Joining the benefits: combining epileptic seizure prediction methods.

PURPOSE: In recent years, a variety of methods developed in the field of linear and nonlinear time series analysis have been used to obtain reliable predictions of epileptic seizures. Because individual methods for seizure prediction so far have shown statistical significance but insufficient performance for clinical applications, we investigated possible improvements by combining algorithms capturing different aspects of electroencephalogram (EEG) dynamics. METHODS: We applied the mean phase coherence and the dynamic similarity index to long-term continuous intracranial EEG data. The predictive performance of both methods was assessed and statistically evaluated separately, as well as by using logical "AND" and "OR" combinations. RESULTS: Used independently, either method resulted in a statistically significant prediction performance in only a few patients. Particularly the "AND" combination led to improved prediction performances, leading to an increase in sensitivity and/or specificity. For a maximum false prediction rate of 0.15/h, the mean sensitivity improved from about 25% for the individual methods to 43.2% for the "AND" and to 35.2% for the "OR" combination. DISCUSSION: This study shows that combinations of prediction methods are promising new approaches to enhance seizure prediction performance considerably. It allows merging the individual benefits of prediction methods in a complementary manner. Because either sensitivity or specificity of seizure prediction methods can be improved depending on the needs of the desired clinical application, the combination opens a new window for future use in a clinical setting.

Inferring direct directed-information flow from multivariate nonlinear time series.

Estimating the functional topology of a network from multivariate observations is an important task in nonlinear dynamics. We introduce the nonparametric partial directed coherence that allows disentanglement of direct and indirect connections and their directions. We illustrate the performance of the nonparametric partial directed coherence by means of a simulation with data from synchronized nonlinear oscillators and apply it to real-world data from a patient suffering from essential tremor.

Parametric versus nonparametric transfer function estimation of cerebral autoregulation from spontaneous blood-pressure oscillations.

Cerebral autoregulation (CAR) is a control mechanism of the brain keeping cerebral blood flow constant albeit the arterial blood pressure varies. Impaired CAR may be associated with an increased risk of cerebral ischemic events in patients with obstructive cerebrovascular disease. Spontaneous blood pressure oscillations are analyzed using a nonparametric and two parametric transfer function estimators, i.e. the autoregressive-moving-average model with exogenous inputs or the vector-autoregressive model. Performance of the methods was compared using data from patients with unilateral stenosis or occlusion. We also analyzed reproducibility by comparing partitions of the data an with data from other patients which have been measured twice. Results show that there is no significant difference between methods (ANOVA, p > 0.27), and that CAR measurements can be performed reproducibly (Kendall's tau, p < 0.0016) by all three methods. In conclusion, CAR measurements by means of spontaneous oscillations can be obtained stably and the presented parametric approaches can serve for future online application of CAR measurement.

Estimating causal dependencies in networks of nonlinear stochastic dynamical systems.

The inference of causal interaction structures in multivariate systems enables a deeper understanding of the investigated network. Analyzing nonlinear systems using partial directed coherence requires high model orders of the underlying vector-autoregressive process. We present a method to overcome the drawbacks caused by the high model orders. We calculate the corresponding statistics and provide a significance level. The performance is illustrated by means of model systems and in an application to neurological data.

Time-variant estimation of directed influences during Parkinsonian tremor.

The inference of interaction structures in multidimensional time series is a major challenge not only in neuroscience but in many fields of research. To gather information about the connectivity in a network from measured data, several parametric as well as non-parametric approaches have been proposed and widely examined. Today a lot of interest is focused on the evolution of the network connectivity in time which might contain information about ongoing tasks in the brain or possible dynamic dysfunctions. Therefore an extension of the current approaches towards time-resolved analysis techniques is desired. We present a parametric approach for time variant analysis, test its performance for simulated data, and apply it to real-world data.