Alan W. L. Chiu

Prediction of seizure onset in an in-vitro hippocampal slice model of epilepsy using Gaussian-based and wavelet-based artificial neural networks.

We propose that artificial neural networks (ANNs) can be used to predict seizure onsets in an in-vitro hippocampal slice model capable of generating spontaneous seizure-like events (SLEs) in their extracellular field recordings. This paper assesses the effectiveness of two ANN prediction schemes: Gaussian-based artificial neural network (GANN) and wavelet-based artificial neural network (WANN). The GANN prediction system consists of a recurrent network having Gaussian radial basis function (RBF) nonlinearities capable of extracting the estimated manifold of the system. It is able to classify the underlying dynamics of spontaneous in-vitro activities into interictal, preictal and ictal modes. It is also able to successfully predict the onsets of SLEs as early as 60 s before. Improvements can be made to the overall seizure predictor design by incorporating time-varying frequency information. Consequently, the idea of WANN is considered. The WANN design entails the assumption that frequency variations in the extracellular field recordings can be used to compute the times at which onsets of SLEs are most likely to occur in the future. Progressions of different frequency components can be captured by the ANN using appropriate frequency band adjustments via pruning, after the initial wavelet transforms. In the off-line processing comprised of 102 spontaneous SLEs generated from 14 in-vitro rat hippocampal slices, with half of them used for training and the other half for testing, the WANN is able to predict the forecoming ictal onsets as early as 2 min prior to SLEs with over 75% accuracy within a 30 s precision window.

The effects of high-frequency oscillations in hippocampal electrical activities on the classification of epileptiform events using artificial neural networks

The existence of hippocampal high-frequency electrical activities (greater than 100 Hz) during the progression of seizure episodes in both human and animal experimental models of epilepsy has been well documented (Bragin A, Engel J, Wilson C L, Fried I and Buzsáki G 1999 Hippocampus 9 137-42; Khosravani H, Pinnegar C R, Mitchell J R, Bardakjian B L, Federico P and Carlen P L 2005 Epilepsia 46 1-10). However, this information has not been studied between successive seizure episodes or utilized in the application of seizure classification. In this study, we examine the dynamical changes of an in vitro low Mg2+ rat hippocampal slice model of epilepsy at different frequency bands using wavelet transforms and artificial neural networks. By dividing the time-frequency spectrum of each seizure-like event (SLE) into frequency bins, we can analyze their burst-to-burst variations within individual SLEs as well as between successive SLE episodes. Wavelet energy and wavelet entropy are estimated for intracellular and extracellular electrical recordings using sufficiently high sampling rates (10 kHz). We demonstrate that the activities of high-frequency oscillations in the 100-400 Hz range increase as the slice approaches SLE onsets and in later episodes of SLEs. Utilizing the time-dependent relationship between different frequency bands, we can achieve frequency-dependent state classification. We demonstrate that activities in the frequency range 100-400 Hz are critical for the accurate classification of the different states of electrographic seizure-like episodes (containing interictal, preictal and ictal states) in brain slices undergoing recurrent spontaneous SLEs. While preictal activities can be classified with an average accuracy of 77.4 +/- 6.7% utilizing the frequency spectrum in the range 0-400 Hz, we can also achieve a similar level of accuracy by using a nonlinear relationship between 100-400 Hz and <4 Hz frequency bands only.

Control of state transitions in an in silico model of epilepsy using small perturbations

We propose the use of artificial neural networks in an in silica epilepsy model of biological neural networks: 1) to predict the onset of state transitions from higher complexities, possibly chaotic to lower complexity possibly rhythmic activities; and 2) to restore the original higher complexity activity. A coupled nonlinear oscillators model (Bardakjian and Diamant, 1994) was used to represent the spontaneous seizure-like oscillations of CA3 hippocampal neurons (Bardakjian and Aschebrenner-Scheibe, 1995) to illustrate the prediction and control schemes of these state transition onsets. Our prediction scheme consists of a recurrent neural network having Gaussian nonlinearities. When the onset of lower complexity activity is predicted in the in silica model, then our control scheme consists of applying a small perturbation to a system variable (i.e., the transmembrane voltage) when it is sufficiently close to the unstable higher complexity manifold. The system state can be restored back to its higher complexity mode utilizing the forces of the systems vector field.

Wavelet-based Gaussian-mixture hidden Markov model for the detection of multistage seizure dynamics: A proof-of-concept study

BackgroundEpilepsy is a common neurological disorder characterized by recurrent electrophysiological activities, known as seizures. Without the appropriate detection strategies, these seizure episodes can dramatically affect the quality of life for those afflicted. The rationale of this study is to develop an unsupervised algorithm for the detection of seizure states so that it may be implemented along with potential intervention strategies.MethodsHidden Markov model (HMM) was developed to interpret the state transitions of the in vitro rat hippocampal slice local field potentials (LFPs) during seizure episodes. It can be used to estimate the probability of state transitions and the corresponding characteristics of each state. Wavelet features were clustered and used to differentiate the electrophysiological characteristics at each corresponding HMM states. Using unsupervised training method, the HMM and the clustering parameters were obtained simultaneously. The HMM states were then assigned to the electrophysiological data using expert guided technique. Minimum redundancy maximum relevance (mRMR) analysis and Akaike Information Criterion (AICc) were applied to reduce the effect of over-fitting. The sensitivity, specificity and optimality index of chronic seizure detection were compared for various HMM topologies. The ability of distinguishing early and late tonic firing patterns prior to chronic seizures were also evaluated.ResultsSignificant improvement in state detection performance was achieved when additional wavelet coefficient rates of change information were used as features. The final HMM topology obtained using mRMR and AICc was able to detect non-ictal (interictal), early and late tonic firing, chronic seizures and postictal activities. A mean sensitivity of 95.7%, mean specificity of 98.9% and optimality index of 0.995 in the detection of chronic seizures was achieved. The detection of early and late tonic firing was validated with experimental intracellular electrical recordings of seizures.ConclusionsThe HMM implementation of a seizure dynamics detector is an improvement over existing approaches using visual detection and complexity measures. The subjectivity involved in partitioning the observed data prior to training can be eliminated. It can also decipher the probabilities of seizure state transitions using the magnitude and rate of change wavelet information of the LFPs.

Stochastic and Coherence Resonance in an In Silico Neural Model

We show that it is possible for chaotic systems to display the main features of stochastic and coherence resonance. In particular, a model of coupled nonlinear oscillators which emulates the transmembrane voltage activities in CA3 neurons, operating in a chaotic regime and in the presence of noise, can exhibit coherence resonance and stochastic resonance. Certain firing frequencies become more “rhythmic” for some optimal values of noise intensity. The effect of noise in different coupling pathways is investigated. We found that the effect of coherence resonance and stochastic resonance are more prominent if noise is presented in either electric field or gap junction coupling pathways. Frequency sensitivity of the model is investigated as a preliminary step in illustrating the principles of possible epileptic seizure control strategies using “chaos control” concepts. Significant effects of stochastic resonance are observed in the 4–8 Hz range. Weaker effects can be found in the 1–4 Hz and 8–10 Hz ranges whereas 0.5 Hz does not exhibit any resonance phenomenon. Our results suggest that: (a) Stochastic resonance could enhance the intrinsic 4–8 Hz rhythms in CA3 neurons more prominently via field coupling pathways. It could also help explain why some reported seizure control strategies using pulse-trains would only be effective at 0.5 Hz. (b) Stochastic resonance-like behavior can occur in the gamma range only if noise is presented via chemical synaptic pathways.

In Vitro Epileptic Seizure Prediction Microsystem

The architecture and VLSI implementation of an epileptic seizure prediction microsystem are presented. The microsystem comprises a neural recording interface and a seizure prediction processor. The two functional blocks have been prototyped in a 0.35 mum CMOS technology and experimentally characterized. The integrated microsystem is validated in predicting the onsets of seizures off line in an in vitro epilepsy model of recurrent spontaneous seizures in the hippocampus of mice.

Prediction of S-glutathionylated Proteins Progression in Alzheimer's Transgenic Mouse Model Using Principle Component Analysis

To date, prediction of Alzheimers disease (AD) is mainly based on clinical criteria because no well-established biochemical biomarkers for routine clinical diagnosis of AD currently exist. We developed an approach to aid in the early diagnosis of AD by using principal component analysis (PCA)-based spectral analysis of oxidized protein electrophoretic profiling. We found that the combination of capillary electrophoresis and PCA analysis of S-glutathionylation distribution characterization can be used in the sample classification and molecular weight (Mw) prediction. The comparison of leave-one-out AD versus non-AD gives the sensitivity of 100% and 93.33% in brain tissues and blood samples, respectively, while the specificity of 100% in brain and 90.0% in blood samples. Our findings demonstrate that PCA of S-glutathionylation electrophoretic profiling detects AD pathology features, and that the molecular weight based electrophoretic profiling of blood and brain S-glutathionylated proteins are sensitive to change, even at the early stage of the disease. Our results offer a previously unexplored diagnostic approach by using electrophoretic characteristics of oxidized proteins to serve as a predictor of AD progression and early stage screening.

Evaluation of a Neural Network-Based Control Strategy for a Cost-Effective Externally-Powered Prosthesis

ABSTRACT This paper presents a control strategy that compensates for the nonlinearity in the inexpensive sensors and hardware of a cost effective prosthetic hand. The control strategy uses neural network-based force control and sensory feedback to detect disturbance induced by slippage. The neural network approach is chosen over other nonlinear models because it is easy to implement and it offered the additional advantage of having its parameters easily adjusted over the life span of the device. The proposed strategy was evaluated on a functional multi-digit underactuated prosthetic hand. The initial and incremental forces exerted from each finger were adjusted to balance the amount of disturbance and the deformation of the objects. Experiments were conducted to test the performance of the protocol in situations encountered in activities of daily living. The displacement of each object under three grasping configurations was measured as a performance criterion while the objects mass was changed. The results showed that with the adjusted parameters for each grasping configuration, the control strategy was able to detect the dynamic changes in mass of the object and was also able to successfully adjust the grasping force before the object drops from the hand.

Long-Term Electrophysiological and Behavioral Analysis on the Improvement of Visual Working Memory Load, Training Gains, and Transfer Benefits

Recent evidence demonstrates that with training, one can enhance visual working memory (VWM) capacity and attention over time in the near transfer tasks. Not only do these studies reveal the characteristics of VWM load and the influences of training, they may also provide insights into developing effective rehabilitation for patients with VWM deficiencies. However, few studies have investigated VWM over extended periods of time and evaluated transfer benefits on non-trained tasks. Here, we combined behavioral and electroencephalographical approaches to investigate VWM load, training gains, and transfer benefits. Our results reveal that VWM capacity is directly correlated to the difference of event-related potential waveforms. In particular, the “magic number 4” can be observed through the contralateral delay amplitude and the average capacity is 3.25-item over 15 participants. Furthermore, our findings indicate that VWM capacity can be improved through training; and after training exercises, participants from the training group are able to dramatically improve their performance. Likewise, the training effects on non-trained tasks can also be observed at the 12th week after training. Therefore, we conclude that participants can benefit from training gains, and augmented VWM capacity sustained over long periods of time on specific variety of tasks.

Classification of intended motor movement using surface EEG ensemble empirical mode decomposition

Noninvasive electroencephalography (EEG) brain computer interface (BCI) systems are used to investigate intended arm reaching tasks. The main goal of the work is to create a device with a control scheme that allows those with limited motor control to have more command over potential prosthetic devices. Four healthy subjects were recruited to perform various reaching tasks directed by visual cues. Independent component analysis (ICA) was used to identify artifacts. Active post parietal cortex (PPC) activation before arm movement was validated using EEGLAB. Single-trial binary classification strategies using support vector machine (SVM) with radial basis functions (RBF) kernels and Fisher linear discrimination (FLD) were evaluated using signal features from surface electrodes near the PPC regions. No significant improvement can be found by using a nonlinear SVM over a linear FLD classifier (63.65% to 63.41% accuracy). A significant improvement in classification accuracy was found when a normalization factor based on visual cue “signature” was introduced to the raw signal (90.43%) and the intrinsic mode functions (IMF) of the data (93.55%) using Ensemble Empirical Mode Decomposition (EEMD).