another while EEG of the whole head was recorded. The hand laterality judgement was the mental rotation task used to induce implicit motor imagery. The time-frequency analysis and sLORETA localisation of the EEG showed that the activities in the sensorimotor areas had similar spatial and time-frequency characteristics in explicit motor imagery and implicit motor imagery conditions. Furthermore this sensorimotor activity was different for the left and for the right hand in both explicit and implicit motor imagery. This result supports that motor imagery is used during mental rotation and that it can be detected and studied with EEG technology. This result should encourage the use of mental rotation of body parts in rehabilitation programmes in a similar manner as motor imagery.

Influence of motor imagination on cortical activation during functional electrical stimulation

Highlights • In a motor imagery based BCI system to control FES, practicing imagery both before and during FES additionally increases intensity of event related desynchronisation throughout the whole period of electrical stimulation.• Discontinuing to practice motor imagery following the onset of FES, reduces subsequent event-related desynchronisation.• Motor imagery and FES produce event-related desynchronisation in similar frequency ranges.

Movement-related cortical potentials in paraplegic patients: abnormal patterns and considerations for BCI-rehabilitation

Non-invasive EEG-based Brain-Computer Interfaces (BCI) can be promising for the motor neuro-rehabilitation of paraplegic patients. However, this shall require detailed knowledge of the abnormalities in the EEG signatures of paraplegic patients. The association of abnormalities in different subgroups of patients and their relation to the sensorimotor integration are relevant for the design, implementation and use of BCI systems in patient populations. This study explores the patterns of abnormalities of movement related cortical potentials (MRCP) during motor imagery tasks of feet and right hand in patients with paraplegia (including the subgroups with/without central neuropathic pain (CNP) and complete/incomplete injury patients) and the level of distinctiveness of abnormalities in these groups using pattern classification. The most notable observed abnormalities were the amplified execution negativity and its slower rebound in the patient group. The potential underlying mechanisms behind these changes and other minor dissimilarities in patients’ subgroups, as well as the relevance to BCI applications, are discussed. The findings are of interest from a neurological perspective as well as for BCI-assisted neuro-rehabilitation and therapy.

Quantification and visualisation of differences between two motor tasks based on energy density maps for brain–computer interface applications

OBJECTIVE To determine the most discriminative features for a brain-computer interface (BCI) system based on statistically significant differences between two energy density maps calculated from EEG signals during two different motor tasks. METHODS EEG was recorded in ten healthy volunteers while performing different cue based, 3s sustained, real and imaginary right hand movements. Energy density maps were calculated over fixed 240 ms and 2 Hz time-frequency windows (called resels) for each movement and statistically significant resels were determined. After that, normalised energy values of the statistically significant resels were compared between two real as well as between two imaginary movements using a parametric test. RESULTS The largest differences between energy density maps between two motor tasks were noticed on electrode location Cp3 in the higher alpha and the beta bands (i.e., 12-30 Hz), for both real and imaginary movements. The method reduced a total number of discriminative features between two motor tasks to fewer than 2% for the imaginary and fewer than 3% for the real movements on the electrode location Cp3. CONCLUSIONS The method can be used for visualisation and feature extraction for BCI and other applications where event related desynchronisation/synchronisation (ERD/ERS) maps should be compared. SIGNIFICANCE If a reliable on-line classification of imaginary movements of the same limb would be achieved it could be combined with classification of movements of different parts of the body. That would increase a number of separable classes of a BCI system, thereby providing a larger number of command signals to control the external devises such as computers and robotic devices.