.

ISSN 2063-5346
For urgent queries please contact : +918130348310

DEEP LEARNING NETWORK WITH LONG SHORT-TERM MEMORY FOR THE PREDICTION OF EPILEPTIC SEIZURE USING ELECTROENCEPHALOGRAM SIGNALS

Main Article Content

S SIVASARAVANA BABU , M SHERIFF , BALIVADA SAISHANKAR PATNAIK, BARATH S, R.PRADEEPKUMAR ,S.VINAYKUMAR, V.SANJAY
» doi: 10.31838/ecb/2023.12.s1.158

Abstract

Early prediction of seizures can help epileptic patients manage their crises before they occur. A new automated system Combining the process of features extraction and classification solution. The input to deep learning models is a EEG signal, Consequently, fewer computations are needed. An electroencephalogram (EEG), it records and measures the electrical activity of the brain, is now a common component, the range of tools available for studying neural disorders. As a result, these models extract the most discriminative features, Prediction time can be reduced while classification accuracy is improved. This method extracts spatial characteristics from various scalp positions using the LSTM technique. The expressed seizure features from EEG data were studied by using Deep Convolutional Variational Auto - encoder, and a tunable Q-factor wavelet transform has been employed for pre-processing. A suggested channel selection technique makes it suitable for use in real-time. To make sure the data gathered is accurate, a trustworthy test procedure is employed. One of the most prevalent neurological conditions in the world is epilepsy. In the lives of epileptic patients, early seizure prediction has a significant impact. This study proposes a unique deep learning-based method for predicting seizures in individual patients using long-term scalp EEG records. Accurately identifying the preictal brain state, separating it from the dominant interictal state as early as possible, and making it acceptable for real-time are the objectives. A single automated system combines the procedures of feature extraction and categorization. The system considers the raw EEG signal, which has not undergone any preprocessing, as its input, which further streamlines computations. The strongest discriminative characteristics that improve classification accuracy and prediction are extracted using four deep learning models. Convolutional neural networks are used to extract important spatial characteristics from various scalp sites in the proposed method, while recurrent neural networks are used to predict the occurrence of seizures earlier than current techniques. To better solve the optimisation problem, a semi-supervised method based on transfer learning methodology is developed. The most pertinent EEG channels are chosen using a suggested channel selection technique, making the suggested system a strong option for real-time use. Robustness is guaranteed via an efficient test methodology. The proposed method is the most effective on the market today due to the 99.6% accuracy achieved, the low false alarm rate of 0.004 h 1, and the hour-early seizure forecast time.

Article Details