Title Predicting Epileptic Seizure: the Expectations and the RealityName : Dr. Antonio Dourado
Affliation : Professor
University : University of Coimbra
Country : Portugal
More than 25 five years of research in epileptic seizure prediction using multichannel EEG signals have given rise to numerous published results with high sensitivity and specificity. Most of the studies have been developed using short time, pre?prepared data. When real long term data is used, throw several days of continuous records, the results are far behind the previously published ones. This probably explains the fact that there is no real device yet to predict seizures. Instead, this calls the attention for that fact that realistic studies are needed, using appropriate data, developing new features, new classifiers, and using multiple biosignals (EEG, ECG, and more).
Studies developed with 275 refractory patients (with scalp and intracranial EEG) from the European Epilepsy Database will be reported, using time and frequency features and machine learning classifiers, namely support vector machines with a multi?channel high dimensional features space. The features spaces are build from the relative power content of different frequency bands, statistical moments, wavelet transform, phase synchronization, and more. The obtained results wee compared with an analytical random predictor. Depending on the study, performance above chance level was validated for about 15% to 30% of the patients. This low value shows how far we still are from a practical transportable device to be used by patients in daily life.Deep learning with artificial neural networks may have a good potential. Some preliminary studies with autoencoders and LSTM (Long Short Time Memory) neural networks show it, but the results need yet substantial improvements.
Full Professor in the Department of Informatics Engineering, University of Coimbra, Portugal, and scientific coordinator of the Adaptive Computation Group of the Center for Informatics and Systems of the University of Coimbra (CISUC).
Has been active researcher in numerous national and European projects, namely the Principal Investigator of EPILEPSIAE- Evolving Platform for Improving the Living Expectations of Patients Suffering from Ictal Events, a FP 7 EU Project, researching algorithms and devices for epileptic seizures prediction.
Has more that 250 international publications in journals and conferences. Is member of IEEE and ACM and interested in machine learning for diagnosis and prognosis.