Evaluation of recurrent neural networks as epileptic seizure predictor

Evaluation of recurrent neural networks as epileptic seizure predictor

Blog Article

The occurrence of epileptic seizures is a problem that makes everyday life difficult for many people who suffer from this disorder, which may causes risks to the subjects and people around them.The arcade smokey the bear belt possibility of predicting epileptic seizures could create a way to minimize risks associated with these attacks.Through computational analysis of Electroencephalography (EEG) signals, a variation of Recurrent Artificial Neural Network (RNN), of Long Short Term Memory (LSTM) type, was developed in an attempt to classify segments of EEG signals that occur before the onset of epileptic seizures, segments which evince the pre-ictal state.Using an EEG database of pediatric subjects with intractable seizures, groups of patients were created, then redundant EEG channels were selected and reduced by footjoy weste herren only 25%.Temporal and spectral features were extracted to improve the evaluation of data patterns related to seizures.

Besides using the unprocessed data, two techniques were proposed, first using the superposition of inputs, then anticipating targets in the training data.A third technique was also proposed by joining the first two.Finally, two LSTM structures were created.The proposed Neural Network achieved high accuracies above 99%; however, the results showed that it was not possible to classify pre-ictal regions by using this combination of architecture and individuals.Furthermore, it was observed that the Neural Network was able to classify ictal regions with up to 61% sensitivity and 99% specificity, confirming the capacity of RNNs of LSTM type to assimilate temporal patterns in EEG data.