Objective: This paper investigates the hypothesis that focal seizures can be predicted using scalp electroencephalogram (EEG) data. Our first aim is to learn features that distinguish between the interictal and preictal regions. The second aim is to define a prediction horizon in which the prediction is as accurate and as early as possible, clearly two competing objectives. Methods: Convolutional filters on the wavelet transformation of the EEG signal are used to define and learn quantitative signatures for each period: interictal, preictal, and ictal. The optimal seizure prediction horizon is also learned from the data as opposed to making an a priori assumption. Results: Computational solutions to the optimization problem indicate a 10-min seizure prediction horizon. This result is verified by measuring Kullback–Leibler divergence on the distributions of the automatically extracted features. Conclusion: The results on the EEG database of 204 recordings demonstrate that (i) the preictal phase transition occurs approximately ten minutes before seizure onset, and (ii) the prediction results on the test set are promising, with a sensitivity of 87.8% and a low false prediction rate of 0.142 FP/h. Our results significantly outperform a random predictor and other seizure prediction algorithms. Significance: We demonstrate that a robust set of features can be learned from scalp EEG that characterize the preictal state of focal seizures.

Focal Onset Seizure Prediction Using Convolutional Networks

Electroencephalography EEG and functional magnetic resonance imaging fMRI record a mixture of ongoing neural processes, physiological and nonphysiological noise. The pattern of interest, such as epileptic activity, is often hidden within this noisy mixture. Therefore, blind source separation BSS techniques, which can retrieve the activity pattern of each underlying source, are very useful. Tensor decomposition techniques are very well suited to solve the BSS problem, as they provide a unique solution under mild constraints. Uniqueness is crucial for an unambiguous interpretation of the components, matching them to true neural processes and characterizing them using the component signatures. Moreover, tensors provide a natural representation of the inherently multidimensional EEG and fMRI, and preserve the structural information defined by the interdependencies among the various modes such as channels, time, patients, etc. Despite the well-developed theoretical framework, tensor-based analysis of real, large-scale clinical datasets is still scarce. Indeed, the application of tensor methods is not straightforward. Finding an appropriate tensor representation, suitable tensor model, and interpretation are application dependent choices, which require expertise both in neuroscience and in multilinear algebra. The aim of this paper is to provide a general guideline for these choices and illustrate them through successful applications in epilepsy. WIREs Data Mining Knowl Discov 2017, 7:e1197. doi: 10.1002/widm.1197

https://onlinelibrary.wiley.com/doi/pdf/10.1002/widm.1197

Tensor decompositions and data fusion in epileptic electroencephalography and functional magnetic resonance imaging data

The discrimination of the preictal state in EEG signals is of great importance in neuroscience and the epileptic seizure prediction field has yet to provide conclusive evidence. In this study, three different classification approaches, including the Repeated Incremental Pruning to Produce Error Reduction (RIPPER) algorithm, Support Vector Machines (SVMs) and Neural Networks (NNs), are investigated for their ability to discriminate preictal from interictal EEG segments. Using public EEG data, a wide range of features is extracted from each segment and then applied to the classifiers. The analysis covers a patient-specific approach, so as to optimize the decision to each patient individually and a patient-independent approach in order to explore a global prediction approach that can discriminate randomly selected preictal and interictal segments from all patients. Overall, the first approach aims at revealing patient-specific epileptic characteristics, whereas the second seeks for potential general preictal-related signs. The results reveal that in the patient-specific case, the SVM classifier exhibits the highest classification accuracy in both preictal and interictal classes reaching 85.75% sensitivity and specificity. As it is expected, the classification performance is lower for the patient-independent case at 68.5%, due to the complicated nature of preictal activity and the variations among patients condition.

Discrimination of Preictal and Interictal Brain States from Long-Term EEG Data

Analysis of epileptic seizures based on EEG using recurrence plot images and deep learning

Graph theory is a well-established theory with many methods used in mathematics to study graph structures. In the field of medicine, electronic health records (EHR) are commonly used to store and analyze patient data. Consequently, it seems straight-forward to perform research on modeling EHR data as graphs. This systematic literature review aims to investigate the frontiers of the current research in the field of graphs representing and processing patient data. We want to show, which areas of research in this context need further investigation. The databases MEDLINE, Web of Science, IEEE Xplore and ACM digital library were queried by using the search terms health record, graph and related terms. Based on the "Preferred Reporting Items for Systematic Reviews and Meta-Analysis" (PRISMA) statement guidelines the articles were screened and evaluated using full-text analysis. Eleven out of 383 articles found in systematic literature review were finally included for analysis in this literature review. Most of them use graphs to represent temporal relations, often representing the connection among laboratory data points. Only two papers report that the graph data were further processed by comparing the patient graphs using similarity measurements. Graphs representing individual patients are hardly used in research context, only eleven papers considered such kind of graphs in their investigations. The potential of graph theoretical algorithms, which are already well established, could help increasing this research field, but currently there are too few papers to estimate how this area of research will develop. Altogether, the use of such patient graphs could be a promising technique to develop decision support systems for diagnosis, medication or therapy of patients using similarity measurements or different kinds of analysis.

/pdf/graph-representation-of-patient-data-a-systematic-literature-x58sonz14p.pdf

Graph-Representation of Patient Data: a Systematic Literature Review.

This work presents a novel modeling of neuronal activity of the brain by capturing the synchronization of EEG signals along the scalp. The pair-wise correspondence between electrodes recording EEG signals are used to establish edges between these electrodes which then become the nodes of a synchronization graph. As EEG signals are recorded over time, we discretize the time axis into overlapping epochs, and build a series of time-evolving synchronization graphs for each epoch and for each traditional frequency band.We show that graph theory provides a rich set of graph features that can be used for mining and learning from the EEG signals to determine temporal and spatial localization of epileptic seizures. We present several techniques to capture the pair-wise synchronization and apply unsupervised learning algorithms, such as k-means clustering and multiway modeling of third-order tensors, to analyze the labeled clinical data in the feature domain to detect the onset and origin location of the seizure. We use k-means clustering on two-way feature matrices for detection of seizures, and Tucker3 tensor decomposition for localization of seizures. We conduct an extensive parametric search to determine the best configuration of the model parameters including epoch length, synchronization metrics, and frequency bands, to achieve the highest accuracy.Our results are promising: we are able to detect the onset of seizure with an accuracy of 88.24%, and localize the onset of the seizure with an accuracy of 76.47%.

http://www.cs.rpi.edu/~dhulena/acmbcb2014_finalsubmission.pdf

Gulcin Irim

Papers

Graph-theoretic analysis of epileptic seizures on scalp EEG recordings