Abstracts

The fast and robust detection of both transient spectral patterns as well as transient synchronization phenomena in the human electroencephalogram (EEG) is a general problem in multivariate time series analysis that plays an important role in clinical practice. A vital example is a problem of efficient detection of patho-physiological electrical activities (e.g. epileptic seizures, pre-ictal states) as recorded by the EEG. Despite of the seemingly simplicity to recognize ongoing seizure activities, corresponding patterns are usually patient-specific, noisy, and exhibit varying waveform morphologies. A miniaturized detector of transient spatiotemporal patterns in noisy and non-stationary signals may offer the possibility to design a portable warning system for patients suffering from epileptic seizures., The proposed method exploits the phenomenon of frequency-selective induction of excitation waves in excitable media with respect to the frequency of a locally applied periodic perturbation. By perturbing several such excitable media, each tuned to a different characteristic frequency, a broad frequency band of the applied perturbation can be scanned for the presence of rhythmic components. The resulting system resembles the functioning of the mammalian cochlear and thus represents some sort of a filter bank instantaneously converting analog waveforms into excitation patterns., In this retrospective study, we applied our method to automatically detect seizures in intracranial multi-channel, multi-day EEG recordings from patients undergoing presurgical evaluation. When adjusted appropriately our technique allows to detect seizures with a high sensitivity and specificity. However, we observed a trade-off between the number of false detections and the latency of detection relative to the electrical seizure onset as defined by an expert reader., Our preliminary findings indicate that, in principle, the proposed method can indeed find a broad range of applications in clinical practice. However, further improvements are necessary in order to achieve a high degree of generalization. Future very large scale integrated implementations of such excitable media as a special-purpose digital signal processor or in form of a hardware Cellular Neural Network may represent a miniaturized portable device for mobile applications in fields where immediate detection of transient rhythmic as well as spatially-synchronized activities is crucial., (Supported by Deutsche Forschungsgemeinschaft.)